ArticlePDF Available

The Architecture of the Nehalem Processor and Nehalem-EP SMP Platforms

Authors:

Abstract and Figures

Nehalem is an implementation of the CISC Intel64 instruction speci�cation based on 45nm and high-k + metal gate transistor technology. Nehalem micro-architectures and system platforms employ a number of state-of-the-art technologies which enable high computation rates for scienti�c and other demanding workloads. Nehalem based processors incorporate multiple cores, on-chip DDR3 memory controller, a shared Level 3 cache and high-speed Quick-Path Interconnect ports for connectivity with other chips and the I/O sub-system. Each core has superscalar, out-of-order and speculative execution pipelines and supports 2-way simultaneous multi-threading. Each core o�ers multiple functional units which can sustain high instruction level parallelism rates with the assistance of program development tools, compilers or special coding techniques. A prominent feature of Intel64 is the processing of SIMD instructions at a nominal rate of 4 double or 8 single precision oating-point instructions per clock cycle. Nehalem platforms are cc-NUMA shared-memory processor systems. Complex processors and platforms, such as those based on Nehalem, present several challenges to application developers, as well as, system level engineers. Developers are faced with the task of writing e�cient code on increasingly complex platforms. System engineers need to understand the system level bottlenecks in order to con�gure and tune the system to yield good performance for the application mix of interest. This report discusses technical details of the Nehalem �􀀀architecture and platforms with an emphasis on inner workings and the cost of instruction execution. The discussion presented here can assist developers and engineers in their respective �elds. The �rst to produce e�cient scalar and parallel code on the Nehalem platform and the latter ones to con�gure and tune a system to perform well under complex application workloads.
Content may be subject to copyright.
The Architecture of the Nehalem Processor
and
Nehalem-EP SMP Platforms
Michael E. Thomadakis, Ph.D.
Supercomputing Facility
miket AT tamu DOT edu
Texas A&M University
March, 17, 2011
Abstract
Nehalem is an implementation of the CISC Intel64 instruction specification based on 45nm and high-k+ metal
gate transistor technology. Nehalem micro-architectures and system platforms employ a number of state-of-the-art
technologies which enable high computation rates for scientific and other demanding workloads. Nehalem based
processors incorporate multiple cores, on-chip DDR3 memory controller, a shared Level 3 cache and high-speed
Quick-Path Interconnect ports for connectivity with other chips and the I/O sub-system. Each core has super-
scalar, out-of-order and speculative execution pipelines and supports 2-way simultaneous multi-threading. Each
core offers multiple functional units which can sustain high instruction level parallelism rates with the assistance
of program development tools, compilers or special coding techniques. A prominent feature of Intel64 is the
processing of SIMD instructions at a nominal rate of 4 double or 8 single precision floating-point instructions per
clock cycle. Nehalem platforms are cc-NUMA shared-memory processor systems.
Complex processors and platforms, such as those based on Nehalem, present several challenges to application
developers, as well as, system level engineers. Developers are faced with the task of writing efficient code on
increasingly complex platforms. System engineers need to understand the system level bottlenecks in order to
configure and tune the system to yield good performance for the application mix of interest. This report discusses
technical details of the Nehalem µarchitecture and platforms with an emphasis on inner workings and the cost of
instruction execution. The discussion presented here can assist developers and engineers in their respective fields.
The first to produce efficient scalar and parallel code on the Nehalem platform and the latter ones to configure
and tune a system to perform well under complex application workloads.
Keywords:Intel64, Nehalem, Core and Uncore, Superscalar Processors, Micro-architecture, Cache Memory Hier-
archy, cc-NUMA, Global Queue, Integrated-Memory Controllers, Local and Remote Memory Performance.
1
Intel Nehalem A Research Report
Contents
1 Introduction 1
1.1 Motivation for this Study . . . . . . . . . . . . . . . . . . 2
1.2 Overview of Features in the Intel Core Micro-Architecture 2
1.3 Summary of New Features in the Intel Micro-Architecture 3
2 The “IntelR
64” Architecture 4
3 The Nehalem Pro cessor 6
3.1 Instruction and Data Flow in Modern Processors . . . . . 6
3.2 Overview of the Nehalem Processor Chip . . . . . . . . . 7
3.3 Nehalem Core Pipeline . . . . . . . . . . . . . . . . . . . 9
3.3.1 Instruction and Data Flow in Nehalem Cores . . . 9
3.3.2 Nehalem Core: Front-End Pipeline . . . . . . . . . 10
3.3.3 Nehalem Core: Out-of-Order Execution Engine . . 13
3.3.4 Execution pipelines of a Nehalem core . . . . . . . 13
3.3.5 Nehalem Core: Load and Store Operations . . . . 17
3.4 Intel SSE Instructions . . . . . . . . . . . . . . . . . . . . 18
3.4.1 Floating-Point SIMD operations in Nehalem core . 19
3.4.2 Floating-Point Registers in Nehalem core . . . . . 19
3.5 Floating-Point Processing and Exception Handling . . . . 22
3.6 Intel Simultaneous Multi-Threading . . . . . . . . . . . . 22
3.6.1 Basic SMT Principles . . . . . . . . . . . . . . . . 23
3.6.2 SMT in Nehalem cores . . . . . . . . . . . . . . . 23
3.6.3 Resource Sharing in SMT . . . . . . . . . . . . . . 24
3.7 CISC and RISC Processors . . . . . . . . . . . . . . . . . 24
4 Cache Hierarchy and Enhancements 25
4.1 Cache-Memory and the Locality Phenomenon . . . . . . . 25
4.2 Cache-Memory Organization in Nehalem . . . . . . . . . 26
4.3 Nehalem Memory Access Enhancements . . . . . . . . . . 27
4.3.1 Store Buffers . . . . . . . . . . . . . . . . . . . . . 27
4.3.2 Load and Store Enhancements . . . . . . . . . . . 28
5 Nehalem-EP Main Memory Organization 30
5.1 Integrated Memory Controller . . . . . . . . . . . . . . . 30
5.2 Cache-Coherence Protocol for Multi-Processors . . . . . . 30
5.2.1 Cache-Coherence Protocol (MESI+F) . . . . . . . 30
5.2.2 Basic MESI Protocol . . . . . . . . . . . . . . . . 32
5.2.3 The Un-core Domain . . . . . . . . . . . . . . . . 33
5.2.4 Local vs. Remote Memory Access . . . . . . . . . 34
6 Virtual Memory in Nehalem Pro cessors 37
6.1 Virtual Memory . . . . . . . . . . . . . . . . . . . . . . . 37
6.2 Nehalem Address Translation Process . . . . . . . . . . . 37
7 Nehalem Main Memory Performance 38
7.1 Ideal Performance Limits . . . . . . . . . . . . . . . . . . 38
7.2 Memory Bandwidth Results . . . . . . . . . . . . . . . . 39
7.2.1 Single Reader . . . . . . . . . . . . . . . . . . . . 39
7.2.2 Single Writer . . . . . . . . . . . . . . . . . . . . 40
7.2.3 Multiple Readers . . . . . . . . . . . . . . . . . . 41
7.2.4 Multiple Writers . . . . . . . . . . . . . . . . . . . 42
7.2.5 Reader and Writer Pair . . . . . . . . . . . . . . . 42
7.2.6 NReader and Writer Pairs . . . . . . . . . . . . . 42
7.3 Memory Latency . . . . . . . . . . . . . . . . . . . . . . 45
8 Intel Turbo Boost Technology 47
List of Figures
1 Intel64 Execution Environment . . . . . . . . . . . . . . . 5
2 Nehalem Processor/Memory Module . . . . . . . . . . . . 8
3 High-level diagram of a Nehalem core pipeline. . . . . . . 9
4 In-Order Front-End Pipeline . . . . . . . . . . . . . . . . 11
5 OoO Back-End Pipeline . . . . . . . . . . . . . . . . . . . 14
6 SIMD Operation . . . . . . . . . . . . . . . . . . . . . . . 18
7 Floating-Point SIMD Operations in Nehalem. . . . . . . . 19
8 Floating-Point Registers in a Nehalem core. . . . . . . . . 20
9 Simultaneous Multi-Treading . . . . . . . . . . . . . . . . 23
10 On-chip Cache Hierarchy . . . . . . . . . . . . . . . . . . 26
11 On-Chip Data Flow . . . . . . . . . . . . . . . . . . . . . 31
12 MESI State Transitions . . . . . . . . . . . . . . . . . . . 32
13 Nehalem-EP 8-way cc-NUMA SMP . . . . . . . . . . . . 35
14 Nehalem Local Memory Access Event Sequence. . . . . . 35
15 Nehalem Remote Memory Access Event Sequence. . . . . 36
16 Bandwidth of a Single Reader Thread . . . . . . . . . . . 39
17 Bandwidth of Single Writer Thread . . . . . . . . . . . . 40
18 Aggregate Bandwidth of Multiple Reader Threads . . . . 41
19 Aggregate Bandwidth of Multiple Writer Threads . . . . 42
20 Bandwidth of a Single Pair of Reader and Writer Threads 43
21 Aggregate Bandwidth of 8 Pairs . . . . . . . . . . . . . . 44
22 Latency to Read a Data Block in Processor Cycles . . . 45
23 Latency to Read a Data Block in Seconds . . . . . . . . 46
1 Introduction
Intel “Nehalem” is the nickname for the “Intel Micro-architecture”, where the latter is a specific implementation
of the “Intel64” Instruction Set Architecture (ISA) specification [1, 2, 3]. For this report, “Nehalem” refers to the
particular implementation where a processor chip contains four cores, the fabrication process is 45nm with high-k+
metal gate transistor technology. We further focus on the Nehalem-EP platform which has two processor sockets per
node and where the interconnection between sockets themselves and between processors and I/O is through Intel’s
Quick-Path Interconnect. Nehalem is the foundation of Intel Core i7 and Xeon processor 5500 series. Even though
Intel64 is a classic Complex-Instruction Set Computer (“CISC”) instruction set type, its Intel Micro-architecture
implementation shares many mechanisms in common with modern Reduced-Instruction Set Computer (“RISC”)
implementations.
Michael E. Thomadakis 1Texas A&M University
Intel Nehalem A Research Report
Each Nehalem chip is a multi-core chip multiprocessor, where each core is capable of sustaining high degrees
of instruction-level parallelism. Nehalem based platforms are cache-coherent non-uniform memory access shared
memory multi-processors. However, to sustain the high instruction completion rates, the developer must have an
accurate appreciation of the cost of using the various system resources.
1.1 Motivation for this Study
Every application, scalar or parallel, experiences a “critical-path” in the system resources it accesses. For a developer
a basic objective is to deploy algorithms and coding techniques which minimize the total application execution
time. A developer has to identify and improve the code which utilizes critical path resources. One “well” known
consequence is that right after of alleviating “the” bottleneck, the next bottleneck emerges. Developers should strike
a balance between the effort to remove a bottleneck and the expected benefits. They need to understand the degree of
Instruction-Level Parallelism (ILP) or Task-Level Parallelism (TLP) available in their code. They have to strike the
right balance at what should be done in parallel between these two extreme levels. A consequence of the ccNUMA
memory architecture of Nehalem is that there is very different cost in accessing data items residing in the local or
the remote physical memories. Developers have to make wise data and computation thread placement decisions for
their code in order to avoid accessing data remotely.
System engineers objectives include increasing system resource utilization, tuning a system to perform well under
mixed application workloads and reducing overall cost of system operation. We can increase system utilization by
allowing more workload into the system to minimize the times resources remain idle. However, it is well known
that the more utilized the service centers (i.e., the resources) get the longer the queuing effects become with the
consequence of individual tasks experiencing higher latencies through the system. Increasing system utilization
therefore, requires one to know which tasks to allow to proceed in parallel so that individual task performance is not
seriously compromised.
Developers and engineers tasks become increasingly challenging as platforms become more complex, while in-
formation is not available to them in a coherent fashion about the inner workings or about system performance
cost.
In this study we attempt to provide in a single place, a coherent discussion of the inner workings and performance
of the Nehalem processor and platform. The sources are scattered around and are in different forms. We proceed
by studying the system operation and cost in layers, namely, from the individual cores all the way out to the
shared memory platform level. Note that this study is the result of continuous and challenging effort and should be
considered to be in the state of continuous expansion. A subsequent publication will focus on the architecture and
performance of Nehalem based clusters, interconnected by high speed low latency interconnects.
Nehalem builds upon and expands the new features introduced by the previous micro-architecture, namely the
45nm “Enhanced Intel Core Micro-architecture” or “Core-2” for short [4, 5, 6, 7]. A short discussion below focuses on
the innovations introduced by the “Penryn” the predecessor to Nehalem micro-architecture and finally the additional
enhancements Nehalem has implemented.
1.2 Overview of Features in the Intel Core Micro-Architecture
The “Core-2” micro-architecture introduced [1, 7] a number of interesting features, including the following
1. “Wide Dynamic Execution” which enabled each processor core to fetch, dispatch, execute and retire up to four instruc-
tions per clock cycle. This architecture had
45nm fabrication technology,
14-stage core pipeline,
4decoders to decode up to 5 instructions per cycle,
3clusters of arithmetic logical units,
Michael E. Thomadakis 2Texas A&M University
Intel Nehalem A Research Report
macro-fusion and micro-fusion to improve front-end throughput,
peak dispatching rate of up to 6micro-ops per cycle,
peak retirement rate of up to 4micro-ops per cycle,
advanced branch prediction algorithms,
stack pointer tracker to improve efficiency of procedure entries and exits.
2. “Advanced Smart Cache” which improved bandwidth from the second level cache to the core, and improved support
for single- and multi-threaded applications computation
2nd level cache up to 4 MB with 16-way associativity,
256 bit internal data path from L2 to L1 data caches.
3. “Smart Memory Access” which pre-fetches data from memory responding to data access patterns, reducing cache-miss
exposure of out-of-order execution
hardware pre-fetchers to reduce effective latency of 2nd level cache misses,
hardware pre-fetchers to reduce effective latency of 1st level data cache misses,
“memory disambiguation” to improve efficiency of speculative instruction execution.
4. “Advanced Digital Media Boost” for improved execution efficiency of most 128-bit SIMD instruction with single-cycle
throughput and floating-point operations
single-cycle inter-completion latency (“throughput”) of most 128-bit SIMD instructions,
up to eight single-precision floating-point operation per cycle,
3 issue ports available to dispatching SIMD instructions for execution.
1.3 Summary of New Features in the Intel Micro-Architecture
Intel Micro-architecture (Nehalem) provides a number of distinct feature enhancements over those of the Enhanced
Intel Core Micro-architecture, discussed above, including:
1. “Enhanced” processor core:
improved branch prediction and recovery cost from mis-prediction,
enhancements in loop streaming to improve front-end performance and reduce power consumption,
deeper buffering in out-of-order engine to sustain higher levels of instruction level parallelism,
enhanced execution units with accelerated processing of CRC, string/text and data shuffling.
2. Hyper-threading technology (SMT):
support for two hardware threads (logical processors) per core,
a4-wide execution engine, larger L3, and large memory bandwidth.
3. “Smarter” Memory Access:
integrated (on-chip) memory controller supporting low-latency access to local system memory and overall scalable
memory bandwidth (previously the memory controller was hosted on a separate chip and it was common to all
dual or quad socket systems),
new cache hierarchy organization with shared, inclusive L3 to reduce snoop traffic,
two level TLBs and increased TLB sizes,
faster unaligned memory access.
4. Dedicated Power management:
Michael E. Thomadakis 3Texas A&M University
Intel Nehalem A Research Report
integrated micro-controller with embedded firmware which manages power consumption,
embedded real-time sensors for temperature, current, and power,
integrated power gate to turn off/on per-core power consumption;
Versatility to reduce power consumption of memory and QPI link subsystems.
In the sections which follow we study and analyze in-depth all the technologies and innovations which make
Nehalem platforms state-of-the-art systems for high-performance computing. This study brings together in one
place information which is not available in a one place. The objective is to study the system at various detail levels
with an emphasis on the cost of execution of code which can be purely scalar, shared or distributed-memory parallel
and hybrid combinations.
2 The “Intel R
64” Architecture
An Instruction Set Architecture (ISA) is the formal definition of the logical or “architected” view of a computer [8, 9].
This is the view that machine language code has of the underlying hardware. An ISA specifies precisely all machine
instructions, the objects which can be directly manipulated by the processor and their binary representation. Data
objects can be, for instance, n-bit integers, floating point numbers of a certain precision, single-byte characters or
complex structures consisting of combinations of simpler objects. Machine instructions determine which elementary
operations the processor can carry out directly on the various data operands. Instruction sets have traditionally been
classified as Complex Instruction Set Computer (CISC) and Reduced Instruction Set Computer (RISC) specifications.
The “Intel64” ISA, historically derives from the 64-bit extensions AMD applied on Intel’s popular 32-bit “IA-
32” ISA for its “K8” processor family. Later on AMD used the name “AMD64” while Intel the names “EM64T”
and “IE-32e”. Finally, Intel settled on “Intel64” as their official 64-bit ISA deriving from the IA-32. The Intel64
architecture supports IA-32 ISA and extends it to fully support natively 64-bit OS and 64-bit applications [1]. The
physical address space in the Intel64 platform can reach up to 48 bits which implies that 256 Tera-binary-Bytes (TiB)
can by directly addressed by the hardware. The logical address size of Intel64 is 64-bit to allow support for a 64-bit
flat linear address space. However, Intel64 hardware addressing currently uses only the last 48-bits. Furthermore,
the address size of the physical memory itself in a Nehalem processor can be up to 40 bits. Intel is reserving the
address filed sizes for future expansions. Intel64 is one of the most prominent CISC instruction sets. Fig. 1 (see
pp. 5) presents the logical (or “architected”) view of the Intel64 ISA [1]. The architected view of an ISA is the
collection of objects which are visible at the machine language code level and can be directly manipulated by machine
instructions. In the 64-bit mode of Intel64 architecture, software may access
a 64-bit linear (“flat”) logical address space,
uniform byte-register addressing,
16 64-bit-wide General Purpose Registers (GPRs) and instruction pointers
16 128-bit “XMM” registers for streaming SIMD extension instructions (SSE, SSE2, SSE3 and SSSE3, SSE4), in
addition to 8 64-bit MMX registers or the 8 80-bit x87 registers, supporting floating-point or integer operations,
fast interrupt-prioritization mechanism, and
a new instruction-pointer relative-addressing mode.
64-bit applications can use 64-bit register operands and 64-bit address pointers through a set of modifier prefixes in
the code. Intel compilers can produce code which takes full advantage of all the features in Intel64 ISA. Application
optimization requires a fair level of understanding of the hardware resources and the cost in using them.
Michael E. Thomadakis 4Texas A&M University
Intel Nehalem A Research Report














 





 
 

!

"
!

#"!$


%&
%'
Figure 1: “Intel64” 64-bit execution environment for Nehalem processor
Michael E. Thomadakis 5Texas A&M University
Intel Nehalem A Research Report
3 The Nehalem Processor
3.1 Instruction and Data Flow in Modern Processors
Nehalem implements a number of techniques to process efficiently the stream of Intel64 ISA CISC “macro-instructions
in the user code. A core internally consists of a large number of functional units (FUs) each capable of carrying
out an elementary “micro-operation” (micro-op). An example of a FU is an ALU (arithmetic and logic unit) which
can carry out an operation against input operands. Micro-ops would specify the operation type and its operands.
Micro-ops are RISC-like type of instructions and they require similar effort and resources to process.
Micro-operations having no dependencies on the results of each other could proceed in parallel if separate FUs
are available. The CISC type of Intel64 macro-instructions are translated by the early stages of the core into one or
more micro-ops. The micro-operations eventually reach the execution FUs where they are dispatched to FUs and
“retire”, that is, have their results saved back to visible (“architected”) state (i.e., data registers or memory). When
all micro-ops of a macro-instruction retire, the macro-instruction itself retires. It is clear that the basic ob jective of
the processor is to maximize the macro-instruction retirement rate.
The fundamental approach Nehalem (and other modern processors) take to maximize instruction completion rates
is to allow the micro-ops of as many instructions as feasible, proceed in parallel with micro-op occupying independent
FUs at each clock cycle. We can summarize the Intel64 instruction flow through the core as follows.
1. The early stages of the processor fetch-in several macro-instructions at a time (say in a cache block) and
2. decode them (break them down) into sequences of micro-ops.
3. The micro-ops are buffered at various places where they can be picked up and scheduled to use the FUs in
parallel if data dependencies are not violated. In Nehalem, micro-ops are issued to stations were they reserve
their position for subsequent,
4. dispatching as soon as their input operands become available.
5. Finally, completed micro-ops retire and post their results to permanent storage.
The entire process proceeds in stages, in a “pipelined” fashion. Pipelining is used to break down a lengthy task
into sub-tasks where intermediate results flow downstream the pipeline stages. In microprocessors, subtasks handled
within each stage take one clock cycle. The amount of hardware logic which goes into each stage has been carefully
selected so that there is approximately an equal amount of work which takes place in every stage. Since adding a
pipeline stage includes some additional fixed overhead for buffering intermediate results, pipeline designs carefully
balance the total number of stages and the duration per stage.
Complex FUs are usually themselves pipelined. A floating-point ALU may require several clock cycles to produce
the results of complex FP operations, such as, FP division or square root. The advantage of pipelining here is that
with proper intermediate result buffering, we could supply a new set of input operands to the pipelined FU in each
clock cycle and then correspondingly expect a new result to be produced at each clock cycle at the output of the FU.
A pipeline bubble takes place when the input operands of a downstream stage are not available. Bubbles flow
downstream at each clock cycle. When the entire pipeline has no input to work with it can stall, that is, it can
suspend operation completely. Bubbles and stalls are detrimental to the efficiency of pipelined execution if they
take place with a “high” frequency. Common reasons for a bubble is when say data has to be retrieved from slower
memory or from a FU which takes multiple cycles to produce them. Compilers and processor designers invest heavily
in minimizing the occurrence and the impact of stalls. A common way to alleviate the frequency of stalls is to allow
micro-ops proceed out of chronological order and use any available FUs. Dynamic instruction scheduling
logic in the processor determines which micro-ops can proceed in parallel while the program execution remains
semantically correct. Dynamic scheduling utilizes the “Instruction Level Parallelism” (ILP) which is possible within
the instruction stream of a program. Another mechanism to avoid pipeline stalling is called *speculative* execution.
Michael E. Thomadakis 6Texas A&M University
Intel Nehalem A Research Report
A processor may speculatively start fetching and executing instructions from a code path before the outcome of a
conditional branch is determined. Branch prediction is commonly used to “predict” the outcome and the target
of a branch instruction. However, when the path is determined not to be the correct one, the processor has to
cancel all intermediate results and start fetching instructions from the right path. Another mechanism relies on data
pre-fetching when it is determined that the code is retrieving data with a certain pattern. There are many other
mechanisms which are however beyond the scope of this report to describe.
Nehalem, as other modern processors, invests heavily into pre-fetching as many instructions, from a predicted
path and translating them into micro-ops, as possible. A dynamic scheduler then attempts to maximize the number
of concurrent micro-ops which can be in progress (“in-flight”) at a time, thus increasing the completion instruction
rates. Another interesting feature of Intel64 is the direct support for SIMD instructions which increase the effective
ALU throughput for FP or integer operations.
3.2 Overview of the Nehalem Processor Chip
A Nehalem processor chip is a “Chip-Multi Processor” (CMP), consisting of several functional parts within a single
silicon die. Fig. 2 illustrates a Nehalem CMP chip and its major parts.
Referring to Fig. 2, a Nehalem chip consists of the following components
four identical compute cores,
UIU: Un-Core Interface Unit (switch connecting the 4 cores to the 4 L3 cache segments, the IMC and QPI
ports),
L3: level-3 cache controller and data block memory,
IMC: 1 integrated memory controller with 3 DDR3 memory channels,
QPI: 2 Quick-Path Interconnect ports, and
auxiliary circuitry for cache-coherence, power control, system management and performance monitoring logic.
A Nehalem chip is divided into two broad domains, namely, the “core” and the “un-core”. Components in the
core domain operate with the same clock frequency as that of the actual computation core. In EOS’s case this is
2.8GHz. The un-core domain operates under a different clock frequency. This modular organization reflects one of
Nehalem’s objectives of being able to consistently implement chips with different levels of computation abilities and
power consumption profiles. For instance, a Nehalem chip may have from two to eight cores, one or more high-speed
QPI interconnects, different sizes for L3 caches, as well as, memory sub-systems with different DRAM bandwidths.
Similar partitioning of CMP chip into different clock domains can be found in other processors, such as, in IBM’s
Power5, 6 and 7, in AMDs multi-core chips and serves very similar purposes.
Outside the Nehalem chip, but at close physical proximity, we find the DRAM which is accessible by means of
three 8-byte DDR3 channels, each capable to operate at up to 1.333 GigaTransfers/sec. The aggregate nominal
main memory bandwidth is 31.992 GB/s per chip, or on the average 7.998 GB/s per core. This is a significant
improvement over all previous Intel micro-architectures. The maximum operating frequency of the DDR3 buses is
determined by the number of DIMMs in the slots.
In essence the “un-core” domain contains the memory controller and cache coherence logic which in earlier Intel
architectures used to be implemented by the separate “North-bridge” chip.
The high performance of the Nehalem architecture relies, among other things, on the fact that the DRAM
controller, the L3 and the QPI ports are all housed within the same silicon die as the four cores. This saves a
significant amount of off-chip communications and makes possible a tightly coupled, low-latency, high bandwidth
CMP system. This particular processor to memory implementation is a significant departure from all previous ones by
Intel. Prior to Nehalem, the memory controller was housed on a separate “Northbridge” chip and it was shared by all
Michael E. Thomadakis 7Texas A&M University
Intel Nehalem A Research Report

Power
Power
&
&
Clock
Clock
 


®

 
!!"#!"$%#&'(
)###*+&'(
(,,-,(
&-



*+
.'(/0'
A. Nehalem Chip and DDR3 Memory Module. The processor chip contains four cores, a shared L3 cache and
DRAM controllers, and Quick-path Interconnect ports.
B. Nehalem micro-photograph.
Figure 2: A Nehalem Processor/Memory module and Nehalem micro-photograph
Michael E. Thomadakis 8Texas A&M University
Intel Nehalem A Research Report








! 
"#

$!
%$&
'
%$&
'
$()&
$!'
*+,-
%.-
+"
+
"#"/
++
0-

1)&-
Figure 3: High-level diagram of a Nehalem core pipeline.
processor chips. The Northbridge has been one of the often cited bottlenecks in previous Intel architectures. Nehalem
has substantially increased the main memory bandwidth and shortened the latency to access main memory. However,
now that a separate DRAM is associated with every IMC and chip, platforms with more than one chips are Non-
Uniform Memory Access (“NUMA”). NUMA organizations have distinct performance advantages and disadvantages
and with proper care multi-threaded computation can make efficient use of the available memory bandwidth. In
general data and thread placement becomes an important part of the application design and tuning process.
3.3 Nehalem Core Pipeline
3.3.1 Instruction and Data Flow in Nehalem Cores
Nehalem cores are modern micro-processors with in-order instruction issue, super-scalar, out-of-order execution
data-paths, which are coupled with a multilevel storage hierarchy. Nehalem cores have extensive support for branch
prediction, speculative instruction execution, data pre-fetching and multiple pipelined FUs. An interesting feature
is the direct support for integer and floating point SIMD instructions by the hardware.
Nehalem’s pipeline is designed to maximize the macro-instruction flow through the multiple FUs. It continues
the four-wide micro-architecture pipeline pioneered by the 65nm “Intel Core Micro-architecture” (“Merom”) and the
45nm “Enhanced Core Micro-architecture” (“Penryn”). Fig. 3 illustrates a functional level overview of a Nehalem
instruction pipeline. The total length of the pipeline, measured by branch mis-prediction delay, is 16 cycles, which
is two cycles longer than that of its predecessor. Referring to Fig. 3, the core consists of
an in-order Front-End Pipeline (FEP) which retrieves Intel64 instructions from memory, uses four decoders
to decode them into micro-ops and buffers them for the downstream stages;
Michael E. Thomadakis 9Texas A&M University
Intel Nehalem A Research Report
an out-of-order super-scalar Execution Engine (EE) that can dynamically schedule and dispatch up to six
micro-ops per cycle to the execution units, as soon as source operands and resources are ready,
an in-order Retirement Unit (RU) which ensures the results of execution of micro-ops are processed and the
“architected” state is updated according to the original program order, and
multi-level cache hierarchy and address translation resources.
We describe in the next two Sub-sections in detail the front-end and back-end pf the core.
3.3.2 Nehalem Core: Front-End Pipeline
Fig. 4 illustrates in more detail key components of Nehalem’s Front-End Pipeline (FEP). The FEP is responsible
for retrieving blocks of macro-instructions from memory and translating them into micro-ops and buffering them for
handling at the execution back-end. FEP handles the code instructions “in-order”. It can decode up to 4macro-
instructions in a single cycle. It is designed to support up to two hardware SMT threads by decoding the instruction
streams of the two threads in alternate cycles. When SMT is not enabled, the FEP handles the instruction stream
of only one thread. Front-End pipeline of a Nehalem core
The Instruction Fetch Unit (IFU) consists of the Instruction Translation Look-aside Buffer (ITLB, discussed
in section 6), an instruction pre-fetcher, the L1 instruction cache and the pre-decode logic of the Instruction Queue
(IQ). The IFU always fetches 16 bytes (128 bits) of aligned instruction bytes on each clock cycle from the Level 1
instruction cache into the Instruction Length Decoder. There is a 128-bit wide direct path from L1 to the IFU. The
IFU always brings in 16 byte blocks.
The IFU uses the ITLB to locate the 16-byte block in the L1 instruction cache and instruction pre-fetch buffers.
Instructions are referenced by virtual address and translated to physical address with the help of a 128 entry ITLB.
A hit in the instruction cache causes 16 bytes to be delivered to the instruction pre-decoder. Programs average
slightly less than 4 bytes per instruction, and since most instructions can be decoded by all decoders, an entire fetch
can often be consumed by the decoders in one cycle. Instruction fetches are always 16-byte aligned. A non-16 byte
aligned target reduces the number of instruction bytes by the amount of offset into the 16 byte fetch quantity. A
taken branch reduces the number of instruction bytes delivered to the decoders since the bytes after the taken branch
are not decoded.
The Branch-Prediction Unit (BPU) allows the processor to begin fetching and processing instructions before
the outcome of a branch instruction is determined. For microprocessors with lengthy pipelines successful branch
prediction allows the processor to fetch and execute speculatively instructions over the “predicted” path without
“stalling” the pipeline. When a prediction is not successful, Nehalem simply cancels all work already done by the
micro-ops already in the pipeline on behalf of instructions along the wrong path. This may get costly in terms of
resources and execution cycles already spent. Modern processors invest heavily in silicon estate and algorithms for
the BPU in order to minimize the frequency and impact of wrong branch predictions.
On Nehalem the BPU makes predictions for the following types of branch instructions
direct calls and jumps: targets are read as a target array, without regarding the taken or not-taken prediction,
indirect calls and jumps: these may either be predicted as having a fixed behavior, or as having targets that
vary according to recent program behavior,
conditional branches: BPU predicts the branch target and whether the branch will be taken or not.
Nehalem improves branch handling in several ways. The Branch Target Buffer (BTB) has been increased
in size to improve the accuracy of branch predictions. Furthermore, hardware enhancements improve the handling
of branch mis-prediction by expediting resource reclamation so that the front-end would not be waiting to decode
instructions in an “architected” code path (the path in which instructions will reach retirement) while resources were
Michael E. Thomadakis 10 Texas A&M University
Intel Nehalem A Research Report








 !
"#

$ 

µ
"%
& '()
$*'(
"#%
!
"&)
)+
",



-)-
 µ.--




!.
(/





01230-


µ



%1%

21%

,









Figure 4: High-level diagram of the In-Order Front-End Nehalem Pipeline (FEP).
Michael E. Thomadakis 11 Texas A&M University
Intel Nehalem A Research Report
allocated to executing mis-predicted code path. Instead, new micro-ops stream can start forward progress as soon
as the front end decodes the instructions in the architected code path. The BPU includes the following mechanisms
Return Stack Buffer (RSB) A 16-entry RSB enables the BPU to accurately predict RET instructions.
Renaming is supported with return stack buffer to reduce mis-predictions of return instructions in the code.
Front-End Queuing of BPU lookups. The BPU makes branch predictions for 32 bytes at a time, twice the
width of the IFU. Even though this enables taken branches to be predicted with no penalty, software should
regard taken branches as consuming more resources than do not-taken branches.
Instruction Length Decoder (ILD or “Pre-Decoder”) accepts 16 bytes from the L1 instruction cache or pre-
fetch buffers and it prepares the Intel64 instructions found there for instruction decoding downstream. Specifically
the ILD
determines the length of the instructions,
decodes all prefix modifiers associated with instructions and
notes properties of the instructions for the decoders, as for example, the fact that an instruction is a branch.
The ILD can write up to 6 instructions per cycle, maximum, into the downstream Instruction Queue (IQ). A
16-byte buffer containing more than 6 instructions will take 2 clock cycles. Intel64 allows modifier prefixes which
dynamically modify the instruction length. These length changing prefixes (LCPs) prolong the ILD process to up to
6 cycles instead of 1.
The Instruction Queue (IQ) buffers the ILD-processed instructions and can deliver up to five instructions in
one cycle to the downstream instruction decoder. The IQ can buffer up to 18 instructions.
The Instruction Decoding Unit (IDU) translates the pre-processed Intel64 macro-instructions into a stream
of micro-operations. It can handle several instructions in parallel for expediency.
The IDU has a total of four decoding units. Three units can decode one simple instruction each, per cycle. The
other decoder unit can decode one instruction every cycle, either a simple instruction or complex instruction, that
is one which translates into several micro-ops. Instructions made up of more than four micro-ops are delivered from
the micro-sequencer ROM (MSROM). All decoders support the common cases of single micro-op flows, including,
micro-fusion, stack pointer tracking and macro-fusion. Thus, the three simple decoders are not limited to decoding
single micro-op instructions. Up to four micro-ops can be delivered each cycle to the downstream instruction decoder
queue (IDQ).
The IDU also parses the micro-op stream and applies a number of transformations to facilitate a more efficient
handling of groups of micro-ops downstream. It supports the following.
Loop Stream Detection (LSD). For small iterative segments of code whose micro-ops fit within the 28-slot In-
struction Decoder Queue (IDQ), the system only needs to decode the instruction stream once. The LSD detects
these loops (backward branches) which could be streamed directly from the IDQ. When such a loop is detected,
the micro-ops are locked down and the loop is allowed to stream from the IDQ until a mis-prediction ends it.
When the loop plays back from the IDQ, it provides higher bandwidth at reduced power, (since much of the
rest of the front end pipeline is shut off. In the previous micro-architecture the loop detector was working with
the instructions within the IQ upstream. The LSD provides a number of benefits, including,
no loss of bandwidth due to taken-branches,
no loss of bandwidth due to misaligned instructions,
no LCP penalties, as the pre-decode stage are used once for
the instruction stream within the loop,
Michael E. Thomadakis 12 Texas A&M University
Intel Nehalem A Research Report
reduced front-end power consumption, because the instruction cache, BPU and pre-decode unit can go
to idle mode. However, note that loop unrolling and other code optimizations may make the loop too
big to fit into the LSD. For high performance code, loop unrolling is generally considered superior for
performance even when it overflows the loop cache capability.
Stack Pointer Tracking (SPT) implements the Stack Pointer Register (RSP) update logic of instructions which
manipulate the program stack (PUSH, POP, CALL, LEAVE and RET) within the IDU. These macro-instructions
were implemented by several micro-ops in previous architectures. The benefits with SPT include
using a single micro-op for these instructions improves decoder bandwidth,
execution resources are conserved since RSP updates do not compete for them,
parallelism in the execution engine is improved since the implicit serial dependencies have already been
taken care of,
power efficiency improves since RSP updates are carried out by a small hardware unit.
Micro-Fusion The instruction decoder supports micro-fusion to improve pipeline front-end throughput and increase
the effective size of queues in the scheduler and re-order buffer (ROB). Micro-fusion fuses multiple micro-ops
from the same instruction into a single complex micro-op. The complex micro-op is dispatched in the out-
of-order execution core. This reduces power consumption as the complex micro-op represents more work in a
smaller format (in terms of bit density), and reduces overall “bit-toggling” in the machine for a given amount
of work. It virtually increases the amount of storage in the out-of-order execution engine. Many instructions
provide register and memory flavors. The flavor involving a memory operand will decodes into a longer flow
of micro-ops than the register version. Micro-fusion enables software to use memory to register operations to
express the actual program behavior without worrying about a loss of decoder bandwidth.
Macro-Fusion The IDU supports macro-fusion which translates adjacent macro-instructions into a single micro-op
if possible. Macro-fusion allows logical compare or test instructions to be combined with adjacent conditional
jump instructions into one micro-operation.
3.3.3 Nehalem Core: Out-of-Order Execution Engine
The execution engine (EE) in a Nehalem core selects micro-ops from the upstream IDQ and dynamically schedules
them for dispatching and execution by the execution units downstream. The EE is a dynamically scheduled “out-
of-order”, super-scalar pipeline which allows micro-ops to use available execution units in parallel when correctness
and code semantics are not violated. The EE scheduler can dispatch up to 6micro-ops in one clock cycle through
the six dispatch ports to the execution units. There are several FUs, arranged in three clusters, for integer, FP and
SIMD operations. Finally, four micro-ops can retire in one cycle, which is the same as in Nehalem’s predecessor
cores. Results can be written-back at the maximum rate of one register per per port per cycle. Fig. 5 presents a
high-level diagram of the Execution Engine along with its various functional units.
3.3.4 Execution pipelines of a Nehalem core
The execution engine includes the following major components:
Register Rename and Allocation Unit (RRAU) – Allocates EE resources to micro-ops in the IDQ and moves
them to the EE.
Reorder Buffer (ROB) – Tracks all micro-ops in-flight,
Unified Reservation Station (URS) – Queues up to 36 micro-ops until all source operands are ready, schedules
and dispatches ready micro-ops to the available execution units.
Michael E. Thomadakis 13 Texas A&M University
Intel Nehalem A Research Report



 










!
!"#$%
&$'



#

"#$%
!#

#
( #
) *+,
-.


!+
!*
!/
!
!0
!1
0µ$
0µ$
0µ$
2


"%)( #
")
*+,.
*+,.
+3 *
&


&

#(($
'#
($#
*

4%
*

4%
5
4%4$
Figure 5: High-level diagram of a the out-of-order execution engine in the Nehalem core. All units are fully pipelined
and can operate independently.
Michael E. Thomadakis 14 Texas A&M University
Intel Nehalem A Research Report
Memory Order Buffer (MOB) – Supports speculative and out of order loads and stores and ensures that writes
to memory take place in the right order and with the right data.
Execution Units and Operand Forwarding Network The execution units are fully pipelined and can produce
a result for most micro-ops with latency 1 cycle.
The IDQ unit (see Fig. 4) delivers a stream of micro-ops to the allocation/renaming stage of the EE pipeline. The
execution engine of Nehalem supports up to 128 micro-ops in flight. The input data associated with a micro-op are
generally either read from the ROB or from the retired register file. When a “dependency chain” across micro-ops
causes the machine to wait for a “slow” resource (such as a data read from L2 data cache), the EE allows other
micro-ops to proceed. The primary objective of the execution engine is to increase the flow of micro-ops, maximizing
the overall rate of instructions reaching completion per cycle (IPC), without compromising program correctness.
Resource Allocation and Register Renaming for micro-ops The initial stages of the out of order core
advance the micro-ops from the front end to the ROB and RS. This process is called micro-op *issue*. The RRAU
in the out of order core carries out the following steps.
1. It allocates resources to micro-ops, such as,
an entry in the re-order buffer (ROB),
an entry in the reservation station (RS),
and a load/store buffer if a memory access is required.
2. It binds the micro-op to an appropriate “dispatch” port.
3. It “renames” source and destination operands of micro-ops in-flight, enabling out of order execution. Operands
are registers or memory in general. Architected (program visible) registers are renamed onto a larger set
of “micro-architectural” (or “non-architectural”) registers. Modern processors contain a large pool of non-
architectural registers, that is, registers which are not accessible from the code. These registers are used
to capture results which are produced by independent computations but which happen to refer to the same
architected register as destination. Register renaming eliminates these false dependencies which are known as
“write-after-write” and “write-after-read” hazards. A “hazard” is any condition which could force a pipeline
to stall to avoid erroneous results.
4. It provides data to the micro-op when the data is either an immediate value (a constant) or a register value
that has already been calculated.
Unified Reservation Station (URS) queues micro-ops until all source operands are ready, then it schedules
and dispatches ready micro-ops to the available execution units. The RS has 36 entries, that is, at any moment there
is a window of up to 36 micro-ops waiting in the EE to receive input. A single scheduler in the Unified-Reservation
Station (URS) dynamically selects micro-ops for dispatching to the execution units, for all operation types, integer,
FP, SIMD, branch, etc. In each cycle, the URS can dispatch up to six micro-ops, which are ready to execute. A
micro-op is ready to execute as soon as its input operands become available. The URS dispatches micro-ops through
the 6 issue ports to the execution units clusters. Fig. 5 shows the 6 issue ports in the execution engine. Each
cluster may contain a collection of integer, FP and SIMD execution units.
The result produced by an execution unit computing a micro-op are eventually written back permanent storage.
Each clock cycle, up to 4results may be either written back to the RS or to the ROB. New results can be forwarded
immediately through a bypass network to a micro-op in-flight that requires it as input. Results in the RS can be
used as early as in the next clock cycle.
The EE schedules and executes next common micro-operations, as follows.
Michael E. Thomadakis 15 Texas A&M University
Intel Nehalem A Research Report
Micro-ops with single-cycle latency can be executed by multiple execution units, enabling multiple streams of
dependent operations to be executed quickly.
Frequently-used micro-ops with longer latency have pipelined execution units so that multiple micro-ops of
these types may be executing in different parts of the pipeline simultaneously.
Operations with data-dependent latencies, such as division, have data dependent latencies. Integer division
parses the operands to perform the calculation only on significant portions of the operands, thereby speeding
up common cases of dividing by small numbers.
Floating point operations with fixed latency for operands that meet certain restrictions are considered excep-
tional cases and are executed with higher latency and reduced throughput. The lower-throughput cases do not
affect latency and throughput for more common cases.
Memory operands with variable latency, even in the case of an L1 cache hit, are not known to be safe for
forwarding and may wait until a store-address is resolved before executing. The memory order buffer (MOB)
accepts and processes all memory operations.
Nehalem Issue Ports and Execution Units The URS scheduler can dispatch up to six micro-ops per cycle
through the six issue ports to the execution engine which can execute up to 6 operations per clock cycle, namely
3 memory operations (1 integer and FP load, 1 store address and 1 store data) and
3 arithmetic/logic operations.
The ultimate goal is to keep the execution units utilized most of the time. Nehalem contains the following
components which are used to buffer micro-ops or intermediate results until the retirement stage
36 reservation stations
48 load buffers to track all allocate load operations,
32 store buffers to track all allocate store operations, and
10 fill buffers.
The execution core contains the three execution clusters, namely, SIMD integer, regular integer and SIMD floating-
point/x87 units. Each blue block in Fig. 5 is a cluster of execution units (EU) in the execution engine. All EUs are
fully pipelined which means they can deliver one result on each clock cycle. Latencies through the EU pipelines vary
with complexity of the micro-op from 1 to 5 cycles Specifically, the EUs associated with each port are the following:
Port 0 supports
Integer ALU and Shift Units
Integer SIMD ALU and SIMD shuffle
Single precision FP MUL, double precision FP MUL, FP MUL (x87), FP/SIMD/SSE2 Move and Logic
and FP Shuffle, DIV/SQRT
Port 1 supports
Integer ALU, integer LEA and integer MUL
Integer SIMD MUL, integer SIMD shift, PSAD and string compare, and
FP ADD
Michael E. Thomadakis 16 Texas A&M University
Intel Nehalem A Research Report
Port 2 Integer loads
Port 3 Store address
Port 4 Store data
Port 5 Supports
Integer ALU and Shift Units, jump
Integer SIMD ALU and SIMD shuffle
FP/SIMD/SSE2 Move and Logic
The execution core also contains connections to and from the memory cluster (see Fig. 5).
Forwarding and By-pass Operand Network Nehalem can support write back throughput of one register
file write per cycle per port. The bypass network consists of three domains of integer, FP and SIMD. Forwarding
the result within the same bypass domain from a producer micro-op to a consumer micro-op is done efficiently in
hardware without delay. Forwarding the result across different bypass domains may be subject to additional bypass
delays. The bypass delays may be visible to software in addition to the latency and throughput characteristics of
individual execution units.
The Re-Order Buffer (ROB) is a key structure in the execution engine for ensuring the successful out-of-
order progress-to-completion of the micro-ops. The ROB holds micro-ops in various stages of completion, it buffers
completed micro-ops, updates the architectural state in macro-instruction program order, and manages ordering of
the various machine exceptions. On Nehalem the ROB has *128* entries to track micro-ops in flight.
Retirement and write-back of state to architected registers is only done for instructions and micro-ops that
are on the correct instruction execution path. Instructions and micro-ops of incorrectly predicted paths are flushed
as soon as mis-prediction is detected and the correct paths are then processed.
Retirement of the correct execution path instructions can proceed when two conditions are satisfied:
1. all micro-ops associated with the macro-instruction to be retired have completed, allowing the retirement of
the entire instruction. In the case of instructions that generate very large numbers of micro-ops, enough to fill
the retirement window, micro-ops may retire.
2. Older instructions and their micro-ops of correctly predicted paths have retired.
These requirements ensure that the processor updates the visible state consistently with the in-order execution of
the macro-instructions of the code.
The advantages of this design is that older instructions which have to block waiting, for example, for the arrival
of data from memory, cannot block younger, but independent, instructions and micro-ops, whose inputs are available.
The micro-ops of these younger instructions can be dispatched to the execution units and warehoused in the ROB
until completion.
3.3.5 Nehalem Core: Load and Store Operations
The memory cluster in the Nehalem core supports:
peak issue rate of one 128-bit (16 bytes) load and one 128-bit store operation per clock cycle
deep buffers for data load and store operations:
48 load buffers,
32 store buffers and
Michael E. Thomadakis 17 Texas A&M University
Intel Nehalem A Research Report








Figure 6: SIMD instructions apply the same FP or integer operation to collections of input data pairs simultaneously.
10 fill buffers;
fast unaligned memory access and robust handling of memory alignment hazards;
improved store-forwarding for aligned and non-aligned scenarios, and
store-to-load data forwarding for most address alignments.
Note that the h/w for memory access and its capabilities as seen by the core are described in detail in a later
subsection.
3.4 Nehalem Core: Intel Streaming SIMD Extensions Instruction Set
Single-Instruction Multiple-Data (SIMD) is a processing technique were the same operation is applied simultaneously
to different sets of input operands. Vector operations, such as, vector additions, subtractions, etc. are examples of
computation where SIMD processing can be applied directly. SIMD requires the presence of multiple Arithmetic
and Logic Units (ALUs) and multiple source and destination operands for these operations. The multiple ALUs
can produce multiple results simultaneously using input operands. Fig. 6 illustrates an example SIMD computation
against four operands. SIMD operating principle
Nehalem supports SIMD processing to integer or floating-point ALU intensive code with the Streaming SIMD
Extensions (SSE) instruction set. This technology has evolved with time and now it represents a rather significant
capability in Nehalem’s micro-architectures. Fig. 7 illustrates the SIMD computation mode in Nehalem. On the left
part of Fig. 7, two double-precision floating-point operations are applied to 2 DP input operands. On the right part
of Fig. 7, four single-precision floating-point operations are applied to 4 SP input operands.
Michael E. Thomadakis 18 Texas A&M University
Intel Nehalem A Research Report























 
Figure 7: Floating-Point SIMD Operations in Nehalem.
3.4.1 Floating-Point SIMD operations in Nehalem core
Nehalem’s execution engine (see Fig. 5) contains the ALU circuitry necessary to carry out two double-precision,
or four single-precision “simple” FP operations, such as addition or subtraction, in each one of the two FP units
accessible through ports 0 and 1. Note that Nehalem execution engine can retire up to 4operations per clock cycle,
including the SIMD FP ones.
Ideal Floating-Point Throughput For the Xeon 5560 which operates at 2.8GHz, we can say that in the steady
state and under ideal conditions each core can retire 4double-precision or 8single-precision floating-point operations
each cycle. Therefore, the nominal, ideal throughput of a Nehalem core, a quad core and a 2-socket system are,
respectively,
11.2Giga FLOPs/sec/core = 2.8GHz ×4FLOPs/Hz
44.8Giga FLOPs/sec/socket = 11.2Giga FLOPs/sec/core ×4cores
89.6Giga FLOPs/sec/node = 44.8Giga FLOPs/sec/socket ×2sockets,
(1)
in terms of double-precision FP operations.
3.4.2 Floating-Point Registers in Nehalem core
SIMD instructions use sets of separate core registers called MMX and XMM registers (shown in Fig. 8). The MMX
registers are 64-bit in size and are aliased to the operand stack for the legacy x87 instructions. XMM registers are
128-bit in size and each can store either 4SP or 2DP floating-point operands. The load and store units can retrieve
and save 128-bit operands from cache or from the main memory.
One interesting feature of Nehalem’s memory subsystem is that certain memory areas can be treated as non-
temporal, that is, they can be used as buffers for vector data streaming in and out of the core, without requiring
Michael E. Thomadakis 19 Texas A&M University
Intel Nehalem A Research Report
Figure 8: Floating-Point Registers in a Nehalem core.
their temporary storage in a cache. This is an efficient way to retrieve a stream of sub-vector operands from memory
to XMM registers, carry out SIMD computation and then stream the results out directly to memory.
Overview of the SSE Instruction Set Intel introduced and extended the support for SIMD operations in stages
over time as new generations of micro-architectures and SSE instructions were released. Below we summarize the
main characteristics of the SSE instructions in the order of their appearance.
MMX(TM) Technology Support for SIMD computations was introduced to the architecture with the “MMX
technology”. MMX allows SIMD computation on packed byte, word, and double-word integers. The integers are
contained in a set of eight 64-bit MMX registers (shown in Fig. 8).
Streaming SIMD Extensions (SSE) SSE instructions can be used for 3D geometry, 3D rendering, speech
recognition, and video encoding and decoding. SSE introduced 128-bit XMM registers, 128-bit data type with four
packed single-precision floating-point operands, data pre-fetch instructions, non-temporal store instructions and other
cache-ability and memory ordering instructions, extra 64-bit SIMD integer support.
Streaming SIMD Extensions 2 (SSE2) SSE2 instructions are useful for 3D graphics, video decoding/encoding,
and encryption. SSE2 add 128-bit data type with two packed double-precision floating-point operands, 128-bit data
types for SIMD integer operation on 16-byte, 8-word, 4-double-word, or 2-quadword integers, support for SIMD
arithmetic on 64-bit integer operands, instructions for converting between new and existing data types, extended
support for data shuffling and extended support for cache-ability and memory ordering operations.
Streaming SIMD Extensions 3 (SSE3) SSE3 instructions are useful for scientific, video and multi-threaded
applications. SSE3 add SIMD floating-point instructions for asymmetric and horizontal computation, a special-
purpose 128-bit load instruction to avoid cache line splits, an x87 FPU instruction to convert to integer independent
of the floating-point control word (FCW) and instructions to support thread synchronization.
Michael E. Thomadakis 20 Texas A&M University
Intel Nehalem A Research Report
Supplemental Streaming SIMD Extensions 3 (SSSE3) SSSE3 introduces 32 new instructions to accelerate
eight types of computations on packed integers.
SSE4.1 SSE4.1 introduces 47 new instructions to accelerate video, imaging and 3D applications. SSE4.1 also
improves compiler vectorization and significantly increase support for packed dword computation.
SSE4.2 Intel during 2008 introduced a new set of instructions collectively called as SSE4.2. SSE4 has been
defined for Intel’s 45nm products including Nehalem. A set of 7 new instructions for SSE4.2 were introduced in
Nehalem architecture in 2008. The first version of SSE4.1 was present in the Penryn processor. SSE4.2 instructions
are further divided into 2 distinct sub-groups, called “STTNI” and “ATA”.
STring and Text New Instructions (STTNI) operate on strings of bytes or words of 16bit size. There are four new
STTNI instructions which accelerate string and text processing. For example, code can parse XML strings faster and
can carry out faster search and pattern matching. Implementation supports parallel data matching and comparison
operations.
Application Targeted Accelerators (ATA) are instructions which can provide direct benefit to specific application targets.
There are two ATA instructions, namely “POPCNT” and “CRC32”.
POPCNT is an ATA for fast pattern recognition while processing large data sets. It improves performance for
DNA/Genome Mining and handwriting/voice recognition algorithms. It can also speed up Hamming distance or
population count computation.
CRC32 is an ATA which accelerates in hardware CRC calculation. This targets Network Attached Storage (NAS)
using iSCSI. It improves power efficiency and reduces time for software I-SCSI, RDMA, and SCTP protocols by
replacing complex instruction sequences with a single instruction.
Intel Advanced Vector Extensions AVX are several vector SIMD instruction extensions of the Intel64
architecture that will be introduced to processors based on 32nm process technology. AVX will expand current
SIMD technology as follows.
AVX introduces 256-bit vector processing capability and includes two components which will be introduced
on processors built on 32nm fabrication process and beyond:
the first generation Intel AVX will provide 256-bit SIMD register support, 256- bit vector floating-point
instructions, enhancements to 128-bit SIMD instructions, support for three and four operand syntax.
FMA is a future extension of Intel AVX, which provides fused floating-point multiply-add instructions
supporting 256-bit and 128-bit SIMD vectors.
General-purpose encryption and AES: 128-bit SIMD extensions targeted to accelerate high-speed block encryp-
tion and cryptographic processing using the Advanced Encryption Standard.
AVX will be introduced with the new Intel 32nm micro-architecture called “Sandy-Bridge”.
Compiler Optimizations for SIMD Support in Executables User applications can leverage the SIMD capa-
bilities of Nehalem through the Intel Compilers and various performance libraries which have been tuned up to take
advantage of this feature. On EOS, use the following compiler options and flags.
-xHost (or the -xSSE4.2) compiler options to instruct the compiler to use the entire set of SSE instructions
in the generated binary
-vec This option enables “vectorization” (better term would be “SIMDizations”) and transformations enabled
for vectorization. This effectively asks the compiler to attempt to use the SIMD SSE instructions available in
Nehalem. Use the -vec-reportNoption to see which lines could use SIMD and which could not and why.
-O2 or-O3
Michael E. Thomadakis 21 Texas A&M University
Intel Nehalem A Research Report
Libraries Optimized for SIMD Support Intel provides user Libraries tuned up for SIMD computation. These
include, Intel’s Math-Kernel Library (MKL), Intel’s standard math library (libimf) and the Integrated-Performance
Primitive library (IPP). Please review the “~/README” file on your EOS home directory with information on the
available software and instructions how to access it. This document contains, among other things, a useful discussion
on compiler flags used for optimization of user code, including SIMD.
3.5 Floating-Point Processing and Exception Handling
Nehalem processors implement a floating-point system compliant with the ANSI/IEEE Standard 754-1985, “IEEE
Standard for Binary Floating-Point Arithmetic”. IEEE 754 defines required arithmetic operations (addition, sub-
traction, sqrt, etc.), the binary representation of floating and fixed point quantities and conditions which render
machine arithmetic valid or invalid. Before this standard, different vendors used to have their own incompatible
FP arithmetic implementations making portability of FP computation virtually impossible. When the result of an
arithmetic operation cannot be considered valid or when precision is lost, the h/w handles a Floating-Point Exception
(FPE).
The following floating-point exceptions are detected by the processor:
1. IEEE standard exception: invalid operation exception for invalid arithmetic operands and unsupported formats
(#IA)
Signaling NaN
∞ − ∞
∞ ÷ ∞
0÷0
∞ × 0
Invalid Compare
Invalid Square Root
Invalid Integer Conversion
2. Zero Divide Exception (#Z)
3. Numeric Overflow Exception (#O)
4. Underflow Exception (#U)
5. Inexact Exception (#P)
The standard defines the exact conditions raising floating point exceptions and provides well-prescribed procedures
to handle them. A user application has a set of choices in how to treat and/or respond, if necessary, to these
exceptions. However, detailed treatment of FPEs is far beyond the scope of this write up.
Please review the following presentation on IEEE 754 Floating-Point Standard and Floating Point Exception
handling http://sc.tamu.edu/systems/hydra/SC-FP.pdf which apply to Nehalem. Note that this presentation is
under revision but it is provides useful material for FP arithmetic.
3.6 Intel Simultaneous Multi-Threading
A Nehalem core supports “Simultaneous Multi-Threading” (SMT), or as Intel calls it “Hyper-Threading”. SMT
is a pipeline design and implementation scheme which permits more than one hardware threads to execute simulta-
neously within each core and share its resources. For Nehalem, two threads can be simultaneously executing within
each core. Fig. 5 shows the different execution units within a Nehalem core which the two SMT threads can share.
Michael E. Thomadakis 22 Texas A&M University
Intel Nehalem A Research Report









Figure 9: Simultaneous Multi-Treading (SMT) concept on Nehalem cores.
3.6.1 Basic SMT Principles
The objective of SMT is to allow the 2nd hardware thread to utilize functional units in a core which the 1st hardware
thread leaves idle. In Fig. 9, the right-hand side part demonstrates the case where two threads execute simultaneously
within a core with SMT enabled. The horizontal dimension shows the occupancy of the functional units of a core
and the vertical one shows consecutive clock cycles. As you can see, both SMT threads may “simultaneously” (i.e.,
at the same clock period) utilize these units, making progress.
The alternative to SMT would be to let a thread run until it has to stall (e.g., waiting for a lengthy FP operation
to finish or a cache memory miss to be handled), at which point in time the OS dispatcher would have to carry
out a costly context-switching operation with processor state swapping. This is illustrated in an idealized fashion
(i.e., without accounting for the resource waste due to context-switching overhead) on the right-hand side part of
the figure. SMT can potentially exploit “task-level” concurrency at a very fine level and produces cost saving by
avoiding context-switching.
In short, the potential advantages of SMT are several, including among others, the increased utilization of
functional units that would have remained idle, the overall increased throughput in instructions completed per clock
cycle and the overhead savings from the lower number of thread switching operations. It implicitly can save power
consumed by the idle units.
3.6.2 SMT in Nehalem cores
When SMT is ON, each Nehalem core appears to the Operating System as two logical processors. An SMT enabled
dx360-M2 node appears as 16 logical processors to Linux.
On Nehalem, SMT takes advantage of the 4-wide execution engine. The units are kept busy with the two threads.
SMT hides the latency experienced by a single thread. One prominent advantage is that with SMT it is more likely
that an active unit will be producing some result on behalf of a thread as opposed to consuming power while it is
Michael E. Thomadakis 23 Texas A&M University
Intel Nehalem A Research Report
waiting for work. Overall, SMT is much more efficient in terms of power than adding another core. One Nehalem,
SMT is supported by the high memory bandwidth and the larger cache sizes.
3.6.3 Resources on Nehalem Cores Shared Among SMT Threads
The Nehalem core supports SMT by replicating, partitioning or sharing existing functional units in the core. Specif-
ically the following strategies are used:
Replication The unit is replicated for each thread.
register state
renamed RSB
large page ITLB
Partitioning The unit is statically allocated between the two threads
load buffer
store buffer
reorder buffer
small page ITLB
Competitive Sharing The unit is dynamically allocated between the two threads
reservation station
cache memories
data TLB
2nd level TLB
SMT Insensitive All execution units are SMT transparent
3.7 CISC and RISC Processors
From the discussion above, it is clear that on the Nehalem processor, the CISC nature of the Intel64 ISA exits the
scene after the instruction decoding phase by the IDU. By that time all CISC macro-instructions have been converted
into RISC like micro-ops which are then scheduled dynamically for parallel processing at the execution engine. The
specific execution engine of the Nehalem we described above could have been be part of any RISC or CISC processor.
In deed one cannot tell by examining it if it is part of a CISC or a RISC processor. Please see a companion article
on Power5+ http://sc.tamu.edu/systems/hydra/hardware.php processors and systems to make comparisons and
draw some preliminary conclusions.
Efficient execution of applications is the ultimate objective and this requires an efficient flow of ISA macro-
instructions through the processor. This implies accurate branch prediction and efficient fetching of instructions,
their efficient decoding into micro-ops and a *maximal flow* of micro-ops from issue to retirement through the
execution engine.
This points to one of the successes of the RISC approach where sub-tasks are simple and can be executed in
parallel in multiple FUs by dynamic dispatching. Conversely, Nehalem has invested heavily in silicon real estate and
clock cycles into preprocessing the CISC macro-instructions so that can be smoothly converted into sequences of
micro-ops. The varying length of the CISC instructions requires the additional overhead in the ILD. A RISC ISA
would had avoided this overhead and instructions would had moved directly from fetch to decoding stage.
At the same time, it is obvious that Intel has done a great job in processing very efficiently a heavy-weight CISC
ISA, using all the RISC techniques. Thus the debate of RISC vs. CISC remains a valid and open question.
Michael E. Thomadakis 24 Texas A&M University
Intel Nehalem A Research Report
Given modern back-end engines, which ISA style is more efficient to capture at a higher-level the semantics of
applications?
Is it more efficient to use a RISC back-end engine with a CISC or a RISC ISA and front-ends?
It would be very interesting to see how well the Nehalem back-end execution engine would perform when fitted
in a RISC processor front-end, handling a classical RISC ISA. For instance, how would a classical RISC, such
as a Power5+ would perform if the Nehalem execution engine were to replace its own?
Conversely, how would the Nehalem perform if it were fitted with the back-end execution engine of a classical
RISC, such as that of an IBM Power5+ processor ?
From the core designer point of view, can I select different execution engines for the same ISA ?
The old CISC vs. RISC debate is resurfacing as a question of how more aptly and concisely RISC or a CISC ISA can
express the semantics of applications, so that when the code is translated into micro-ops powerful back-end execution
engines can produce results at a lower cost, i.e., in shorter amount of time and/or using less power?
4 Memory Organization, Hierarchy and Enhancements in Nehalem
Processors
4.1 Cache-Memory and the Locality Phenomenon
The demand for increasingly larger data and instruction sections in applications requires that the size of the main
memory hosting them be also sufficiently large. Experience with modern processors suggests that 2 to 4 GiB are
needed per compute core to provide a comfortable size for a main memory. However, cost and power consumption
for this large amounts of memory, necessitates the use of the so called, Dynamic Random Access Memory (DRAM)
technology. DRAM allows the manufacturing of large amounts of memory using simpler memory elements (i.e., by
a transistor and a capacitor which needs to be dynamically refreshed every a few 10s of Milli-seconds). However, the
bandwidth rates at which modern processors require to access memory in order to operate efficiently, far exceed the
memory bandwidth that can be supported with current DRAM technologies.
Another type of memory, called “Static RAM” (SRAM) implements memory elements with more complex orga-
nization (5-6 transistors). SRAM is much faster than the DRAM and it does not require periodic refreshing of the
bit contents. However, with more electronic components per bit, memory density per chip decreases dramatically
while the power consumption grows. We cannot currently provide 2-4 GiB of RAM per core using just SRAM with
a viable cost.
Computer architects design modern processors with multiple levels of faster, smaller and more expensive cache
memories. Cache memories, mostly implemented with SRAM logic, maintain copies of recently and frequently used
instruction and data blocks of the “main” (DRAM) memory. When an object is accessed for the first time, the
hardware retrieves a block of memory containing it from the DRAM and stores it in the cache. Subsequent object
accesses go directly to the faster cache and avoid the lengthy access to DRAM.
This is a viable approach due to the phenomenon of “temporal” and “spatial locality” in the memory access
patterns exhibited by executable code. Simply speaking, temporal locality means that objects (data or instructions)
accessed recently, have a higher probability to get accessed in the near future than other memory objects. Spatial
locality means that objects physically adjacent in memory to those accessed recently have a higher probability of
getting accessed “soon”. Temporal locality stems from the fact that within in a short span of time instructions
in iterations (loops) are executed repeatedly likely accessing common data. Spatial locality is the result of code
accessing dense array locations in linear order or simply accessing the next in sequence instruction. Hardware and
compiler designers invest heavily in mechanisms which can leverage the locality phenomenon. Compilers strive to
Michael E. Thomadakis 25 Texas A&M University
Intel Nehalem A Research Report








!"
#$%!
&'


$()#

(
*#$
#+"',$-
(
#&+",$-'
%(
.-/
$+"$* -
%(
.-/

,+"'%0
# (
.-/
1"

02
+
3
+&2)+&2
02405
.
.
.

6
1
6
1
&







$*7,8
&
"
,$
(


9!"
(
9


+
Figure 10: Overview of Cache Memory Hierarchy and Data Flow Paths to and from Nehalem’s Core.
co-locate items which are likely to be accessed together within short time spans. Hardware logic detects sequential
memory access and attempts to pre-fetch subsequent blocks ahead of time. The cache memories eventually have to
evict least used contents to make room for incoming new ones.
4.2 Cache-Memory Organization in Nehalem
Nehalem [1, 2] divides the physical memory into blocks 64 byte in size. These blocks, referred to as “cache blocks”
or “cache lines”, are the units of data the memory system transfers among the major subsystems.
The architecture supports a hierarchy of up to three levels of cache memory and DRAM memory. Fig. 10 shows
the different caches in a Nehalem chip, their connectivity with the common L3, QPI and IMC, along with the TLBs
translation structures.
Referring to Fig. 10, a Nehalem core contains an instruction cache, a first-level data cache and a second-level
unified cache. Each physical processor chip may contain several processor cores and a shared collection of subsystems
that are referred to as “uncore”. Specifically, in Intel Xeon 5560 processors, there are four cores and the uncore
provides a unified third-level cache shared by all cores in the chip, Intel QuickPath Interconnect ports and auxiliary
logic such as, a performance monitoring unit, control configuration registers and power management units, among
others.
The processor always reads a cache line from system memory beginning on a 64-byte boundary (which has an
Michael E. Thomadakis 26 Texas A&M University
Intel Nehalem A Research Report
address with its 6 least-significant bits zero). A cache line can be filled from memory with 8-byte transfer burst
transactions. The caches do not support partially-filled cache lines, so caching even a single double-word requires
caching an entire line.
L1 Cache At Level 1 (L1), separate instruction and data caches are part of the Nehalem core (called a “Harvard”
style). The instruction and the data cache are each 32 KiB in size. The L1 data-cache has a single access data port,
and a block size of 64 bytes. In SMT mode, the caches are shared by the two hardware threads running in the core.
The instruction and the data caches have 4-way and 8-way set associative organization, respectively. The access
latency to retrieve data already in L1 data-cache is 4clocks and the “throughput” period is 1 clock.
L2 Cache Each core also contains a private, 256KiB, 8-way set associative, unified level 2 (L2) cache (for both
instructions and data). L2’s block size is 64 bytes and access time for data already in the cache is 10 clocks. The
write policy is write-back and the cache is non-inclusive.
L3 Cache The Level 3 (L3) cache is a unified, 16-way set associative, 8 MiB cache shared by all four cores on
the Nehalem chip. The latency of L3 access may vary as a function of the frequency ratio between the processor and
the uncore sub-system. Access latency is around 3540+ clock cycles, depending on the clock ration of the core and
the uncore domains.
The L3 is inclusive (unlike L1 and L2), meaning that a cache line that exists in either L1 data or instruction, or
the L2 unified caches, also exists in L3. The L3 is designed to use the inclusive nature to minimize “snoop” traffic
between processor cores and processor sockets. A 4-bit valid vector indicates if a particular L3 block is already cached
in the L2 or L1 cache of a particular core in the socket. If the associated bit is not set, it is certain that this core
is not caching this block. A cache block in use by a core in a socket, is cached by its L3 cache which can respond
to snoop requests by other chips, without disturbing (snooping into) L2 or L1 caches on the same chip. The write
policy is write-back.
4.3 Nehalem Memory Access Enhancements
The data path from L1 data cache to the memory cluster is 16 bytes in each direction. Nehalem maintains load and
store buffers between the L1 data cache and the core itself.
4.3.1 Store Buffers
Intel64 processors temporarily store data for each write (store operation) to memory in a store buffer (SB). SBs
are associated with the execution unit in Nehalem cores. They are located between the core and the L1 data-cache.
SBs improve processor performance by allowing the processor to continue executing instructions without having to
wait until a write to memory and/or to a cache is complete. It also allows writes to be delayed for more efficient use
of memory-access bus cycles.
In general, the existence of store buffers is transparent to software, even in multi-processor systems like the
Nehalem-EP. The processor ensures that write operations are always carried out in program order. It also insures
that the contents of the store buffer are always drained to memory when necessary.
when an exception or interrupt is generated;
when a serializing instruction is executed;
when an I/O instruction is executed;
when a LOCK operation is performed;
when a BINIT operation is performed;
when using an SFENCE or MFENCE instruction to order stores.
Michael E. Thomadakis 27 Texas A&M University
Intel Nehalem A Research Report
4.3.2 Load and Store Enhancements
The memory cluster of Nehalem supports a number of mechanisms which speed up memory operations, including
out of order execution of memory operations,
peak issue rate of one 128-bit load and one 128-bit store operation per cycle from L1 cache,
“deeper” buffers for load and store operations: 48 load buffers, 32 store buffers and 10 fill buffers,
data pre-fetching to L1 caches,
data pre-fetch logic for pre-fetching to the L2 cache
fast unaligned memory access and robust handling of memory alignment hazards,
memory disambiguation,
store forwarding for most address alignments and
pipelined read-for-ownership operation (RFO).
Data Load and Stores Nehalem can execute up to one 128-bit load and up to one 128-bit store per cycle, each
to different memory locations. The micro-architecture enables execution of memory operations out-of-order with
respect to other instructions and with respect to other memory operations.
Loads can
issue before preceding stores when the load address and store address are known not to conflict,
be carried out speculatively, before preceding branches are resolved
take cache misses out of order and in an overlapped manner
issue before preceding stores, speculating that the store is not going to be to a conflicting address.
Loads cannot
speculatively take any sort of fault or trap
speculatively access the uncacheable memory type
Faulting or uncacheable loads are detected and wait until retirement, when they update the programmer visible state. x87
and floating point SIMD loads add 1 additional clock latency.
Stores to memory are executed in two phases:
Execution Phase Prepares the store buffers with address and data for store forwarding (see below). Consumes dispatch
ports 3 and 4.
Completion Phase The store is retired to programmer-visible memory. This may compete for cache banks with executing
loads. Store retirement is maintained as a background task by the Memory Order Buffer, moving the data from the
store buffers to the L1 cache.
Data Pre-fetching to L1 Caches Nehalem supports hardware logic (DPL1) for two data pre-fetchers in the L1
cache. Namely
Data Cache Unit Prefetcher DCU (also known as the “streaming prefetcher”), is triggered by an ascending access
to recently loaded data. The logic assumes that this access is part of a streaming algorithm and automatically
fetches the next line.
Michael E. Thomadakis 28 Texas A&M University
Intel Nehalem A Research Report
Instruction Pointer-based Strided Prefetcher IPSP keeps track of individual load instructions. When load
instructions have a regular stride, a prefetch is sent to the next address which is the sum of the current address
and the stride. This can prefetch forward or backward and can detect strides of up to half of a 4KB-page, or
2 KBytes.
Data prefetching works on loads only when loads is from write-back memory type, the request is within the page
boundary of 4 KiB, no fence or lock is in progress in the pipeline, the number of outstanding load misses in progress
are below a threshold, the memory is not very busy and there is no continuous stream of stores waiting to get
processed.
L1 prefetching usually improves the performance of the memory subsystem, but in rare occasions it may degrade
it. The key to success is to issue the pre-fetch to data that the code will use in the near future when the path from
memory to L1 cache is not congested, thus effectively spreading out the memory operations over time. Under these
circumstances pre-fetching improves performance by anticipating the retrieval of data in large sequential structures
in the program. However, it may cause some performance degradation due to bandwidth issues if access patterns
are sparse instead of having spatial locality.
On certain occasions, if the algorithm’s working set is tuned to occupy most of the cache and unneeded pre-fetches
evict lines required by the program, hardware prefetcher may cause severe performance degradation due to cache
capacity of L1.
In contrast to hardware pre-fetchers, software prefetch instructions relies on the programmer or the compiler to
anticipate data cache miss traffic. Software prefetch act as hints to bring a cache line of data into the desired levels
of the cache hierarchy.
Data Pre-fetching to L2 Caches DPL2 pre-fetch logic brings data to the L2 cache based on past request patterns
of the L1 to the L2 data cache. DPL2 maintains two independent arrays to store addresses from the L1 cache, one for
upstreams (12 entries) and one for down streams (4 entries). Each entry tracks accesses to one 4K byte page. DPL2
pre-fetches the next data block in a stream. It can also detect more complicated data accesses when intermediate
data blocks are skipped. DPL2 adjusts its pre-fetching effort based on the utilization of the memory to cache paths.
Separate state is maintained for each core.
Memory Disambiguation A load instruction micro-op may depend on a preceding store. Many micro-architectures
block loads until all preceding store address are known. The memory disambiguator predicts which loads will not
depend on any previous stores. When the disambiguator predicts that a load does not have such a dependency, the
load takes its data from the L1 data cache. Eventually, the prediction is verified. If an actual conflict is detected,
the load and all succeeding instructions are re-executed.
Store Forwarding When a load data follows a store which reloads the data the store just wrote to memory, the
micro-architecture can forward the data directly from the store to the load in many cases. This is called “tore-to-load”
forwarding, and it saves several cycles by allowing a data requester receive data already available on the processor
instead of waiting for a cache to respond. However several conditions must be met for store to load forwarding to
proceed without delays:
the store must be the last store to that address prior to the load,
the store must be equal or greater in size than the size of data being loaded and
the load data must be completely contained in the preceding store.
In previous micro-architectures specific address alignments and data sizes between the store and load operations would
determine whether a store-to-load forwarding might proceed directly or get delayed going through the cache/memory
sub-system. Intel micro-architecture (Nehalem) allows store-to-load forwarding to proceed regardless of store address
alignment.
Michael E. Thomadakis 29 Texas A&M University
Intel Nehalem A Research Report
Efficient Access to Unaligned Data The cache and memory subsystems handle a significant amount of instruc-
tions and data with different address alignment scenarios. Different address alignments have varying performance
impact on memory and cache operations based on the implementation of these subsystems. On Nehalem the data
path to the L1 caches are 16 bytes wide. The L1 data cache can deliver 16 bytes of data in every cycle, regardless
how their addresses are aligned. However, if a 16-byte load spans across a cache line boundary, the data transfer will
suffer a mild delay in the order of 4 to 5 clock cycles. Prior micro-architectures imposed much heavier delays.
5 Nehalem-EP Main Memory Organization
5.1 Integrated Memory Controller
The integrated memory controller (IMC) for Nehalem supports three 8-byte channels of DDR3 memory operating
at up to 1.333 GigaTransfer/sec (GT/s). Fig. 11 shows the IMC in a Nehalem chip. Total theoretical bandwidth
between DRAM and the IMC in the un-core domain of the chip is 31.992 GB/s. The memory controller supports both
registered and un-registered DDR3 DRAM. Each channel of memory can operate independently and the controller
services requests out-of-order to minimize latency. Each core supports up to 10 data cache misses and 16 total
outstanding misses. This places a strict upper bound on the memory bandwidth per core.
5.2 Cache-Coherence Protocol for Multi-Processors
The conveniences of the cache memories come with some extra cost when the system has multiple processors. Copies
of data which have been retrieved and modified by a processor in its local cache are inconsistent with the original
copy in main memory. When another processor accesses the same data item it should receive the latest up-to-
date copy and not an older stale version of it. This problem of Memory Consistency is addressed with Cache
Coherence (CC) mechanisms. CC ensures that the value of an item retrieved by any processor in the system is
the most up-to-date one. CC may add considerable overhead in accessing memory in multi-processors. CC logic is
in the critical path of accessing memory and can easily become the main bottleneck, exacerbating the processor and
memory speed gap. Recent processors provide increasingly tuned and adaptive CC protocols which try to stay to any
extend feasible out of the way in accessing memory. Ideally, accesses to disjoint data by separate processors should
proceed without any additional overhead. Conflicting access to the same data item (reads and writes) by different
processors should extend the latency as minimally as possible, maintain fairness and avoid indefinite postponement.
Cache coherence mechanism have been studied extensively in the literature and still are hot topics as there is always
an increasing demand for larger multi-processors and more efficient concurrent access to shared memory.
5.2.1 Cache-Coherence Protocol (MESI+F)
Practical reasons require that CC protocols maintain memory consistency in terms of 64-byte memory blocks and
not in individual bytes or words. Memory blocks are the units of physical memory transfer. Each block has a unique
identification, and belongs to a unique Nehalem socket (“home location”) and is managed by the local IMC. Based
on the way and time they propagate modifications of the blocks, CC protocols are divided into 2 categories, namely
write-update and write-invalidate. Based on the way they locate multiple copies of the same block, are divided into
“snoopy” and the directory based protocols. For the discussion which follows please refer to Fig. 11.
Nehalem processors use the MESIF (Modified, Exclusive, Shared, Invalid and Forwarding) cache protocol to
maintain cache coherence with caches on the same chip and on other chips via the QPI. MESIF belongs to the write-
invalidate, snoopy (with a small directory part) category, and it is a variation of the well known MESI protocol. The
designations used in the its acronym are the possible states that cache memory blocks can be in as they are transfered
among cores, caches, I/O and DRAM. When a core reads or modifies memory objects causes the their corresponding
Michael E. Thomadakis 30 Texas A&M University
Intel Nehalem A Research Report
 








 















 



® !!"
#$%&!'
(
!"
!"
#
$
%$&
'
()
*!
)()*+
#(#*+#(#*+
),(),*
),(),*
Figure 11: Nehalem On-Chip Memory Hierarchy and Data Traffic through the Chip and Global Queue.
Michael E. Thomadakis 31 Texas A&M University
Intel Nehalem A Research Report

 








Figure 12: The State Transitions of Cache Blocks in the Basic MESI Cache-Coherence Protocol.
block to transition from one of these states to another. The current state of a block and the requested operation
against it prescribes the h/w to follow a different sequence of tasks which provably maintain memory consistency.
5.2.2 Basic MESI Protocol
Initially all blocks in a cache do not store actual data and they are in the “Invalid” state. When a core reads a data
object, it always checks first the L1 memory to see if the block is already there. The first time a block is accessed
results in a cache read miss and it is in the Invalid state. If the block is not in L1 but is found in the L2 cache
(L1 miss and L2 hit), it is transferred to the L1 data cache and the data access instruction proceeds. If the block is
neither in the L2, then it must be retrieved from the “un-core”. In general, a read-miss causes the core to retrieve
the entire 64-byte cache block containing the object into the appropriate cache (L1, L2, L3, or all). This operation
is called a cache-line fill. A block read for the first time by any core transitions to the “Exclusive” state.
The next time a core needs to access the same or nearby memory locations, if the block is still in the cache the
data object is retrieved directly from the cache instead of going back to memory. This is called a cache hit.
When a core A has already retrieved a block in the Exclusive state and core B requires to read the same block,
the cache coherence h/w stores a copy of this block in the cache of core B and changes its state to “Shared”.
When a core wants to write an operand to memory, it first checks if the corresponding block is already in the
cache. If a valid cache line does exist, the processor can write the operand into the cache instead of writing it out to
system memory. This operation is called a write hit.
A write which refers to a memory location not currently in the cache causes a write-miss. In this case the
core performs a cache line-fill, write allocation and proceeds to modify the value of the operand in the cache line
without writing directly to memory. The Intel64 architecture does not write-allocate on a write miss if when the
write operation is “non-temporal”. When data is “streamed” in and out of the processor, as the case may be with
SIMD/vector operands, there is no need to store this data in the cache as it is not expected to be needed in the
near future. Write allocate operations are costly especially when the block to enter the cache would have to evict
Michael E. Thomadakis 32 Texas A&M University
Intel Nehalem A Research Report
a modified block already resident in a conflicting cache location. The Intel compilers may select the non-temporal
write operations when it is evident that the written data are streamed out of the processor.
When a core attempts to modify data in a block in the Shared state, the h/w issues a “Request-for-Ownership”
(RfO) transaction which invalidates all copies in other caches and transitions its own (unique now) copy to Modified
state. The owning core can read and write to this block without having to notify the other cores. If any of the cores
previously sharing this block attempts to read this block, it will receive a cache-miss since the block is Invalid in that
core’s cache. Note that when a core attempts to modify data in a Exclusive state block, NO “Request-for-Ownership”
transaction is necessary since it is certain that no other processor is caching copies of this block.
For Nehalem which is a multi-processor platform, the processors have the ability to “snoop” (eavesdrop) the
address bus for other processor’s accesses to system memory and to their internal caches. They use this snooping
ability to keep their internal caches consistent both with system memory and with the caches in other interconnected
processors.
If through snooping one processor detects that another processor intends to write to a memory location that it
currently has cached in Shared state, the snooping processor will invalidate its cache block forcing it to perform a
cache line fill the next time it accesses the same memory location.
If a core detects that another core is trying to access a memory location that it has modified in its cache, but
has not yet written back to system memory, the owning core signals the requesting core (by means of the “HITM#”
signal) that the cache block is held in Modified state and will perform an implicit write-back of the modified data.
The implicit write-back is transferred directly to the requesting core and snooped by the memory controller to assure
that system memory has been updated. Here, the processor with the valid data can transfer the block directly to the
other core without actually writing it to system memory; however, it is the responsibility of the memory controller
to snoop this operation and update memory.
Each memory block can be stored in a unique set of cache locations, based on a subset of their memory block
identification. A cache memory with associativity Kcan store each memory block to up to Kalternative locations.
If all Kcache slots are occupied by memory blocks, the K+1 request will not have room to store this latest memory
block. This requires that one of the existing Kblocks has to be written out to memory (or the inclusive L3 cache) if
this block is in Modified state. Cache memories commonly use a Least Recently Used (LRU) cache replacement
strategy where they evict the block which has not been accessed recently.
As we mentioned, before written out to memory, data operands are first saved in a store buffer. They are then
written from the store buffer to memory when the system path to memory is available.
Note that when all 10 of the line-fill buffers in a core become occupied, outstanding data access operations queue
up in the load and store buffers and cannot proceed. When this happens the core’s front end suspends issuing new
micro-ops to the RS and OOO engine to maintain pipeline consistency.
5.2.3 The Un-Core Domain and Multi-Socket Cache Coherence
In the Nehalem processor the “un-core” domain essentially is a shared last level L3 cache (“LLC”), a memory access
chip-set (“Northbridge”), and a QPI socket interconnection interface [10]. Fig. 11 illustrates a quad-core Nehalem
processor chip and the connections between the cores, the cache memories, the local DRAM and the inter-socket QPI
links. Technical information published about the un-core does not go into sufficient depth to permit an understanding
of the underlying h/w capabilities and limitations. Note that in Section 7 we attempt to quantify the performance
of the Nehalem-EP based actual low level cache and memory access experiments.
The un-core supports cache line access requests (such as, L2 cache misses, “un-cacheable” loads and stores) from
the on-chip cores and transfers data among the L3 cache, the local IMC, the remote socket and the I/O Hub.
Memory consistency in a multi-core, multi-socket system like the Nehalem-PE, is maintained across sockets. With
the introduction of the Intel Quick-Path Interconnect protocol the, 4 MESI states are supplemented with a fifth,
“Forward” (F) state. A specific cache memory is chosen to store a block in the Forward state and it is allowed to
forward it to other requesters. Blocks in the Shared state cannot be directly forwarded. Forwarding of a block from
Michael E. Thomadakis 33 Texas A&M University
Intel Nehalem A Research Report
a single cached location only is done to avoid cache coherence protocol complexity and unnecessary data transfers.
This would occur if multiple caches were allowed to forward the same block to a requester. Forwarding is done
beneficially when a block belongs to a remote IMC, it is already cached by the remote L3 and it can be forwarded
by th L3 faster than accessing the remote DRAM.
Cache line requests from the on-chip four cores, from a remote chip or the I/O hub are handled by the Global
Queue (GQ) (see Fig. 11) which resides in the Uncore. The GQ buffers, schedules and manages the flow of data
traffic through the uncore. The operations of the GQ is most-critical for the efficient exchange of data within and
among Nehalem processor chips. The GQ contains 3request queues for the different request types:
Write Queue (WQ): is a 16-entry queue for store (write) memory access operations from the local cores.
Load Queue (LQ): is a 32-entry queue for load (read) memory requests by the local cores.
QPI Queue (QQ): is a 12-entry queue for off-chip requests delivered by the QPI links.
According to [10], each pair of cores share one port to the Global Queue. Thus for the two core pairs on a chip there
is a total of 64 request buffers for read operations to the local DRAM. We suspect that this buffer limit places an
artificial upper bound in the available read bandwidth to the local DRAM. Unfortunately, no other information is
available to elucidate the limitations of the GQ h/w. A “cross-bar” switch assists GQ to exchange data among the
connected parts.
When the GQ receives a cache line request from one of the cores, it first checks the on-chip Last Level Cache (L3)
to see if the line is already cached there. As the L3 is inclusive, the answer can be quickly determined. If the line is
in the L3 and was owned by the requesting core it can be returned to the core from the L3 cache directly. If the line
is being used by multiple cores, the GQ snoops the other cores to see if there is a modified copy. If so the L3 cache is
updated and the line is sent to the requesting core. In the event of an L3 cache miss, the GQ sends out requests for
the line. Since the cache line could be cached in the other Nehalem chip, a request through the QPI to the remote
L3 cache is made. As each Nehalem processor chip has its own local integrated memory controller, the GQ must
identify the “home” location of the requested cache line from the physical address. If the address identifies home as
being on the local chip, then the GQ makes a simultaneous request to the local IMC. If home belongs to the remote
chip, the request sent by the QPI will also be used to access the remote IMC.
This process can be viewed in the terms of the QPI protocol as follows. Each socket has a “Caching Agent” (CA)
which might be thought of as the GQ plus the L3 cache and a “Home agent” (HA) which is the IMC. An L3 cache
miss results in simultaneous queries for the line from all the CAs and the HA (wherever home is). In a Nehalem-EP
system there are 3caching agents, namely the 2 sockets and an I/O hub. If none of the CAs has the cache line, the
home agent ultimately delivers it to the caching agent that requested it. Clearly, the IMC has queues for handling
local and remote, read and write requests.
5.2.4 Local vs. Remote Memory Access
In Nehalem, the integrated memory controller substantially improved memory latency and bandwidth, compared to
predecessor micro-architectures. For the two socket implementations of Nehalem EP (see Fig. 13 ), the remote latency
is higher, since the memory request and response must go through a QPI link. This shared memory organization is
called “cache-coherent Non-Uniform Memory Access” (cc-NUMA) and it is very common in modern SMP platforms.
The latency to access the local memory is, approximately, 65 nano-seconds. The latency to access the remote memory
is, approximately, 105 nano-seconds. That is, remote accesses are 1.6 to 1.7 times the latency of local memory access.
The available bandwidth through the QPI link is 12.8 GB/s which is approximately %40 of the theoretical
bandwidth of the three local DDR3 channels.
Access to Local Memory DRAM Fig. 14 demonstrates access to a memory block whose home location is in
the directly (locally) attached DRAM. The sequence of steps are the following:
Michael E. Thomadakis 34 Texas A&M University
Intel Nehalem A Research Report
 

 

 





 


!"
#!$
%
&'
"("
 

 

 



 

"
("




)&'

&'
*

+
 
!"
Figure 13: Nehalem-EP 8-way cc-NUMA SMP, Memory Hierarchies and Local vs. Remote Memory Organization.










Figure 14: Nehalem Local Memory Access Event Sequence.
Michael E. Thomadakis 35 Texas A&M University
Intel Nehalem A Research Report


 







Figure 15: Nehalem Remote Memory Access Event Sequence.
1. Proc0 requests a cache line which is not in its L1, l2 nor in shared L3 cache
- Proc0 requests data from its DRAM,
- Proc0 snoops Proc1 to check if data is present there.
2. Response
- local DRAM returns data;
- Proc1 returns snoop response;
- Proc0 installs block in its L3, L2 and L1 cache and retrieves target memory word.
The local memory access latency is the maximum of the above two steps.
Access to Remote Memory DRAM Fig. 15 illustrates access to a memory block whose home location is the
remote DRAM memory (directly attached to the other Nehalem chip). The steps to access the remote memory block
are the following:
1. Proc0 requests a cache line which is in not in Proc0’s L1, L2 or L3 cache;
2. Request sent over QPI to Proc1
3. Proc1’s probes for cache line
- Proc1’s IMC makes requests to its own DRAM;
- Proc1 snoops internal caches;
4. Response
- Data block returns to Proc0 via the QPI;
- Proc0 installs cache block in L3, L2 and L1;
- Proc0 sets the state of cache block to Shared or Exclusive.
Michael E. Thomadakis 36 Texas A&M University
Intel Nehalem A Research Report
The remote memory access latency is the sum of steps 1, 2, 3 and 4 and clearly is a function of QPI latencies.
The cache coherency protocol messages, among the multiple sockets, are exchanged over the Intel QPI. The
inclusive L3 cache mode allows this protocol to operate rather fast, with the latency to the L3 cache of the adjacent
socket being even less than the latency to the local memory.
One of the main virtues of the integrated memory controller is the separation of the cache coherency traffic and
the memory access traffic. This enables a tangible increase in memory access bandwidth, compared to previous
architectures, but it results in a non-uniform memory access (NUMA). The latency to the memory DIMMs attached
to a remote socket is considerably longer than to the local DIMMs. A second advantage is that the memory control
logic can run at processor frequencies and thereby reduce the latency.
6 Virtual Memory in Nehalem Processors
6.1 Virtual Memory
In modern processors, an executable is logically divided into “sections”, or “segments” which have distinct function.
For instance, an executable consists, among others, of a “code”, a “stack” and a “heap” segment. For reasons of
effective space management and efficient utilization, each segment is divided into logical units, called pages. A page
in different varies from a few KiBytes in size, (e.g., 4 KiB) to several MiBytes or even GiBytes.
Each page can actually be stored anywhere in the physical memory, in any of the available main memory slots
known as page frames. We can consider the main memory as an array of pages frames. As an example, a physical
memory with 4GiBytes capacity has available exactly 1 Mi page frames for pages having 4KiB size. Applications can
define vast amounts of memory, but they usually refer or access a very small subset of it. When a program references
for the first time a memory location (for instance a new subroutine call or a reference to a data item in an array)
the system selects a free page frame and “pages in” the corresponding page. Application pages already in page slots
which have not been recently used become candidates for eviction to make room for new pages.
The mechanism which dynamically manages the page frames and keeps track of the mapping between pages and
page frames is called the Virtual Memory (VM) management system.
While applications execute refer to memory using “Effective Addresses” (EA) which are virtual memory addresses.
“Physical Addresses” (PA) are actual addresses the memory hardware uses to identify specific memory locations.
The VM system dynamically translates EAs into PAs, in process called “Virtual Address Translation”. VM systems
keeps track of the various program segments and corresponding pages with in memory in data structures called VM
segment and page tables. These structures end up taking plenty of memory space. Multiple levels of page tables are
used to cut down on the actual space used. This multi-level indirection requires the traversal of multiple tables for
each address an application uses and it is in the critical path of the tasks the processor has to carry out in order to
retire each macro-instruction. For this reason, special hardware called “Translation Look-aside Buffers” (TLBs) is
used to speed up this process.
6.2 Nehalem Address Translation Process
Address Sizes Intel64 architecture defines translation from a “flat”, linear 64-bit “Effective Address” (EA) into a
“Physical Address” (PA) with a width to up to 52 bits [2, 11]. Note that even though this mode produces 64-bit
linear addresses, the processor ensures that bits 63:47 of such an address are identical (“sign extension” of bit 47).
This implies that the Virtual Address (VA) the paging system uses has an effective width of 48 bits. Although 52
bits corresponds to 4 PiBytes, since linear addresses are limited to 48 bits, at most 256 TiBytes of linear-address
space may be accessed at any given time by a single process. Note finally that on Nehalem, the physical address size
is actually 40 bits.
VM Page Sizes Supported Nehalem support virtual memory page sizes of 4KiB, 2MiB and 1GiB “huge” pages.
Please refer to Fig. 11 which shows in detail the hardware components involved in the EA to PA translation process
Michael E. Thomadakis 37 Texas A&M University
Intel Nehalem A Research Report
and the various levels of cache memories.
TLBs The processor architecture specifies two-levels of translation look-aside buffers, TLB0and TLB1to speed-up
the EA to PA translation process. The TLB is a cache of recent EA to PA translations. Intel64 allows TLBs and other
h/w structures to cache address mappings information which is stored in main memory. When a process executes
it is associated with an integer Process-Context Identifier (PCId). In this way TLBs and other structures may
cache multiple mappings associated with different processes simultaneously. The first level TLB0consists of separate
TLBs for data DTLB0and instructions ITLB0. DTLB0handles address translation for data accesses, it provides
64 entries to support 4KiB pages and 32 entries for large pages. The ITLB0provides 64 entries (per thread) for
4KiB pages and 7 entries (per thread) for large pages. The second level unified UTLB1handles both code and data
accesses for 4KiB pages. It support 4KiB page translation operation that missed ITLB0or DTLB0.
Here is a list of entries in each TLB
ITLB0for 4-KiB pages: 64 entries / SMT thread, 4-way associative;
ITLB0for 2-MiB “huge” pages: 7 entries / SMT thread, fully associative;
DTLB0for 4-KiBe pages: 64 entries, 4-way associative;
DTLB0for 2-MiB pages : 32 entries, 4-way associative;
UTLB1for 4-KiB pages: 512 entries for both data and instruction look-ups.
An DTLB0miss and UTLB1hit causes a penalty of 7cycles. Software only pays this penalty if the DTLB0is
used in some dispatch cases. The delays associated with a miss to the UTLB1and Page-Miss Handler are largely
non-blocking.
7 Nehalem Main Memory Performance
This section presents actual performance measurements of the cache and memory systems in a Nehalem-EP platforms.
The experiments investigate the performance of basic, low-level memory operations ad allow us to quantify the
actual performance limits. We compare latencies and bandwidth of different access patterns with those allowed by
the hardware under ideal conditions. Even though not comprehensive, these results offer a good insight into actual
performance limits of the memory subsystems in the Nehalem-EP platform. A subsequent publication will present
more thorough memory system performance evaluation of Nehalem and likely of the Westmere which is the follow-on
platform for Intel64 systems.
7.1 Ideal Performance Limits
Ideal or “Theoretical” data transfer figures are obtained by simply multiplying the transfer rates times the width
of the various system channels. Please review Sections 4 and 5 for a discussion of the memory architecture on
Nehalem-EP platforms and the ideal performance figures. Here we focus on the Xeon 5560 processor which has a
core domain clock of 2.8GHz. Each core in a socket should be able to access the DRAM attached to the local IMC at
the maximum bandwidth of the DDR3 paths connecting the IMC to the memory DIMMs. The three DDR3 channels
to local DRAM support a bandwidth of 31.992GB/s = 3 ×8×1.333G transfers/s. The available bandwidth to access
memory blocks on the other socket should be bounded by the QPI inter-core links which can achieve
When remote memory locations are access data is transfered over the QPI link connecting the Nehalem chips.
The available bandwidth through the QPI link is 12.8 GB/s (approximately %40 of the theoretical bandwidth to
the local DRAM) is the absolute upper bound to access remote DRAM.
We divide the experimental results below into bandwidth and latency sections, for single reader, single writer,
combined reader and writer, multiple reader and writer threads. By varying the amount of total memory accessed we
Michael E. Thomadakis 38 Texas A&M University
Intel Nehalem A Research Report
5k
10k
15k
20k
25k
30k
35k
40k
45k
10k 100k 1M 10M 100M 1G
MB/s
memory size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.single-reader
read bandwidth CPU0 locally
read bandwidth CPU0 accessing CPU1 memory
read bandwidth CPU0 accessing CPU2 memory
read bandwidth CPU0 accessing CPU3 memory
read bandwidth CPU0 accessing CPU5 memory
Figure 16: Bandwidth of a Single Reader Thread
are able to demonstrate the bandwidth and latencies to access memory at all 3 levels of cache and of the main memory.
Finally, we present the different cost to access memory blocks owned by the local vs.remote memory controller. This
last results demonstrates the intensity the NUMA effect has on the bandwidth and latency of applications accessing
Nehalem system memory. In all experiments CPU0, . . ., CPU3, and CPU4, . . ., CPU7, belong to chips 0 and 1,
respectively. All experiments were carried out on our Nehalem-EP cluster called EOS [?] and they utilized only 4KiB
size pages.
For all experiments we utilize the “BenchIT” [12, 13] open source package1which was built on our target Nehalem-
EP target system. We used a slightly modified version of a collection of benchmarks called “X86membench” [14] which
are also available at the BenchIT site.
7.2 Memory Bandwidth Results
7.2.1 Single Reader Thread
In this experiment, a single thread reads memory segments of size varying from 10KiBs to 200MBs. All memory
blocks are in the Exlusive state. The top curve in Fig. 16 (labeled “read bandwidth CPU0 locally”) plots the observed
bandwidth when a thread running on CPU0, accesses data which have as home location DRAM managed by the
IMC on the same socket as CPU0. No other thread is reading any of these data blocks. This data traverse only
the memory path which connects the CPU0 core to the common L3, and then the private L2 and L1 data cache
1Visit the official web site of the BenchIt project at http://www.benchit.org.
Michael E. Thomadakis 39 Texas A&M University
Intel Nehalem A Research Report
5k
10k
15k
20k
25k
30k
35k
40k
45k
10k 100k 1M 10M 100M 1G
MB/s
memory size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.single-writer
memory bandwidth: CPU0 writing memory used by CPU0
memory bandwidth: CPU0 writing memory used by CPU1
memory bandwidth: CPU0 writing memory used by CPU2
memory bandwidth: CPU0 writing memory used by CPU3
memory bandwidth: CPU0 writing memory used by CPU5
Figure 17: Bandwidth of Single Writer Thread
memories of CPU0. We can see that the observed bandwidth in accessing data areas which fit entirely within the L1
data cache (32 KiB) is very close to the ideal L1 bandwidth of 44.8GB/sec = 2.8GHz ×16bytes.
Next, the memory ready bandwidth of segments of size up to the L2 cache (256 KiBs) stays a little below 30
GB/sec. As soon as the data cannot entirely fit in the L2, data is retrieved from the inclusive 8MiB L3 cache. While
data sizes stay below 8MiB, performance stays at around 22.5 GB/sec.
Note that the next three from the top curves namely those labeled “CPU0 accessing CPU1”, and so on, up to
“CPU0 accessing CPU3”, represent threads running respectively on CPU and accessing data already read in by
CPU1, CPU2 and CPU3 which are on the same quad-core socket as CPU0. Read performance is essentially that of
L3 as data are served off the inclusive L3 cache to the rest of the cores of the same socket.
Soon after data sizes exceed the 8MiBs of L3 capacity, all data is fetched from DRAM. Since our system has
two sockets, each owning a different set of DRAM chips, we have two distinct performance cases. The first case is
when all data is brought in from the DRAM which is owned by the same socket as the cores on which the threads
execute and this achieves a bandwidth of approximately 11.5 GiB/sec.
The second case is when the thread running on CPU0 accesses data which are homed on the other Nehalem
socket. The achieved bandwidth is around 8.8 GiB/sec when the data size is less than 8MiBs and around 7.8
GiB/sec otherwise.
7.2.2 Single Writer Thread
In this experiment, a single thread writes to memory segments of size varying from 10KBs to 200MBs. When a core
has to write to a data block, the MESI(F) protocol requires a “Read-for-Ownership” operation which snoops and
Michael E. Thomadakis 40 Texas A&M University
Intel Nehalem A Research Report
0
50k
100k
150k
200k
250k
300k
350k
10k 100k 1M 10M 100M 1G
MB/s
Memory Size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.multiple-reader
read bandwidth 0-31
Figure 18: Aggregate Bandwidth of Multiple Reader Threads
invalidates this block already stored on other caches. All data blocks are in the Modified state when the writes are
performed against them.
The top curve in Fig. 17 (labeled “CPU0 writing memory used by CPU0”) demonstrates the case a thread
exclusively writes to data only owned by itself. The next three curves plot the write bandwidth of a thread running
on CPU0 writing to data blocks which are already cached by the other three cores on the same chip and are in the
Modified state. Updating data in the L1 of another core takes longer than writing to the L2 of the same core.
As soon as a core starts writing data blocks which are larger than L3, data spill over to the local DRAM at a
rate of 8.2 GiB/sec, approximately.
Finally, the bottom curve represents the bandwidth of writing to memory which is owned by the IMC on the
other socket. It can be seen that as soon as the L3 capacity of the socket is exceeded, the write bandwidth to remote
DRAM drops to 5.5 GiB/sec, approximately.
7.2.3 Multiple Reader Threads
In this experiment, 8 threads are split evenly across all cores and read simultaneously from disjoint locations memory
segments of sizes up to 200MBs. Cached memory blocks are in the Exclusive state. As we can see, while the L1
data cache can satisfy the requests, the aggregate read bandwidth is very close to the ideal one which is equal to
358GB/sec = 2.8GHz ×8×16bytes.
We can then see the clear pattern of the bandwidths while memory reads can be satisfied from L1, L2 and L3
caches. Finally, when all requests go directly to DRAM, read bandwidth settles to approximately 40 GB/sec, or
correspondingly 20 GB/sec per socket.
Michael E. Thomadakis 41 Texas A&M University
Intel Nehalem A Research Report
0
50k
100k
150k
200k
250k
300k
350k
10k 100k 1M 10M 100M 1G
MB/s
memory size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.multiple-writer
write bandwidth 0-31
Figure 19: Aggregate Bandwidth of Multiple Writer Threads
7.2.4 Multiple Writer Threads
In this experiment, 8 threads are split evenly across all cores and write simultaneously to disjoint locations memory
segments of sizes up to 200MBs. As we can see, while the L1 data cache can satisfy the requests, the aggregate write
bandwidth is also very close to the ideal one which is equal to 358GB/sec = 2.8GHz ×8×16bytes.
We can then again see as with the multiple reader case above, the clear pattern of the attainable bandwidths
while memory writes can stay within L2 and L3 caches. Finally, when all requests go directly to DRAM, aggregate
write bandwidth settles to approximately 20 GB/sec.
7.2.5 Single Pair of Combined Reader and Writer Threads
In this experiment, a single pair of threads reads and writes memory segments of size varying from 10KBs to 200MBs.
The reader thread retrieves data belonging to the remote DRAM, whereas the writer thread writes into data belonging
to the local DRAM. The top curve in Fig. 20 (labeled “bandwidth CPU0–CPU0”) plots the observed bandwidth
when the pair of threads is running on CPU0 and they access data not accessed by other cores. When the L3 capacity
is overran the limitation becomes the QPI link one way bandwidth.
7.2.6 Multiple Pairs of Combined Reader and Writer Threads
This experiment is a combined 8-way experiment of the “single reader and writer pair” of threads above.
Michael E. Thomadakis 42 Texas A&M University
Intel Nehalem A Research Report
0
10k
20k
30k
40k
50k
60k
70k
80k
10k 100k 1M 10M 100M 1G
MB/s
memory size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.single-r1w1
bandwidth: CPU0 - CPU0
bandwidth: CPU0 - CPU1
bandwidth: CPU0 - CPU2
bandwidth: CPU0 - CPU3
bandwidth: CPU0 - CPU5
Figure 20: Bandwidth of a Single Pair of Reader and Writer Threads
Michael E. Thomadakis 43 Texas A&M University
Intel Nehalem A Research Report
0
100k
200k
300k
400k
500k
600k
10k 100k 1M 10M 100M 1G
MB/s
memory size
arch_x86_64.memory_bandwidth.C.pthread.SSE2.multiple-r1w1
bandwidth 0-31
Figure 21: Aggregate Bandwidth of 8 Pairs of Reader and Writer Threads
Michael E. Thomadakis 44 Texas A&M University
Intel Nehalem A Research Report
50
100
150
200
250
300
10k 100k 1M 10M 100M 1G
Clock Cycles
Memory Size
arch_x86_64.memory_latency.C.pthread.0.read
Latency CPU0 locally
Latency CPU0 accessing CPU1 memory
Latency CPU0 accessing CPU2 memory
Latency CPU0 accessing CPU3 memory
Latency CPU0 accessing CPU5 memory
Figure 22: Latency to Read a Data Block in Processor Cycles
7.3 Memory Latency
Michael E. Thomadakis 45 Texas A&M University
Intel Nehalem A Research Report
50
100
10k 100k 1M 10M 100M 1G
Time (nano-seconds)
Memory Size
arch_x86_64.memory_latency.C.pthread.0.read
Latency CPU0 locally
Latency CPU0 accessing CPU1 memory
Latency CPU0 accessing CPU2 memory
Latency CPU0 accessing CPU3 memory
Latency CPU0 accessing CPU5 memory
Figure 23: Latency to Read a Data Block in Seconds
The memory latency experiments measure the absolute amount of time in clock cycles and seconds it takes a
core to retrieve a new cache block directly from the L1, L2, L3 caches and finally from the DRAM. The target
memory blocks are initialized to be in the Exclusive state before access. We have only used 4KiB size pages for these
experiments.
Referring to Fig. 22 and Fig. 23, the bottom “red” curve plots the access time it takes CPU 0 access data residing
on the local DRAM block. The cost to retrieve memory clearly reflects the location of the specific block, namely the
cache level (L1, L2, L3) or the local DRAM.
The three curves labeled as “CPU0 accessing CPUMmemory”, for M= 1,2,3, reflect the time it takes to retrieve
a memory block which is already stored in Exclusive state in another core within the sane Nehalem chip. We can
readily notice that the cost is identical regardless of which particular core owns this block.
Finally, the top curve demonstrates the case where the block to be retrieved resides on the DRAM of the other
Nehalem chip. The latency increases considerably as the UnCore has to send cache coherence and data transfer
messages over the QPI link. The distance of the bottom (CPU0 to local DRAM) and top (CPU0 to remote DRAM)
curves trace the “NUMA effect”. A thread accessing data belonging to the remote DRAM incurs much higher
memory access cost.
One may readily notice that as cores attempt to retrieve at once data which far exceeding the capacity of the
cache memories, the remote and local DRAM access costs start getting close to each other. We believe that this
is due to the fact that the data TLB level 0 and 1 for 4KiB size pages are becoming overwhelmed by the excessive
number of VM addresses requested in sequence. Visit Section 6 for a discussion on the memory address translation
process in Intel64 architectures. There are only 64 and 512 entries, respectively, on the level 0 and 1 DTLBs. 64
DTLB entries cover up to 256 KiBs of memory and the 512 DTLB level 1 cover 2MiBs. As a core accesses memory
Michael E. Thomadakis 46 Texas A&M University
Intel Nehalem A Research Report
sizes which exceed these limits, the address translation hardware becomes the bottleneck as the page tables in DRAM
have to be accessed to locate complete page address information.
Wes strongly recommend that code which accesses lengthy blocks of data use large pages (known as “Huge Pages”
in the Linux world) as there is significant difference in the latencies for memory access.
8 Intel Turbo Boost Technology
Nehalem supports a number of power saving features which switch off the power to idle cores or other parts of the
system2. An interesting related feature is called “Turbo Boost Technology” (TBT) which not only dynamically turns
off unused processor cores but it can also increase the clock speed of the cores in use as long as the power and head
profiles of the chip is maintained. TBT will increase the frequency in steps of 133 MHz (to a maximum of three
steps or 400 MHz) as long as the processors’ predetermined thermal and electrical requirements are still met. For
example, with three cores active, a 2.26 GHz processor can run the cores at 2.4 GHz. With only one or two cores
active, the same processor can run those cores at 2.53 GHz. Similarly, a 2.93 GHz processor can run at 3.06 GHz or
even 3.33 GHz. When the cores are needed again, they are dynamically turned back on and the processor frequency
is adjusted accordingly. This feature can be enabled or disabled in the UEFI BIOS of each node.
References
[1] Intel R
64 and IA-32 Architectures Software Developer’s Manual Volume 1:Basic Architecture, Intel, May 2011.
[2] Intel R
64 and IA-32 Architectures Software Developer’s Manual Volume 3A: System Programming Guide, Part 1, Intel,
May 2011.
[3] Intel R
64 and IA-32 Architectures Software Developer’s Manual Volume 3B: System Programming Guide, Part 2, Intel,
May 2011.
[4] S. Gunther and R. Singhal, “Next generation intel R
microarchitecture (nehalem) family: Architectural insights and power
management,” in Intel Developer Forum. San Francisco: Intel, Mar. 2008.
[5] P. P. Gelsinger, “Intel architecture press briefing,” in Intel Developer Forum. San Francisco: Intel, Mar. 2008.
[6] R. Singhal, “Inside intel next generation nehalem microarchitecture,” in Intel Developer Forum. San Francisco: Intel,
Mar. 2008.
[7] Intel R
64 and IA-32 Architectures Optimization Reference Manual, Intel, May 2010.
[8] D. A. Patterson and J. L. Hennessy, Computer Organization and Design: The Hardware/Software Interface, 4th ed.
Morgan-Kaufmann Publishers Inc., 2009, iSBN: 978-0-12-374493-7.
Publisher’s URL: http://www.elsevierdirect.com/product.jsp?isbn=9780123744937.
[9] J. L. Hennessy and D. A. Patterson, Computer Architecture: A Quantitative Approach, 4th ed. Morgan-Kaufmann
Publishers Inc., 2007, iSBN: 978-0-12-370490-0.
Publisher’s URL: http://www.elsevierdirect.com/product.jsp?isbn=9780123704900.
[10] D. Levinthal, “Performance analysis guide for intelR
core(TM) i7 processor and intel xeon(TM) 5500 processors,” Intel,
Tech. Rep. Version 1.0, 2009.
[11] Intel, “TLBs, paging-structure caches, and their invalidation,” Intel, Application Note, 2008.
[12] G. Juckeland, S.B¨orner, M. Kluge, S. K¨olling, W. Nagel, S. Pfl¨uger, H. oding, S. Seidl, T. William, and
R. Wloch, “BenchIT – performance measurement and comparison for scientific applications,” in Parallel Computing
- Software Technology, Algorithms, Architectures and Applications, ser. Advances in Parallel Computing, F. P.
G.R. Joubert, W.E. Nagel and W. Walter, Eds. North-Holland, 2004, vol. 13, pp. 501–508. [Online]. Available:
http://www.sciencedirect.com/science/article/B8G4S-4PGPT1D-28/2/ceadfb7c3312956a6b713e0536929408
2This discussion is beyond hte scope of the present article and only short description is given.
Michael E. Thomadakis 47 Texas A&M University
Intel Nehalem A Research Report
[13] G. Juckeland, M. Kluge, W. E. Nagel, and S. Pfl¨uger, “Performance analysis with BenchIT: Portable, flexible, easy to
use,” in Proc. of the Inte’l Conf. on Quantitative Evaluation of Systems, QEST2004. Los Alamitos, CA, USA: IEEE
Computer Society, 2004, pp. 320–321.
[14] D. Hackenberg, D. Molka, and W. E. Nagel, “Comparing cache architectures and coherency protocols on x86-64 multicore
smp systems,” in Proceedings of the 42nd Annual IEEE/ACM International Symposium on Microarchitecture, ser. MICRO
42. New York, NY, USA: ACM, 2009, pp. 413–422. [Online]. Available: http://doi.acm.org/10.1145/1669112.1669165
Michael E. Thomadakis 48 Texas A&M University
... However, with the increase of the number of cores (thus CPUs) on commodity servers, the bus rapidly a bottleneck to memory access. To overcome this situation, in modern servers, each CPU is physically connected to its memory module(s), forming a node, and can access remote memory of other nodes in a cache coherent manner via a CPU interconnect [5,37,93,117] as shown on Figure 1. 5 Figure 1.5 -Example of a NUMA architecture. Nodes are linked via Interconnects (such as AMD HyperTransport [7] or Intel QuickPath Interconnect [38]). ...
Thesis
Virtual machine monitors (VMMs) or hypervisors play a crucial role in cloud computing platforms’ software stack. Their design and implementation significantly impact the performance, security, and robustness of cloud tenants applications. Hypervisors classified as Type-I are the most efficient, since they offer stronger isolation and better performance than Type-II pendant. In most of today’s Type-I virtualized systems (e.g., Xen or Hyper-V), the hypervisor relies on a privileged virtual machine (pVM). The pVM accomplishes work both for the hypervisor (e.g., VM life cycle management) and client VMs (I/O management). On uniform and non-uniform memory access (UMA & NUMA) architectures, this pVM-based architecture raises two challenging problems :• (1) pVM’s resource sizing (CPU + memory) and placement — Indeed, an inappropriate pVM sizing and resource placement impact guests’ application performance. It is a tricky issue since there is a tight correlation between pVM’s needs and guest activities. Existing solutions either propose static approaches which lead to over/under-provisioning or do notconsider resource placement in NUMA architectures. • (2) pVM’s fault tolerance — Being a central component, the pVM represents a critical component with a large blast radius in case of a failure. Existing approaches to improve the pVM’s fault tolerance provide limited resilience guarantees or prohibitive overheads. This dissertation presents several design changes brought to the pVM from architectural and logical perspectives to tackle these problems. Concretely, this thesis introduces : 1. Closer, a principle for designing a suitable OS for the pVM. Closer consists of respectively scheduling and allocating pVM’s tasks and memory as close to the target guest as possible. Closer being a dynamic approach, alleviates the need to size the pVM and handles its resource placement in NUMA architectures with its locality strategy. 2. Two new mechanisms that reduce the overhead of page flipping (an efficient scheme used in network I/O virtualization) when used on NUMA architectures. By carefully selecting pVM pages for page flipping depending on their location, the latter mechanisms achieve better performance than the current network virtualization protocol. 3. A set of three design principles (disaggregation, specialization, and pro-activity) and optimized implementation techniques for building a resilient pVM without sacrificing guest application performance.We build prototypes of pVM-based hypervisors (relying on the Xen hypervisor) that implements all the principles above. We validate the effectiveness of our prototypes by conducting several evaluations with a series of benchmarks. The results obtained shows better performance than state-of-the-art approaches and low overhead.This dissertation highlights the critical role of the pVM in a virtualized environment and shows that it requires more attention from the research community.
... Initially, I validated the execution of the COTSon simulator against the Intel i7700 Core [75], by using the architecture parameters reported in Table 2.3. About the AArch64 architecture, we relied on the AXIOM-Board hardware specification, which is a single computer board developed during the AXIOM-Project at the beginning of 2017 [61,35,36,38,4]. ...
Thesis
Full-text available
The end of Dennard scaling [1], Moore’s law [2], and the resulting difficulty of increasing clock frequency forced the engineering community to shift to the multi/many-core processors and multi-node systems as an alternative way to improve performance. An increased core number benefits many workloads, but programming limitations still reduce the performance due to not fully exploited parallelism. From this perspective, new execution models are arising to overcome such limitations to scale up performance. Execution models like Data-Flow can take advantage of the full parallelism, thanks to the possibility of creating many asynchronous threads that can run in parallel. These threads may encapsulate the data to be processed, their dependencies, and, once completed, write their output for other threads. Data-Flow Threads (DF-Threads) is a novel Data-Flow execution model for mapping threads on local or distributed cores transparently to the programmer [3]. This model is capable of being parallelized massively among different cores, and it handles even hundreds of thousands or more Data-Flow threads with their associated data regions. Further implementation and evaluation of the DF-Threads model (previously proposed by R.Giorgi [3]) are presented in this thesis. The proposed model can be able to exploit the full parallelism of modern heterogeneous embedded architectures (e.g., the AXIOM-Board [4]). The work relies on introducing the ”Data-Flow engine” (DF-Engine), which is able to accelerate the function execution and spawn many asynchronous, data-driven threads across several general-purpose cores of a multi-core/node system. The DF-Engine can be placed either in software or directly implemented at the hardware level by using a heterogeneous architecture (e.g., the AXIOM-Board). The DF-Engine can handle the creation, the thread-dependency, and the locality of many fine-grain threads, leaving the general-purpose core focusing only on the execution of the threads. This implementation is a hybrid Data-Flow - von-Neumann model, which harnesses the parallelism and data synchronization inherently to the Data-Flow, and yet maintains the programmability of the von-Neumann model. Starting from the DF-Threads execution model, we analyzed the tradeoffs of a minimalistic API to enable an efficient implementation, which can distribute the DF-Threads either locally across the core of a single multi-core system or/and across the remote cores of a cluster. Implement and evaluate the proposed model directly on a real architecture requires time, resources, and effort. Therefore, the design has been preliminarily evaluated in a simulation framework, and then the validated model has been gradually migrated into a real board in collaboration with my research group. The simulation framework presented in this thesis is based on the COTSon simulator [5] and on a set of tools named ”MY Design Space Exploration” (MYDSE) [6], which has been implemented and adopted by our research group. Then, we explain how the validation phase of the simulation framework has been performed against real architecture like x86 64 and AArch64. Moreover, we analyzed the impact of different Linux distributions on the execution. Afterward, we explain the workflow adopted to migrate the design of the DF-Threads execution model from the COTSon simulator to a High-Level Synthesis framework (e.g., Xilinx HLS), targeting a heterogeneous architecture such as the AXIOM-Board [4]. We used a driving example that models a two-way associative cache to demonstrate the simplicity and rapidness of our framework in migrating the design from the COTSon simulator to the HLS framework. The methodology has been adopted in the context of the AXIOM project [7], which helped our research team in reducing the development time from days/weeks to minutes/hours. In the end, we present the evaluation of the proposed DF-Threads execution model. We are interested in stressing and analyzing the efficiency of the DF-Engine with thousands or more Data-Flow threads. For this goal, we decided to use the Recursive Fibonacci algorithm, which gives us the possibility to generate such a high number of threads easily. Moreover, we want to study the behavior of the execution model with data-intensive applications for evaluating the performance with memory operations and data movements. For this reason, we adopted the Matrix Multiplication benchmark, which is the main computational kernel of widely used applications (e.g., Smart Home Living, Smart Video Surveillance, Artificial Intelligence). The proposed design has been evaluated against OpenMPI, which is typically adopted for cluster programming, and against CUDA, a parallel programming language for GPUs. DF-Threads achieve better performance-per-core compared to both OpenMPI and CUDA. In particular, OpenMPI exhibits much more Operating System (OS) kernel activity with respect to DF-Threads. This OS activity slows down the OpenMPI performance. If we consider the delivered GFLOPS per core, DF-Threads is also competitive with respect to CUDA.