CPU Performance from Socket 7 to Slot 1

Introduction

A comparison of all currently available PC CPUs was more than overdue and although it’s a horrible amount of work I felt forced to finally do one after Intel released not only a CPU for the high end but as well for the low end market. This performance comparison should give you some help deciding which kind of system you will buy or put together or in which direction your next upgrade is going to go.

The office application performance was measured with Ziff Davis’ Winstone 98. Winstone 98 is using more recent software than Winstone 97 and possibly the way ZDBOp is weighing the different applications has changed, because non Intel CPUs perform worse in Winstone 98 than they did in Winstone 97 if you compare their scores against Intel CPUs. The 6x86MX suffered the worst impact, whilst a 6x86MX used to reach pretty much the same Business Winstone score as a Pentium II at the same PR-rating it is quite a way behind Pentium II CPUs in Winstone 98 now. Winstone 98 runs about 40 min., performimg two siutes of ’task switching’, which is mainly responsible for the length of the benchmark run. I was very surprised finding out that the task switching benchmarks don’t influence the Winstone score at all though. Database software and word processors have the highest weighing for the final score.

For testing the performance of the CPU’s floating point unit I used 3D Studio Max, taking the time the system needed to render a scene. 3D Studio Max or 3D Studio Max 2 are still the number one rendering engines used and they obviously don’t require that much of main memory or Level 2 cache performance, which is why the Celeron scored almost as well as the Pentium II.

For gaming I planned using Id’s Quake II, Rage’s Incoming and Acclaim’s Forsaken, but I found out that Forsaken scales exactly the same as Incoming so that I used the special benchmark version of Incoming that Rage was nice enough to make for me. This special gem is a lot easier to run than Forsaken. The benchmarks will show that Incoming depends pretty heavily on L2 cache speed, whilst Quake II still mainly focusses on FPU performance. Hence Incoming benefits quite a bit more from higher bus speeds than Quake II does.

Test Setup

The systems I ran the benchmarks on consisted of the following :

Slot 1 CPUs :

  • AOpen AX6B BX motherboard
  • Corsair PC100 64 MB SDRAM
  • Intel Graphic Express AGP Graphics Card with i740 chip
  • 3Dfx Voodoo2 12 MB reference board, driver from April 8, 1998
  • Quantum Fireball SE 4.3 HDD using UDMA and Intel bus master driver 3.01

Socket 7 CPUs :

  • VIA mVP3 reference board, using VIAGART 1.7
  • Corsair PC100 64 MB SDRAM
  • Intel Graphic Express AGP Graphics Card with i740 chip
  • 3Dfx Voodoo2 12 MB reference board, driver from April 8, 1998
  • Quantum Fireball SE 4.3 HDD using UDMA and VIA bus master driver 2.1.9

Benchmark Runs :

  • Winstone 98 at 1024x768, 16 bit color, 85 Hz refresh
  • Quake II 3.14 at 640x480, 3Dfx GL driver
  • Incoming at 640x480, Voodoo2 driver, special benchmarking version ’Lux et Robur’
  • 3D Studio Max at 1024x768, true color, rendering to 640x480

Office Application Performance Under Windows 95