The Mother of All CPU Charts Part 1
When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.
Introduction
UPDATE: MARCH 22, 2005:
Check out our overview of CPU history in Part 1, with Part 2 detailing AMD’s path to Athlon64, plus 3,300 benchmarks for CPUs since 1995. Now, we’ve updated the benchmarks to include the Intel Pentium 600 series and Extreme Edition .
From 1995 To 2005: 116 CPUs From 100 MHz To 3800 MHz
Ladies and gentlemen, fasten your seatbelts, because here we go. After 12 weeks of grueling and meticulous testing in our Munich THG lab, our biggest CPU test of all time is complete. Marathon, all-night monitoring sessions, system hiccups and crashes and the logistics involved in coordinating such a feat often short-circuited our caffeine-addled nerves. But then again, this project was about more than just tallying up CPU benchmarks. Indeed, it was fascination (or was it an obsession?) with the course taken by CPU technology in the past 11 years that motivated THG to conduct a comparative test of this magnitude. In Part 1 of this series, we take a look at the beginnings of the CPU and move up through the very latest from Intel. Tomorrow, we continue with a focus on AMD’s development in history, and bring you the results of our benchmark tests as well.
Many readers still remember the days when a PC system with Pentium 100 and 16 MB memory was «state of the art» and counted as the ultimate equipment for an enthusiastic student. That was in 1994, when this computer sold for just under $3000. Today the ex-student may work in an IT company or has his own company and makes decisions on investments. How times change.
Overall, 116 different processors dating from 1995 to today are being subjected to a uniform, comprehensive testing procedure. This monster comparison can be seen as a journey through time that very clearly reflects how brilliantly the performance of chips and, ultimately, of PC systems has changed over a very short period. But performance is not the only thing that many users take for granted when spending their money.
What is not visible to many are is the multitude of technical changes that components and standards have undergone. The test will provide insights into this topic as well.
Starting with the CPU platforms — from Socket 5 from 1995 to the current Socket 775 — no less than seven different evolutionary stages were observed in the case of processor manufacturer Intel alone. During the same period, AMD launched five platforms, from Socket 7 to Socket 939.
With subdued pride, we can state that this is the most comprehensive CPU comparison and benchmark marathon on the Internet. The reader will be able to directly compare the performance of a total of 116 processors and analyze them relying on 3,330 benchmark values in various testing categories. Above all, users who entertain the idea of replacing their old PC system with a new one will see exactly what kind of performance increase they can expect in practice. This is possible — even though over the years the PC’s architecture has been transformed with respect to memory, chipset and platform — because we did all of our benchmark tests using WindowsXP.
- 1
Current page:
Introduction
Next Page From 1995 To 2005, Continued
Get instant access to breaking news, in-depth reviews and helpful tips.
Contact me with news and offers from other Future brandsReceive email from us on behalf of our trusted partners or sponsors
Tom’s Hardware’s dedicated news crew consists of both freelancers and staff with decades of experience reporting on the latest developments in CPUs, GPUs, super computing, Raspberry Pis and more.
Mobile CPU Chart Update and Forecast
When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.
Our CPU Charts, have listed current and many outdated processors for easy comparison for several years. However, at the beginning we could not include mobile processors for technical reasons. While state-of-the-art desktop motherboards will typically support the entire desktop processor portfolio from AMD or Intel, it had been very difficult to find motherboards that support Intel’s socket 479 and especially, socket S1 AMD mobile processors.
But then, at last, we found a motherboard that could accommodate the AMD Turion 64 X2, so we could compare it directly against Core 2 Duo mobile CPUs as well as other AMD and Intel processors in our charts.
All of our Interactive Charts are based on platforms that allow us to test various components as they become available. Our 3.5” Hard Drive Charts are the best example: They are based on our reference storage test system that we used for almost every storage review. They have also not undergone major changes for years in order to offer a more accurate comparison between the different hard drives.
While all desktop-type components such as processors and graphics cards can easily be tested on desktop platforms, testing mobile components is not as simple due to limited platform options. It’s certainly possible to use a single laptop to test a large range of AMD or Intel processors. However, it would not be possible to use the same desktop hard drive and desktop graphics solutions we utilized for our desktop processor testing for mobile devices.
Finding an AMD and an Intel laptop that could handle components such as hard drives and graphics cards that were similar enough to each other was even more difficult.
So-called mobile-on-desktop solutions (MoDT) turned out to be the solution. These desktop-type components, which are based on mobility-optimized and thus energy-efficient hardware, have been around for several years now. However, they have mostly been designed for Intel processors because of Intel’s one-stop Centrino platform strategy. Although many people still believe that Centrino refers to a processor, it is actually the name for the entire platform package consisting of a mobile processor, a mobility chipset and a wireless unit. Some manufacturers such as Aopen even run dedicated websites on the topic. Please check out our articles Aopen releases Core Duo to the desktop and Quick and Quiet: Pentium D Desktop Boards by Aopen and DFI for more details.
The situation on the AMD side has been tricky, because the firm had refused to be a chipset provider for many years.
As a consequence, the chip maker had to rely on third-party chipset solutions from ATI, Nvidia, SiS or VIA. Therefore, the product selection of motherboards for AMD’s mobile socket S1 has been minimal, to say the least. Motherboards typically did not support an x16 PCI Express slot for graphics. Things changed with the acquisition of ATI, and AMD has now fully acknowledged the importance of offering entire platform solutions instead of only single components.
The solution we found comes from the German vendor Kontron, and it’s a MiniITX motherboard based on AMD’s 690 chipset. The KT690/mITX is an industrial-type motherboard that offers an impressive level of component integration, and it allowed us to finally compare Intel’s Core 2 Duo mobile lineup to the AMD Turion 64 X2 family based on the same set of test components.
- 1
Current page:
Mobile CPU Charts Updated
Next Page AMD’s New Mobile Platform “Puma”
Get instant access to breaking news, in-depth reviews and helpful tips.
Contact me with news and offers from other Future brandsReceive email from us on behalf of our trusted partners or sponsors
KNOW INTUIT | Lecture | Pipeline organization of the microprocessor
< Lecture 18 || Lecture 9 : 123
Abstract: The purpose of the lecture: to study the principles of the pipeline organization of microprocessors, the impact of conflicts on the operation of the pipeline and ways to reduce this impact.
Keywords: processor, software, instruction, pipeline, instruction fetch, instruction decoder, decoding, operand, cost, TID, overhead, clock time, execution time, analysis, performance, pipeline cycle, operations, pentium, pipeline length , clock speed, space, actual performance, hardware, instruction steps, pipeline bubble, collision, storage, write, memory, data store, access scheme, itanium, ALU, parallel processing, superscalar, buffer, instruction type, performance, pipelining, branch instructions, value, program counter, conditional branch, address, compiler, slot, microprocessor, definition, execution, invalidation, bogus, Instruction, unconditional branch, BTB, branch, interrupt, RAW, data forwarding, bypassing, circuit, device control, internal commands, graph, pipeline scheduling, command flow, instruction scheduling, RA, operand, operator sequence, variable, register, program, register renaming, table, logical, microinstruction, queue, file, ideal pipeline
Pipeline organization of the microprocessor
The number and essence of the stages differ in different processors.
Let’s consider the principles of pipeline processing of information on the example of a five-stage pipeline, in which the execution of a command consists of the following stages:
- ID ( INsTRuction DecodINg ) — command decoding;
- OR ( Operand ReadINg ) — read operands;
- EX ( ExecutINg ) — command execution;
- WB ( Write Back ) — write the result.
The execution of commands in such a pipeline is presented in Table. 9.1.
| Team | Tact | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | |
| i | IF | ID | OR | EX | WB | ||||
| i+1 | IF | ID | OR | EX | WB | ||||
| i+2 | IF | ID | OR | EX | WB | ||||
| i+3 | IF | ID | OR | EX | WB | ||||
| i+4 | IF | ID | OR | EX | WB | ||||
Ideal Pipeline Performance Estimate
Let the following time costs (in some conventional units) be required to complete the individual stages of processing:
TIF = 20, TID = 15, TOR = 20, TEX = 25, TWB = 20. time is t = 5 units, we get the takt time:
Let’s estimate the execution time of one instruction and some group of instructions in sequential and pipelined processing.
With sequential processing, the execution time for N instructions will be:
Examples of the duration of execution of a certain number of commands in sequential and pipelined processing are given in Table. 9.2.
Obviously, if the pipeline runs for a long time, its speed will significantly exceed the speed achieved with sequential processing of commands. This increase will be the greater, the shorter the duration of the pipeline cycle and the greater the number of instructions executed during the period under consideration. Reducing the duration of the cycle can be achieved by dividing the execution of the instruction into a large number of stages, each of which includes relatively simple operations and therefore will be completed in a shorter period of time.
So, if in the Pentium microprocessor the pipeline length was 5 stages (at a maximum clock frequency of 200 MHz), then in the Pentium 4 processors on the Northwood core the pipeline length was 20 stages, and on the Prescott core it was increased to 31 stages at a maximum clock frequency of 3.8 GHz.
| Number of commands | Time | |
|---|---|---|
| for serial execution | when pipelined | |
| 1 | 100 | 150 |
| 2 | 200 | 180 |
| 10 | 1000 | 420 |
| 100 | 10000 | 3120 |
task (process) is a sequence of interrelated actions leading to the achievement of some goal. In computing, a task is understood as a specific entity that is closely related to the processor architecture and has its own virtual address space and state. In a simpler sense, we will assume that task is a program that is running or waiting to be executed while another program is running.
The process can be in the following states:
- active state (count) — the program is being executed on the processor;
- waiting — the program is not executed because some resource is busy;
- readiness — the program is not executed, but all the resources required at the moment are provided for execution, except for the central processor;
- termination — normal or abnormal termination of the program, after which the processor and other resources are not provided to it.
In the case of computers, the definition of a task typically includes the resources needed to achieve the goal. The concept of «resource» is not strictly defined. We will assume that any consumed object (regardless of the form of its existence), which has some practical value for the consumer, is a resource :
modes:
- The processor time is divided between programs that are in memory in the ready state;
- in parallel with the operation of the processor, preparation and exchange with several external devices (ED) takes place.
Multiprogramming is designed to increase the throughput of a computing system by more uniform and complete loading of all its equipment, primarily the processor.
At the same time, the speed of the processor itself and the nominal performance of the computer do not depend on the use of multiprogramming.
Multiprogram mode has hardware and software support in the computer. The hardware used to organize the multiprogram mode includes controllers for RAM and external devices that can work in parallel with the processor, an interrupt system, hardware for protecting programs and data, etc.
The software component contains a multitasking operating system, external device drivers, handlers interrupts and other means. Operating system, implementing multiprogram mode , should distribute (including dynamically) between parallel executable programs system resources (processor time, RAM and external memory, I / O devices) in order to increase throughput and taking into account restrictions on resources and requirements by the urgency of the implementation of individual programs.
The efficiency of a multiprogram computer can be estimated by the number of tasks performed per unit of time (bandwidth) and the execution time of a separate program.
When analyzing the operation of a computer, it is important to determine the degree of use of its resources. For this, the following indicators are widely used (Fig.
6.1):
device load factor:
where — device busy time for the total time of the computer;
average length of the request queue to the device :
6.1.
Evaluation of performance indicators of a multiprogram computer
6.1 cases:
In addition to the average queue length, the dynamics of its length change is also important.
The values of , and changes in time of the Lq value can be used to determine the most scarce resource in the system, its «bottleneck».
Elimination of these «bottlenecks» can be carried out either by increasing the performance of the corresponding resource , or by choosing such a mixture of tasks that would ensure a more uniform use of all resources (for example, some tasks use the processor more actively (counting tasks), others — hard disk (working with databases), others — input-output devices (output of a large amount of graphic information)).
Consider an example of executing four programs in multiprogram mode with a multiprogramming factor equal to 2.
The multiprogramming factor (KM) is the number of programs processed simultaneously in multiprogramming mode .
We assume that the computer has 3 devices that can work in parallel: the central processing unit (CPU), the input device (IN) and the output device (OUT), and the programs go through the following work cycle:
account1 — input — account2 — output. The execution times of the corresponding blocks of programs are given in Table. 6.1.
We assume that programs have relative priority. That is, firstly, if several programs simultaneously claim any resource when it is freed, then it is provided to the program with the smallest number.