PC Player Bench

Doom

Quake

Duke Nukem 3D

Shadow Warrior

Terminal Velocity

A couple additional charts.

The first compares performance with the Pentium Pro (512kb) without FastVid and with FastVid enabled.

The second chart shows the different in performance between the 256kb and 512kb versions of the Pentium Pro. The extra cache resulted in performance gains from between 2% to 12% depending on the benchmark.

I'd been reading about FastVid and that EMM386 apparently can interfere with its function.

As a result, I re-ran all the FastVid benchmarks with EMM386 disabled. The only significant change was PC Player Bench at 640x480. FastVid and no EMM386 resulted in a 50% performance boost.

In contrast, Duke Nukem 3D and Shadow Warrior showed only marginal SVGA gains of about 1-2 FPS at best. Other benchmarks showed little to no gains with EMM386 disabled.

I reformatted the charts to combine the FastVid and no FastVid results into single charts, and recolored the Pentium Pro columns. This should make them easier to read.

Digging through some old CGW magazines, I attached excerpts from their benchmarks (November 1996) and their write-up on the Pentium Pro from their Ultimate Gaming Rig build (January 1997).

They did realize a better performance gain in their benchmarks with FastVid and Duke 3D. However, the Pentium 166 still took the lead.

Awesome!
Miss the days when the PPro was king.

Just for fun, I was curious how the Pentium Pro 200 (256kb) would fare against a Pentium 133. Both processors released in 1995, so I thought I'd do a 1995 showdown.

Unfortunately I don't have a 450KX (Mars) board to make this a true 1995 comparison. I used the same Supermicro P6SNE (Intel 440FX) motherboard I used in previous Pentium Pro benchmarks. The Pentium 133 used a Gigabyte GA-586ATE/P with an Intel 430FX (Triton) chipset and 256k of L2 cache.

Video card used in both sets of benchmarks is a Diamond Stealth64 2001 (S3 Trio 64+). S3VBE20 was enabled where noted. For the Pentium Pro benchmarks, I enabled FastVid with EMM386 disabled.

While the Pentium Pro had generally better performance in most cases, it still was bested by the Pentium 133 in those SVGA Build-engine games. It blows my mind that for Duke 3D in 640x480, the Pentium 133 would have been the better choice than a PPro running at 200 MHz. 😆

Interesting. Why would the P200MMX be faster than the P200? I thought those two were basically identical except for the added multimedia instructions which are only helping if the sofware is built to take advantage of it..

mtest001 wrote on 2023-12-02, 20:28:

Interesting. Why would the P200MMX be faster than the P200? I thought those two were basically identical except for the added multimedia instructions which are only helping if the sofware is built to take advantage of it..

The MMX model had double the L1 cache

mtest001 wrote on 2023-12-02, 20:28:

Interesting. Why would the P200MMX be faster than the P200? I thought those two were basically identical except for the added multimedia instructions which are only helping if the sofware is built to take advantage of it..

The Pentium MMX has 16kb of L1 cache versus the Pentium 200 with only 8kb.

I also ran the Pentium MMX on a slightly newer Intel 430TX board (DFI P5BTX/L). I had wanted to run them on the same board initially, but apparently the board I used with the Pentium 200 (GA-586ATV) has problems with Pentium MMX 200 and assigning the correct multiplier.

I may try to dig out a more compatible motherboard and run both these processors on the same board to do a more apples-to-apples comparison.

Ok, I did not know about the bigger cache, it makes sense now. Thanks.

Shponglefan wrote on 2023-12-02, 20:36:

The Pentium MMX has 512kb of L1 cache versus the Pentium 200 with only 256kb.

This is L2. maxtherabbit means L1. Pentium is 8KB+8KB, MMX is 16KB+16KB. Example of the speed difference in NFS2 using same mobo/graphics https://www.youtube.com/watch?v=7iNgFkmsCcc
When it comes to L2 type of cache and chipset has much more influence than the size. TX and Pipelined can be a good 10-15% boost over VX with SRAM.

Build uses very low level optimizations, most likely means 16bit instructions and partial register accesses. As wiki states:

"Pentium Pro's out-of-order register renaming architecture had trouble running 16-bit code and mixed code (8-bit with 16-bit (8/16), or 16-bit with 32-bit (16/32), as using partial registers cause frequent pipeline flushing."

rasz_pl wrote on 2023-12-02, 23:23:

This is L2. maxtherabbit means L1. Pentium is 8KB+8KB, MMX is 16KB+16KB.

Whoops, had a brain fart when I typed that.

I meant 16kb and 8kb. 😕