RetroPCCupboard wrote on 2026-01-03, 09:58:

It isn't the latest version. Just what was already on the machine. It is from April 2024. I could download the latest and rerun the tests.

That would be interesting if you are willing to do it. I had a look and cannot download the old version from his website to do the check myself

Installed the k6-2 400 today. Unfortunately it did not go smoothly. It bootloops after the initial memory count. I had a search and found out this could be related to “usb keyboard support”. This is disabled by default so I just needed to do a cmos reset. It still boot looped so I tried removing 1 stick of the 2 x 32Mb SDRAM that is in there and it then, as if by magic, booted fine.

I’m still investigating what’s going on. But in the meantime here are some updated results with just 32MB installed comparing to all of the tests with the Pentium with 64MB

66x3.5, 75x3.5, 83x3, 83x3.5 66x6
233, 262.5, 250, 291.6, 400

Quake 320x200
53.5, 59.9, 61.7, 66.5, 63.6

Doom, max settings
84.68, 94.19, 98.15, 103.59, 89.02fps
882, 793, 761, 721, 839 realticks

Chris’s 3d benchmark (320x200)
165.5, 185.8, 189.8, 206.3, 220.4 fps
275.8, 309.7, 316.4, 343.8, 367.3 (score)

PCplayer 320x200
59.5, 66.8, 68.0, 74.2, 95.0

Speedsys
L1: 425.89, 478.13, 460.29, 530.71, 1835 MB/s
L2: 185.05, 207.75, 230.81, 230.81, 202.83 MB/s
Memory: 130.04, 146.30, 162.62, 162.53, 112.92 MB/s
Processor benchmark score, 176.77, 198.42, 188.92, 220.40, 458.59

I’ll need to make a table or something as there’s a bit too much here now to digest.

I have rerun with the latest dosbench from Phil's site.

Quake 320x200 = 71.3
Doom Max Details = 126.59
Chris 3D Bench (320×200) = 221.5fps (score 369.2)
PC Player (320x200) = 78.6

Speedsys
L1: 553.94 MB/s
L2: 278.12 MB/s
Memory: 183.88 MB/s
Processor benchmark score: 226.7

Here is my system specs:

Oh, and I just tried the benchmark in CPU-Z Vintage Edition:

Thank you for this!

I have solved the problem with the RAM. I had to change to another bios. I tried the bios that was on the board originally but it detected the k6-2 as a k6 and the boot looping issue was present. I used bios patcher to add k6-2 support, but this didn’t fix the problem.

I’m now using a Wims bios which was mentioned in another thread about this board

https://web.archive.org/web/20051109005128/ht … ex.php?count=-1

This detects the CPU as a K6-3D. Not sure if this bios supported the k6-2 as supplied, or if it’s part of Wim’s update.

My board had an oem bios in it but Wims is a generic MSI one for the same board revision and seems to work Ok. There seem to be slightly fewer options in the new bios menu, but nothing major is missing. I contacted Jan Steunebrink about modding the original bios a while ago and he has been working on it, as I would like to keep it with the original bios if possible.

Initially the performance of the Wims bios seemed to be much worse than the patched original bios. But k6init seems to have solved this.

It seems doom just doesn’t run great on my board! I can’t imagine getting another 20fps with it at 100MHz FSB. Maybe another 10fps with the Pentium.

Ok I’ve made several more tests now with the k6-2 at 400 and 450MHz with 66 and 75MHz FSB.

I attempted 83Mhz FSB, but this BIOS seems to be less stable than the original one at that frequency. It didn’t post at all at 500MHz, so I tried a lower multiplier. At lower multipliers it will POST but not boot. It freezes at the end of the POST screen.

I think I’m happy with 450MHz, just wanted to see if it was possible to get 500MHz for testing purposes

I am seeing some weird behaviour with the k6 vs the Pentium. The Performance of the motherboard cache is faster when using the k6, but memory performance is much worse. I see this in speedsys and in sisoft sandra98

Performance and stability seem OK. Here are some quick numbers. I’ll finish the spreadsheet this week sometime

450MHz with k6init set to auto
71.2fps in quake 320x200
27fps in quake 640x480 (18.7 fps without k6init)
98.9fps in doom (103.31 without k6init)
2220 3dmarks in 3dmark99
5287 cpu marks in 3dmark99

These numbers look alright apart from maybe the 3dmark99 GPU score. I think the amount of RAM is causing stuttering in the 2nd game which is pulling the score down. Averaging 13fps in the 2nd game vs 32 in the first game

That's with Voodoo 2000 right?

Yeah, voodoo 3 2000 running at 143 MHz

I think, on the whole, K6-2 performance doesn't seem too great considering the clock speed. Not suggesting anything wrong with your system, but I think that probably it just wasn't as fast as Intel at the time. Especially for floating point math that's used for 3D. I believe I chose K6-2 back in the day, as it was far cheaper than Intel. So it had it's place back then.

Anyway, out of interest I just did some tests to compare my Pentium MMX to my Pentium II. Both at 300Mhz, but the MMX is at 100Mhz FSB and the Pentium II is at 66Mhz FSB.

Pentium MMX @ 300Mhz:
Quake 320x200 = 71.3
Doom Max Details = 126.59
Chris 3D Bench (320×200) = 221.5fps (score 369.2)
PC Player (320x200) = 78.6

Speedsys
L1: 553.94 MB/s
L2: 278.12 MB/s
Memory: 183.88 MB/s
Processor benchmark score: 226.7

Pentium II 300Mhz:
Quake 320x200 = 68.7
Doom Max Details = 76.29
Chris 3D Bench (320×200) = 187.2fps (score 312.1)
PC Player (320x200) = 88.3

Interestingly, most of these scores are worse than the MMX scores! For Windows 3D games, it's a different story. The PII, is vastly faster.

Anyhow, if I enable fastvid on the Pentium II, the scores change to these:

Quake 320x200 = 78.8
Doom Max Details = 91.64
Chris 3D Bench (320×200) = 320fps (score 534)
PC Player (320x200) = 106.1

Now, with the exception of Doom, the Pentium II wins all of the tests.

This is Speedsys result for the Pentium II:
L1: 1127.84 MB/s
L2: 430.46 MB/s
Memory: 201.23 MB/s
Processor benchmark score: 339.31

This is much higher than the MMX.

I forgot to mention that both systems have a Riva 128 in. But the MMX actually has a better version , the 128ZX (8mb vs 4mb, and better RAMDAC). I am not actually sure if they are clocked any differently, but hopefully that would make little difference for non-3D-accelerated DOS tests.

From about 11:40 in this video there are benchmarks comparing Pentium MMX 233mhz to K6-2 400Mhz to Pentium II 233Mhz.

https://youtu.be/2fM9OfB4Jvk?si=0g0ZSz1d-oDbTy2C&t=700

Should help you to see if your system is performing as expected.

Thanks for your replies. Interesting to see how much the Pentium 2 bumps up the numbers.

With socket 7 I think it’s a limit of the platform as much as anything else.

I also bought my first k6-2 due to being on a budget ( I was 16 at the time). I did not have much fun with that SS7 platform though. AGP issues (well documented now) stability issues and overall poor performance from the TNT2 (M64). I think I went through 3 motherboards with it. First one was DOA, 2nd one got killed by a PSU failure.

I’ve been comparing my results to videos I can find from Phil’s computer lab, necroware, bitsundbolts with these CPUs. I’m seeing pretty decent numbers in quake, comparable to some results with SS7 chipset boards. If i have time I’ll compile a list.

Edit. Forgot to say I did some tinkering with the system at the weekend. The processor seems to be running just fine at 450MHz (75x6) and stock volts. I had some issues with a couple of older games (day of the tentacle, monkey island) crashing. I solved this by disabling L1 and L2 caches. Both games work flawlessly now.

I had already planned to do so, but I wanted to change out the original 4.3Gb Fujitsu HDD for a faster CF card. Since I’m also running a patched bios I should be able to run with DMA enabled for better performance. I made some tests before and after

Fujitsu 4.3Gb; 8.5MB/s (with and without DMA enabled)
32Gb Sandisk CD Card; 12MB/s (DMA disabled) 30MB/s (DMA enabled)

It’s much snappier now on boot. Runs great. No real change in 3dmark score though. I think more cacheable ram is needed…

I don't actually recall what motherboards and CPUs I used in this era. I know in 1995 I had a Pentium 120Mhz (Not MMX). In 1999 I had a Slot A athlon at 700Mhz (still have it).

I know between those dates, I stuck with AMD. But I don't recall the exact CPUs. Probably it was a K6 in 1997 on a new motherboard, and then I think upgraded to a K6-2 on same motherboard in 1998.

I remember all of them, is that bad? 😁

New CPU day! I just couldn’t help myself (I may have a problem!!) and bought a k6-3 for the socket 7. It’s a k6-3 400 (not a + model) and it’s not a particularly good binning as it’s rated to run at 2.4V but it was available at a relatively good price (half what most 2+/3+ CPUs seem to be on sale for)

It arrived with a lot of slightly bent, but no missing pins. Took some time to get them straight so the socket would accept the new CPU. All done and up and running now!

Some numbers
66x6 k6init on auto
3dmark99 2595 GPU / 5457 CPU
Quake 320x200 70.4fps 640x480 25.6fps
Doom 102.74 fps / 727 ticks

75x6 k6init on auto
3dmark99 2951 GPU / 6225 CPU
Quake 320x200 79.0 fps, 640x480 28.7fps
Doom 113.86 / 656 ticks

I’ve started compiling scores from various sources on YouTube. Since Phil’s computer lab did a video on k6init fairly recently it’s quite straightforward to do a side by side

https://youtu.be/Po_oy7VjLu0?si=pFSgdfvRMIhZi3I8
His board specs
GA-5AX Aladdin V SS7, 128MB SDRAM, k6-3+ TNT2

4x100
Quake 320x200 79.4 fps
Doom 144.7 fps
Quake 640x480 24.2fps

5.5x100
Quake 640x480 28.2fps

So looks like I’m getting reasonable numbers apart from doom which always seem a bit on the low side (but still fine)

More benchmarks to see what happens when I try other memory/cache configurations

Now I’ve got the k6-3 I’m not limited to 64MB ram

So I dropped in a 128MB stick in (double sided / dual rank)

Didn’t bother testing this config in DOS as I didn’t expect increasing the RAM would do anything for DOS performance

Just did a quick check with 3dmark99. Performance increased to 3035 points now GPU. CPU score went down slightly to 6151

Next, I tried disabling the motherboard cache and I was quite surprised by the result. Ran 3dmark again

GPU: 3215
CPU: 6846

Will keep playing around!

Disabling the motherboard cache for the dos benchmarks is a bit of a mixed bag. The quake results were improved by a few %, while there was a 0-3% decrease to almost all of the other benchmarks. The only big increase was with the speedsys memory performance. All memory performance benchmarks have been pretty rubbish with the k6 CPUs compared to the Pentium MMX for some reason. But disabling the cache seems to fix this. The result is now better than the Pentium was at the same FSB, rather than significantly worse

Quake 320x200 increased to 81.2 from 79.0 fps (+2.8%)
Quake 640x480 increased to 30.2 from 28.7 fps (+5.2%)
Speedsys memory result increased from 108.99 to 152.43 MB/s

I’ve got a few other configurations for the RAM to try to see if any make a difference.

2 x 64MB single sided
Id like to try 2 x 128Mb double sided but I’ve only got one double-sided stick. So I may try the 2x64 + 1x128 so I’ve got 4 ranks total
1 x 256MB double sided

Hopefully I’ll also be able to test at 83MHz FSB again at some point.

FullYes wrote on 2026-01-09, 19:49:

New CPU day! I just couldn’t help myself (I may have a problem!!) and bought a k6-3 for the socket 7.

Yes. You have a problem. It will probably get worse. You have the retro bug, and there is no cure.

Congrats on your purchase! Hope you have lots of free space in your home for the next retro PC you buy. 😀

No I’ve definitely reached my limit now. I promise!!