Awesome work!

Really shows how well AMD competed with Intel back in the day.

Very nice test 😀

Google translate isnt really that good when it comes to translating but the charts are clear enough 😁

Huge respect!

So much work went into collecting these results, you must be a very dedicated person. Well done 😀

Thank you.

Actually the benchmarking itself is not that hard - one CPU takes 4 - 8 hours (depends how fast it is). Time measuring is semi-automatic so I just have to start the video encoding for example, then wait till it finishes and time is measured automaticaly... so I don't have to supervise the process with a stopwatch. 😀
Writing down the article is the hardest part I think. 😁

Good job, but 2 questions. First, why did you choose Windows XP especially with SP3 over Windows 98 SE? Second, did you actually use the original Voodoo3 driver from 3dfx?

SSFT A41 custom - Voodoo3 driver

Because I have no idea what kind of driver is that.

1. There are many reasons to use XP. The biggest one is application compatibility - most of these applications I used will not run in windows 98SE. Also the time measuring utility needs java to work, so win98 is again out of the game (testing so many CPUs and tasks manually would be too much time consuming and therefore impossible). I guess I could've used windows 2000 but there isn't really that much difference between 2k and XP. And last but not least - I'm going to test 50 or so more processors up to Athlon 64 FX and Xeon Gallatin with very similar methodology.
//edit: Oh, one more thing - dual CPU configurations would be useless in win 98... and I like dual CPU boards very much. 😎

2. Voodoo3 driver is 3rd party driver - in this case SSFT alpha 41... which is rather old but compatibility with older games (like Q3A, UT, etc) is better than latest SSFT version. http://www.3dfxzone.it/news/reader.php?objid=16199
The custom part is described here - quote from ps47, member of czech 3dfx forum and author of the modification (loosely translated from slovak)

ps47 at forum.3dfx.cz wrote:

A41 custom is a modified driver based on SSFT A41, the most problem-free SSFT driver version for general purpose usage. The custom version was created to run most games out of the box (from the oldest glide games to relatively new D3D ones).
Differences: 1. original glide 2x/3x instead of koolsmoky glides for the best compatibility. 2. 3dfx ICD (openGL driver) instead of mesaFX - ICD is faster for older games. 3. minor registry tweaks

A41 custom works much like the original windows 9x "3dfx Voodoo3 V1.07.00" and is the best winXP Voodoo3/4/5 driver in my opinion.

In CPU benchmark better to use games working in software modes. 3D cards modes good as additional. Quake 1.06 in software mode for example. Unreal has software mode.

Hello, these CPU charts are great, but I've just noticed one strange thing, that I don't understand.
Lets look at Havli's benchamark of GLQuake. And compare them to Philscomputerlabs voodoo2 scaling project.
In case, I provide links to both tests CPU scaling tests:
https://www.cnews.cz/test-historickych-proces … il-i-1995-1999/
https://www.philscomputerlab.com/uploads/3/7/ … ing_project.pdf

There is huge differance between FPS in GLQuake and I don't know why.
GLQuake 800x600

Havli's test on voodoo3:
P233MMX: 27,6 fps
P166: 14,8 fps

Phils test on voodoo2 (but as CPU is bottleneck, voodoo2 SLI ~ voodoo3 2000 values are same)
P233MMX: 69,2 fps
P166: 57.9 fps

Both are on Supersocket 7 platform. Havli used Ali Aladdin V chipset, Phil used Via Apollo MVP3, but differences shouldn't be too much.

Now, when you compare for example Pentium II CPU's

Havli's Pentium II 450: 80.7 fps
Phils Pentium II 400: 71.3 fps (but voodoo2 have there limit around 70, even with faster CPU, FPS is doesn't rise) so let's use voodoo2 SLI value,
that should match voodoo3
Phils Pentium II 400 (voodoo2 SLI~voodoo3 2000) : 126,8 fps

What's going on here? Why have Havli so little score with Voodoo3? Can it be because of Windows XP drivers? I think GLQuake have only Glide mode, so it cannot be because OpenGL Mini driver, instead of Glide.

What's holding back Havli's performance by huge margin (in some cases, like Pentium 166, even 4x less FPS).