Originally Published by Advanced Micro Devices, Cyrix Corporation, IBM Microelectronics Division, and SGS-Thomson. These specification have long been obsolete and have been uploaded here to be preserved for research and educational purposes.

Processor Performance Rating (P-rating) Specification

Introduction

The Processor Performance Rating (P-rating) specification defines a methodology created by several companies including: AMD, Cyrix, IBM Microelectronics, and SGS-Thomson. The specification is designed to provide a credible, consistent, and easy-to-understand method for rating processor performance. Earlier methods of rating processors relied on a combination of architecture (e.g., 5-class, 6-class) and clock speed - a system that was confusing and failed to provide a good measure of real application performance. This system provides a single rating number for a processor (the P-rating) that gives a true measure of application performance and allows an "apples-to-apples" comparison among different processors on the market.

Goals of the Rating Specification

The goals of the processor performance rating specification are:

• To create a credible, industry-wide, and easy to understand method of rating processor performance
• To provide a single, application-based performance number that an end user can easily understand, instead of using frequency (MHz) and processor architecture
• To ensure wide acceptance by the processor manufacturers, system vendors, the press, and end users

Future Revisions

This methodology is judged to be a credible and accurate method as determined by all parties as of January 25, 1996. As newer benchmark tools become available and as the industry evolves to adopt newer OS and application programs this specification will need to be modified accordingly.

Rating Methodology Overview

The process for obtaining a performance rating for a processor is as follows:

1. Assemble two systems: one for the processor you want to rate (the vendor platform) and the other to accommodate a Pentium processor (the reference platform). If the vendor and Pentium processors share the same pinout and bus interface, a single system can be used by replacing the CPU between tests. Section 3 lists the hardware components for the systems and provides examples for each component. The reference and vendor platforms must be identical, and all components used must be fully documented and disclosed. Identical platforms ensure that comparisons are done on an "apples-to-apples" basis, and that the performance-rating tests are repeatable.
2. Load the software specified in section 4 and run the benchmarks specified in section 5 on the vendor platform.
3. Repeat step 2 for the reference platform with Pentium processors at different clock frequencies to build an index of performance numbers.
4. Use the index to obtain the P-rating of the vendor processor based on the algorithm in section 5.
5. Report the P-rating along with the actual configuration used for the test. The configuration list must include information for all the categories listed in the configuration charts in sections 3 and 4.

Hardware Configuration

Hardware Components

The following table lists the hardware components that must be disclosed in the final results as well as examples for each component. All configuration parameters in the vendor and reference platforms must be disclosed in the test results.

==================================================================
Component                   Example
------------------------------------------------------------------
Motherboard                 · Asus P/I - P55SP4 rev 1.4
------------------------------------------------------------------
Chipset                     · SiS 5511
------------------------------------------------------------------
Disk drive and              · Quantum Fireball 1280A drive
drive controller            · ATA-2 PCI controller
------------------------------------------------------------------
Video card, video           · Diamond Stealth 64 3200XL PCI
resolution, and software      graphics card with 2 MB of VRAM
driver                      · First production release of Diamond       
                              Stealth driver
                            · 1024 x 768 resolution, 256 colors 
                              at 75 Hz
------------------------------------------------------------------
Main memory type and speed  · 16 Mbytes of EDO DRAM (60ns)
------------------------------------------------------------------
L2 cache type and speed     · 256-Kbyte synchronous pipeline burst
                              SRAM 3,1,1,1 (8 ns)
------------------------------------------------------------------
PCI bus speed               · 33 MHz (Speed of PCI bus must match
                              clock rates defined in PCI 2.1
                              specification)
==================================================================

The hardware components listed in the table above are examples only. Individual vendors may choose to substitute other components as required. This allows the rating system to accommodate architectural, design, and positioning requirements for the processor being rated, ensuring a credible "apples-to-apples" comparison.

Note that both the vendor system and the reference system must have identical components. If identical systems cannot be provided due to architectural differences (i.e., processors with different bus interface protocols), like components can be substituted. Any substitutions should be fully documented.

Software Configuration

Operating System and Software Configuration Parameters

When running the benchmark tests on both the vendor and reference platforms, use the operating system and software configuration parameters listed in the table below:

==================================================================
Component                   Configuration
------------------------------------------------------------------
Operating System            Windows 95 first production release
------------------------------------------------------------------
File System                 32-bit with 4 Mbytes of cache
------------------------------------------------------------------
Virtual Memory              32-bit
------------------------------------------------------------------
Disk Compression            Not installed
------------------------------------------------------------------
Virtual Memory              Maximum 32 Mbytes, minimum 32 Mbytes
------------------------------------------------------------------
Read-Ahead Optimization     Full
==================================================================

This particular configuration was chosen in order to minimize deviations of the benchmark testing. No performance advantage or disadvantage has been observed by using this configuration relative to the Windows 95 default settings.

Benchmarks

Testing is based on Ziff-Davis's Winstone 96 benchmark. The process used to obtain a benchmark number is as follows:

1. Begin with a clean disk free of previous operating systems or configuration files.
2. Install the operating system (Windows 95) and set up the parameters listed above in section 4.
3. Install the Winstone 96 benchmarking software and defragment the hard drive using the Windows 95 defrag utility.
4. Run the Winstone 96 benchmark three times and average the results. Note that a defrag must be performed between each run to minimize variations.

Obtaining the P-rating

1. Use the reference platform equipped with various Pentium microprocessors (e.g., Pentium-75, Pentium-90, Pentium-100, etc.) to obtain an index table using the methodology described in section 5. Example:

   ======================================================================
   Pentium Frequency    75 MHz    90 MHz    100 MHz    120 MHz    133 MHz
   ----------------------------------------------------------------------
   Winstone 96 score     46.1      52.9      55.8       59.3       63.4
   ======================================================================

   Adjust the values in the index table down by 1.5% to compensate for the natural variation in the Winstone 96 test (no greater than 3% total variation as specified by Ziff-Davis Benchmark Operation). This results in the following table:

   ======================================================================
   Pentium Frequency    75 MHz    90 MHz    100 MHz    120 MHz    133 MHz
   ----------------------------------------------------------------------
   Adjusted score        45.4      52.1      55.0       58.4       62.4
   ======================================================================

2. Using the vendor platform, run the benchmark with the processor you want to rate. Note that the vendor platform and the reference platform must have the same hardware configuration. Example: The vendor then runs the benchmark on their processor and obtains a score of 57.1.
3. The P-rating number as currently defined is chosen to conform to a standard Pentium microprocessor frequency. To obtain the P-rating, compare the vendor processor benchmark score to the table of adjusted values obtained in step 1. The P-rating for the vendor processor is assigned according to a score that is greater than or equal to the scores listed in the adjusted table. Example: Since 57.1 is greater than 55.0 but less than 58.4, the processor has a P-rating of P100.

Reporting Results

Once the P-rating is defined, it can be used to describe the processor's performance in collateral material, on the processor chip, and when providing rating information to the press.

When a vendor reports P-rating results in published literature like collateral, press releases, and web sites, the literature must disclose all configuration parameters used to generate the P-rating. Configuration parameters are the rows in the tables in section 3. Additionally, any deviations from this specification must be clearly documented.

Reporting P-rating results in this manner will allow OEMs, the press, and consumers to reproduce and validate the P-rating for the processor. This ensures credibility for both the vendor's P-rating numbers and for the overall P-rating system.

Issued jointly by Advanced Micro Devices, Cyrix Corporation, IBM Microelectronics Division, SGS-Thomson