CSC/ECE 506 Fall 2007/wiki1 4 a1: Difference between revisions

From Expertiza_Wiki
Jump to navigation Jump to search
 
(133 intermediate revisions by 2 users not shown)
Line 1: Line 1:
== Architectural Trends ==
== Architectural Trends ==
=== VLIW(Very Long Instruction Word) ===
[[Image:MIPSR10000.jpg|thumb|right|300px|Fig.1 MIPS R10000 Block Diagram (From Fig. 2 of [3])]]
Ours.
[[Image:IntelMoorsLaw.jpg|thumb|right|300px|Fig.2 The number of trnasistors on a chip of Intel]]
VLIW is based on statically sceduled processing which is performed by machine dependent compiler. VLIW has mutiple execution pipelines which are aimed specialized function. The compiler analyzes the programmer's instructions and groups multiple independent instructions into a large package. Hence, the length of instruction can be logger and those architecture is called VLIW (Very Long Instruction Word).
Feature size means the minimum size of transistors or a wire width which are used for connectiong transistors and other circuit components. Feature sizes have dramatically decreased from 10 microns in 1971 to 0.18 microns in 2001. These advanced integrated circuit processes allowed the integration of one billion transistors on a single chip and enabled more complicated and faster microprocessor architecure which have evolved to the direction of increasing parallelism; [http://en.wikipedia.org/wiki/Instruction_level_parallelism ILP] and [http://en.wikipedia.org/wiki/Thread_level_parallelism TLP]. With respect to microprocessor architecture, as superscalar processor prevails, several additional exploitable architectures were also proposed during past 10 years as other past decades did. Based on superscalar architecture, VLIW, superspeculative, simultaneous multithreading, chip multiprocessor and so on were proposed and explored. These techniques tried to overcome the control and data hazard as deep pipelining and multiple issue overwhelms as well as to maximize the throughput of computing by TLP.
Ours.


For example, MIPS R10000 is a superscalar processor executed by out of order manner, which has 6.8 million transistors on 16.64mm x 17.934 mm(298mm<sup>2</sup>) dimension using 0.35um process. It fetches 4 instructions simultaneously and has total of 6 pipelines; 5 pipe lines for execution and 1 pipe line for fetching and decoding. Each execution pipelines can be categorized into 3 kinds of execution - integer, float and load/store.


one VLIW instruction encodes multiple operations; specifically, one instruction encodes at least one operation for each execution unit of the device. For example, if a VLIW device has five execution units, then a VLIW instruction for that device would have five operation fields, each field specifying what operation should be done on that corresponding execution unit. To accommodate these operation fields, VLIW instructions are usually at least 64 bits in width, and on some architectures are much wider.


The instruction scheduling logic that makes a superscalar processor is just boolean logic. In the early 1990s, a significant innovation was to realize that the coordination of a multiple-ALU computer could be moved into the compiler, the software that translates a programmer's instructions into machine-level instructions.


This type of computer is called a very long instruction word (VLIW) computer.
----


Statically scheduling the instructions in the compiler (as opposed to letting the processor do the scheduling dynamically) can reduce CPU complexity. This can improve performance, reduce heat, and reduce cost.


Unfortunately, the compiler lacks accurate knowledge of runtime scheduling issues. Merely changing the CPU core frequency multiplier will have an effect on scheduling. Actual operation of the program, as determined by input data, will have major effects on scheduling. To overcome these severe problems a VLIW system may be enhanced by adding the normal dynamic scheduling, losing some of the VLIW advantages.


Static scheduling in the compiler also assumes that dynamically generated code will be uncommon. Prior to the creation of Java, this was in fact true. It was reasonable to assume that slow compiles would only affect software developers. Now, with JIT virtual machines for Java and .net, slow code generation affects users as well.
=== VLIW ===
VLIW(Very Long Instruction Word) is one way to expedite ILP under multiple-issue processors. Multiple-issue processors are attainable by two basics - superscalar and VLIW. The big difference between superscalar and VLIW is located on the scheduling method of instructions. Whlie superscalar processors issue multiple numbers of instructions per clock, which are scheduled either statically or dynamically, VLIWs issue statically sceduled instructions by the compiler. Both superscalar and VLIW have multiple and independent functional units.


There were several unsuccessful attempts to commercialize VLIW. The basic problem is that a VLIW computer does not scale to different price and performance points, as a dynamically scheduled computer can. Another issue is that compiler design for VLIW computers is extremely difficult, and the current crop of compilers (as of 2005) don't always produce optimal code for these platforms.
VLIW processor's compiler analyzes the programmer's instructions and then groups multiple independent instructions into a large packaged instruction. VLIW issues a fixed number of instructions, the format of which can be either one large instruction or a fixed instruction packet with the parallelism.
 
To look into the inside of VLIW operation, assume the below example code for MIPS[1].
 
for (i=1000; i>0; i=i-1)  x[i] = x[i] + s;
 
The standard MIPS code looks like this:
 
[[Image:simpleMIPS.jpg]]
 
If loop-unrolling and scheduling the code are applied, then
 
[[Image:loopunrollMIPS.jpg]]
 
it takes 14 cycles for loop body.
 
If VLIW instructions are used, then
 
[[Image:VLIW.jpg]]
 
it takes 9 cycles assuming 5 execution pipelines.
 
MIPS R10000 is also a good example. It has 2 integer functional units and 3 types of operands. Hence, the compiler can generate one instruction which contains 3 integer operations with the corresponding operands to each operation. Yet another example of VLIW is i860 of Trimedia.
 
 
 
----


Also, VLIW computers optimise for throughput, not low latency, so they were not attractive to the engineers designing controllers and other computers embedded in machinery. The embedded systems markets had often pioneered other computer improvements by providing a large market that did not care about compatibility with older software.


In January 2000, a company called Transmeta took the interesting step of placing a compiler in the central processing unit, and making the compiler translate from a reference byte code (in their case, x86 instructions) to an internal VLIW instruction set. This approach combines the hardware simplicity, low power and speed of VLIW RISC with the compact main memory system and software reverse-compatibility provided by popular CISC.


=== Multi-threading ===
=== Multi-threading ===
Current designs work best when the computer is running only a single program, however nearly all modern operating systems allow the user to run multiple programs at the same time. For the CPU to change over and do work on another program requires expensive context switching. In contrast, multi-threaded CPUs can handle instructions from multiple programs at once.
[[Image:SMTEx.jpg|thumb|right|300px|Fig.3 Four different approaches of using issue slots in superscalar processor (Redrawn from Fig 6.44 of [1])]]
Multi-threading enables exploiting thread-level parallelism(TLP) within a single processor. It allows multiple threads to share the functional units of a single processor by an overlapping manner. For this sharing, the processor has to maintain the duplicated state information of each thread-register file, PC, page table and so on. In addition, the processor can switch the different thread quickly.
 
For attaining multi-threading, there are two basic approaches; fine-grained multi-threading and coarse-grained multi-threading. The former switches each instruction between multiple interleaved threads. For this interleaving, the processor can switch threads on every clock cycle. The advantage of this architecture can prohibit stalling, because other instructions from other threads can be performed when one thread stalls. The disadvantage makes slow down the individual thread's execution, because even though the instruction is ready to be executed, it can be interleaved by another thread's instruction.
 
The latter switches threads when it meets the stall only with a high cost. This policy reduces unnecessary switching of thread, so that the individual thread does not need to slow down its execution contrary to the fine-grained case. However, it has the cost when switching occurs to fill the pipeline. This kind of processor issues instructions from a single thread, although it switches the running thread. If the stall occurs, the pipeline is empty. Then, in order to execute a new thread instead of stalled thread, the pipeline has to be filled, which results in the cost.


To do this, such CPUs include several sets of registers. When a context switch occurs, the contents of the "working registers" are simply copied into one of a set of registers for this purpose.
The Simultaneous multithreading (SMT) is a kind of multithreading that uses the resources of a multiple-issue, dynamically scheduled processor to exploit TLP. At the same time it exploits ILP using the issue slots in a single clock cycle. Figure 3 shows the comparison between three kinds of multi-threading in addition to a superscalar processor.


Such designs often include thousands of registers instead of hundreds as in a typical design. On the downside, registers tend to be somewhat expensive in chip space needed to implement them. This chip space might otherwise be used for some other




Computer architects have become stymied by the growing mismatch in CPU operating frequencies and DRAM access times. None of the techniques that exploited instruction-level parallelism within one program could make up for the long stalls that occurred when data had to be fetched from main memory. Additionally, the large transistor counts and high operating frequencies needed for the more advanced ILP techniques required power dissipation levels that could no longer be cheaply cooled. For these reasons, newer generations of computers have started to exploit higher levels of parallelism that exist outside of a single program or program thread.
----


This trend is sometimes known as throughput computing. This idea originated in the mainframe market where online transaction processing emphasized not just the execution speed of one transaction, but the capacity to deal with massive numbers of transactions. With transaction-based applications such as network routing and web-site serving greatly increasing in the last decade, the computer industry has re-emphasized capacity and throughput issues.


One technique of how this parallelism is achieved is through multiprocessing systems, computer systems with multiple CPUs. Once reserved for high-end mainframes and supercomputers, small scale (2-8) multiprocessors servers have become commonplace for the small business market. For large corporations, large scale (16-256) multiprocessors are common. Even personal computers with multiple CPUs have appeared since the 1990s.


With further transistor size reductions made available with semiconductor technology advances, multicore CPUs have appeared where multiple CPUs are implemented on the same silicon chip. Initially used in chips targeting embedded markets, where simpler and smaller CPUs would allow multiple instantiations to fit on one piece of silicon. By 2005, semiconductor technology allowed dual high-end desktop CPUs CMP chips to be manufactured in volume. Some designs, such as Sun Microsystems' UltraSPARC T1 have reverted back to simpler (scalar, in-order) designs in order to fit more processors on one piece of silicon.
=== Multi-core ===
[[Image:Smithfield_die_med.jpg|thumb|right|80px|Fig.4 Intel® Pentium® processor Extreme Edition processor die [7]]]
Multi-core CPUs have multiple numbers of CPU cores on a single die. They are connected to each other through a shared L2 or L3 cache, or a glue logic like switch and bus on a die. Every CPU core on a die shares interconnect components with which to interface to other processors and the rest of the system. These components include a FSB (Front Side Bus), a memory controller, a cache coherent link to other processors, and a non-coherent link to the southbridge and I/O devices. The advantages of multi-core chips are power-efficiency and simplicity around the processors. Since multiple processors are packed into a single die, the glue logics which are required to connect to each processor are also packed into a die. It saves power and simplifies auxilary circuits than coupled processors, which need PCB circuits.
Intel Pentium Extreme, Coreduo and Coreduo2 are good examples of multi-core processors.
Intel Xeon X7300 series has quad-core in a single die with 65nm processing.


Another technique that has become more popular recently is multithreading. In multithreading, when the processor has to fetch data from slow system memory, instead of stalling for the data to arrive, the processor switches to another program or program thread which is ready to execute. Though this does not speed up a particular program/thread, it increases the overall system throughput by reducing the time the CPU is idle.


Conceptually, multithreading is equivalent to a context switch at the operating system level. The difference is that a multithreaded CPU can do a thread switch in one CPU cycle instead of the hundreds or thousands of CPU cycles a context switch normally requires. This is achieved by replicating the state hardware (such as the register file and program counter) for each active thread.
----


A further enhancement is simultaneous multithreading. This technique allows superscalar CPUs to execute instructions from different programs/threads simultaneously in the same cycle.


=== Multi-core ===
Multi-core CPUs are typically multiple CPU cores on the same die, connected to each other via a shared L2 or L3 cache, an on-die bus, or an on-die crossbar switch. All the CPU cores on the die share interconnect components with which to interface to other processors and the rest of the system. These components may include a front side bus interface, a memory controller to interface with DRAM, a cache coherent link to other processors, and a non-coherent link to the southbridge and I/O devices. The terms multi-core and MPU (which stands for Micro-Processor Unit) have come into general usage for a single die that contains multiple CPU cores


=== Speculative Execution ===
=== Speculative Execution ===
One problem with an instruction pipeline is that there are a class of instructions that must make their way entirely through the pipeline before execution can continue. In particular, conditional branches need to know the result of some prior instruction before "which side" of the branch to run is known. For instance, an instruction that says "if x is larger than 5 then do this, otherwise do that" will have to wait for the results of x to be known before it knows if the instructions for this or that can be fetched.
While trying to get more ILP, managing control dependencies becomes more important but more burden. To reduce the cost of stall because of branch, branch prediction techinque is applied for the instruction fetching stage. However, for the processor which executes multiple instructions per clock, more than just accurate prediction are required. To speculate is to act on these predictions; fetch and execute instructions from the predicted path.[12]


For a small four-deep pipeline this means a delay of up to three cycles — the decode can still happen. But as clock speeds increase the depth of the pipeline increases with it, and modern processors may have 20 stages or more. In this case the CPU is being stalled for the vast majority of its cycles every time one of these instructions is encountered.
[[Image:speculative.jpg]]


The solution, or one of them, is speculative execution, also known as branch prediction. In reality one side or the other of the branch will be called much more often than the other, so it is often correct to simply go ahead and say "x will likely be smaller than five, start processing that". If the prediction turns out to be correct, a huge amount of time will be saved. Modern designs have rather complex prediction systems, which watch the results of past branches to predict the future with greater accuracy.
Under speculative execution, fetch, issue, and execute instructions are performed as if branch predictions were always correct. When misprediction occurs, the recovery mechanism handles this situation. If the processor meets a branch, it predicts the branch target and follows that path as well as does checkpoint. While checkpointing, the processor duplicates the copy of information such as register files and control information and another possible branch target and so on. If the prediction is correct, the processor reclaims the stored information for use by new predicted branches. But if the prediction is incorrect, it resotres the execution information from the corresponding checkpoint. There are examples like PowerPC 603/604/G3/G4, MIPS R10000/R12000, Intel Pentium II/III/4, Alpha 21264, and AMD K5/K6/Athlon.  


=== Real Processors ===


== Updated Figure 1.8 & Figure 1.9 ==
 
----
 
== Updated Figure 1.8 & 1.9 ==
[[Image:fig18.jpg|frame|Figure 1.8 Number of processors in fully configured commercial bus-based shared memory multiprocessors]]
 
Figure 1.8 of our book has been updated to incorporate trends from 2000 to the present. SGI Origin 3000 series were reintroduced as Origin 3400 and Origin 3900 in year 2000 and 2003, respectively. Sun introduced even more powerful enterprise servers than E10000, which are E15000 in year 2002, E20000 and E25000 in year 2006. HP's high-end supercomputer 9000 Superdome with 16, 32, and 64 processors are released this year(2007).
 
 
 
[[Image:fig19.jpg|frame|Figure 1.9 Bandwidth of the shared memory bus in commercial multiprocessors(Y-axis is log-scaled)]]
 
Figure 1.9 shows the bandwidth of shared memory bus of those servers introduced in figure 1.8, which are SGI Origin 3000 series, SUN Enterprise 15K, 20K, and 25K as well as IBM p5 590 and HP9000 Superdome. In the case of Sun E25K, the bandwidth available is 43.2 GBps and the aggregated bandwith exceeds 100GBps. Origin 3900 has 12.8 GBps bandwidth and the aggregate bandwidth of 172.8 GBps.
 
 
----
 
== References ==
== References ==
[1] John L. Hennessy, David A. Patterson, "Computer Architecture: A Quantitative Approach" 3rd Ed., Morgan Kaufmann, CA, USA
[2] CE Kozyrakis, DA Patterson, "A new direction for computer architecture research",
Computer Volume 31 Issue 11, IEEE, Nov 1998, pp24-32
[3] K.C. Yeager, "The MIPS R10000 Superscalar Microprocessor", IEEE Micro Volume 16 Issue 2, Apr. 1996, pp28-41
[4] Geoff Koch, "Discovering Multi-Core: Extending the Benefits of Moore’s Law", Technology@Intel Magazine, Jul 2005, pp1-6
[5] Richard Low, "Microprocessor trends:multicore, memory, and power developments", Embedded Computing Design, Sep 2005
[6] Artur Klauser, "Trends in High-Performance Microprocessor Design", Telematik 1, 2001
[7] http://www.intel.com & http://www.intel.com/pressroom/kits/pentiumee
[8] http://www.alimartech.com/9000_servers.htm
[9] http://www.sun.com/servers/index.jsp?gr0=cpu&fl0=cpu4&gr1=
[10] http://www.sgi.com/pdfs/3867.pdf
[11] http://www-03.ibm.com/systems/p/hardware/highend/590/index.html
[12] Eric Rotenberg, ECE721 Advanced Microarchitecture lecture notes, NCSU, 2007

Latest revision as of 02:28, 11 September 2007

Architectural Trends

Fig.1 MIPS R10000 Block Diagram (From Fig. 2 of [3])
Fig.2 The number of trnasistors on a chip of Intel

Feature size means the minimum size of transistors or a wire width which are used for connectiong transistors and other circuit components. Feature sizes have dramatically decreased from 10 microns in 1971 to 0.18 microns in 2001. These advanced integrated circuit processes allowed the integration of one billion transistors on a single chip and enabled more complicated and faster microprocessor architecure which have evolved to the direction of increasing parallelism; ILP and TLP. With respect to microprocessor architecture, as superscalar processor prevails, several additional exploitable architectures were also proposed during past 10 years as other past decades did. Based on superscalar architecture, VLIW, superspeculative, simultaneous multithreading, chip multiprocessor and so on were proposed and explored. These techniques tried to overcome the control and data hazard as deep pipelining and multiple issue overwhelms as well as to maximize the throughput of computing by TLP.

For example, MIPS R10000 is a superscalar processor executed by out of order manner, which has 6.8 million transistors on 16.64mm x 17.934 mm(298mm2) dimension using 0.35um process. It fetches 4 instructions simultaneously and has total of 6 pipelines; 5 pipe lines for execution and 1 pipe line for fetching and decoding. Each execution pipelines can be categorized into 3 kinds of execution - integer, float and load/store.



VLIW

VLIW(Very Long Instruction Word) is one way to expedite ILP under multiple-issue processors. Multiple-issue processors are attainable by two basics - superscalar and VLIW. The big difference between superscalar and VLIW is located on the scheduling method of instructions. Whlie superscalar processors issue multiple numbers of instructions per clock, which are scheduled either statically or dynamically, VLIWs issue statically sceduled instructions by the compiler. Both superscalar and VLIW have multiple and independent functional units.

VLIW processor's compiler analyzes the programmer's instructions and then groups multiple independent instructions into a large packaged instruction. VLIW issues a fixed number of instructions, the format of which can be either one large instruction or a fixed instruction packet with the parallelism.

To look into the inside of VLIW operation, assume the below example code for MIPS[1].

for (i=1000; i>0; i=i-1) x[i] = x[i] + s;

The standard MIPS code looks like this:

If loop-unrolling and scheduling the code are applied, then

it takes 14 cycles for loop body.

If VLIW instructions are used, then

it takes 9 cycles assuming 5 execution pipelines.

MIPS R10000 is also a good example. It has 2 integer functional units and 3 types of operands. Hence, the compiler can generate one instruction which contains 3 integer operations with the corresponding operands to each operation. Yet another example of VLIW is i860 of Trimedia.



Multi-threading

Fig.3 Four different approaches of using issue slots in superscalar processor (Redrawn from Fig 6.44 of [1])

Multi-threading enables exploiting thread-level parallelism(TLP) within a single processor. It allows multiple threads to share the functional units of a single processor by an overlapping manner. For this sharing, the processor has to maintain the duplicated state information of each thread-register file, PC, page table and so on. In addition, the processor can switch the different thread quickly.

For attaining multi-threading, there are two basic approaches; fine-grained multi-threading and coarse-grained multi-threading. The former switches each instruction between multiple interleaved threads. For this interleaving, the processor can switch threads on every clock cycle. The advantage of this architecture can prohibit stalling, because other instructions from other threads can be performed when one thread stalls. The disadvantage makes slow down the individual thread's execution, because even though the instruction is ready to be executed, it can be interleaved by another thread's instruction.

The latter switches threads when it meets the stall only with a high cost. This policy reduces unnecessary switching of thread, so that the individual thread does not need to slow down its execution contrary to the fine-grained case. However, it has the cost when switching occurs to fill the pipeline. This kind of processor issues instructions from a single thread, although it switches the running thread. If the stall occurs, the pipeline is empty. Then, in order to execute a new thread instead of stalled thread, the pipeline has to be filled, which results in the cost.

The Simultaneous multithreading (SMT) is a kind of multithreading that uses the resources of a multiple-issue, dynamically scheduled processor to exploit TLP. At the same time it exploits ILP using the issue slots in a single clock cycle. Figure 3 shows the comparison between three kinds of multi-threading in addition to a superscalar processor.



Multi-core

Fig.4 Intel® Pentium® processor Extreme Edition processor die [7]

Multi-core CPUs have multiple numbers of CPU cores on a single die. They are connected to each other through a shared L2 or L3 cache, or a glue logic like switch and bus on a die. Every CPU core on a die shares interconnect components with which to interface to other processors and the rest of the system. These components include a FSB (Front Side Bus), a memory controller, a cache coherent link to other processors, and a non-coherent link to the southbridge and I/O devices. The advantages of multi-core chips are power-efficiency and simplicity around the processors. Since multiple processors are packed into a single die, the glue logics which are required to connect to each processor are also packed into a die. It saves power and simplifies auxilary circuits than coupled processors, which need PCB circuits. Intel Pentium Extreme, Coreduo and Coreduo2 are good examples of multi-core processors. Intel Xeon X7300 series has quad-core in a single die with 65nm processing.



Speculative Execution

While trying to get more ILP, managing control dependencies becomes more important but more burden. To reduce the cost of stall because of branch, branch prediction techinque is applied for the instruction fetching stage. However, for the processor which executes multiple instructions per clock, more than just accurate prediction are required. To speculate is to act on these predictions; fetch and execute instructions from the predicted path.[12]

Under speculative execution, fetch, issue, and execute instructions are performed as if branch predictions were always correct. When misprediction occurs, the recovery mechanism handles this situation. If the processor meets a branch, it predicts the branch target and follows that path as well as does checkpoint. While checkpointing, the processor duplicates the copy of information such as register files and control information and another possible branch target and so on. If the prediction is correct, the processor reclaims the stored information for use by new predicted branches. But if the prediction is incorrect, it resotres the execution information from the corresponding checkpoint. There are examples like PowerPC 603/604/G3/G4, MIPS R10000/R12000, Intel Pentium II/III/4, Alpha 21264, and AMD K5/K6/Athlon.


Updated Figure 1.8 & 1.9

Figure 1.8 Number of processors in fully configured commercial bus-based shared memory multiprocessors

Figure 1.8 of our book has been updated to incorporate trends from 2000 to the present. SGI Origin 3000 series were reintroduced as Origin 3400 and Origin 3900 in year 2000 and 2003, respectively. Sun introduced even more powerful enterprise servers than E10000, which are E15000 in year 2002, E20000 and E25000 in year 2006. HP's high-end supercomputer 9000 Superdome with 16, 32, and 64 processors are released this year(2007).


Figure 1.9 Bandwidth of the shared memory bus in commercial multiprocessors(Y-axis is log-scaled)

Figure 1.9 shows the bandwidth of shared memory bus of those servers introduced in figure 1.8, which are SGI Origin 3000 series, SUN Enterprise 15K, 20K, and 25K as well as IBM p5 590 and HP9000 Superdome. In the case of Sun E25K, the bandwidth available is 43.2 GBps and the aggregated bandwith exceeds 100GBps. Origin 3900 has 12.8 GBps bandwidth and the aggregate bandwidth of 172.8 GBps.


References

[1] John L. Hennessy, David A. Patterson, "Computer Architecture: A Quantitative Approach" 3rd Ed., Morgan Kaufmann, CA, USA

[2] CE Kozyrakis, DA Patterson, "A new direction for computer architecture research", Computer Volume 31 Issue 11, IEEE, Nov 1998, pp24-32

[3] K.C. Yeager, "The MIPS R10000 Superscalar Microprocessor", IEEE Micro Volume 16 Issue 2, Apr. 1996, pp28-41

[4] Geoff Koch, "Discovering Multi-Core: Extending the Benefits of Moore’s Law", Technology@Intel Magazine, Jul 2005, pp1-6

[5] Richard Low, "Microprocessor trends:multicore, memory, and power developments", Embedded Computing Design, Sep 2005

[6] Artur Klauser, "Trends in High-Performance Microprocessor Design", Telematik 1, 2001

[7] http://www.intel.com & http://www.intel.com/pressroom/kits/pentiumee

[8] http://www.alimartech.com/9000_servers.htm

[9] http://www.sun.com/servers/index.jsp?gr0=cpu&fl0=cpu4&gr1=

[10] http://www.sgi.com/pdfs/3867.pdf

[11] http://www-03.ibm.com/systems/p/hardware/highend/590/index.html

[12] Eric Rotenberg, ECE721 Advanced Microarchitecture lecture notes, NCSU, 2007

Retrieved from "https://wiki.expertiza.ncsu.edu/index.php?title=CSC/ECE_506_Fall_2007/wiki1_4_a1&oldid=3467"