I just ran Quake 3 Demo001 640x480 normal quality setting and got 37.2 FPS. Looking at a similar benchmark from cpu-central where they used a K6-III+ at 450MHZ and a Voodoo3 3000 AGP they got the results below:

See second section labelled 400mhz, second column for normal quality.

Taken from http://www.cpu-central.com/Articles.asp?artic … ecor_int=28#xvi

This is about as close as I can find as far as similar parts and settings but I have no idea if around 10fps difference between my setup and theirs makes sense.

Is memory running 100 MHz or does the board have an asynchronous mode, running it at 133? That would cost performance if FSB and memory is not running at same clock. Furthermore, it's the memory set to CAS2 in BIOS, or is it just specified to be able to do it?

H3nrik V! wrote on 2022-08-30, 21:14:

Is memory running 100 MHz or does the board have an asynchronous mode, running it at 133? That would cost performance if FSB and memory is not running at same clock. Furthermore, it's the memory set to CAS2 in BIOS, or is it just specified to be able to do it?

CPU-Z reports the memory running at 100mhz in CAS 2. I set CAS 2 in the bios:

I get about 1900 with radeon 7500 66fsb and simms on tyan s1564d (400mhz)

Btw. Set that chip to 2.0-2.2v and 2x=6x see if you can get 600 mhz working.

Your voodoo is probably why your score is low. I think a 7500 may just be a faster card for this bench

Sphere478 wrote on 2022-08-30, 23:21:

I get about 1900 with radeon 7500 66fsb and simms on tyan s1564d (400mhz)

Btw. Set that chip to 2.0-2.2v and 2x=6x see if you can get 600 mhz working.

Your voodoo is probably why your score is low. I think a 7500 may just be a faster card for this bench

I will aim for 550mhz and if that goes well try 600 and post about it in the system specs forum. Is there much point to using the Voodoo3 beyond playing glide supported games? If not I'll play those and then swap the card out for a later card from 2000 - 2004 that I have laying around.

Don’t try to use anything from ati or nvidia that is newer than nvidia 5000 or ati 9000 series the drivers don’t like 586 after that.

Also, use windows drivers or really old driver versions.

This is really important it will save you a lot of time.

While windows xp works on 586 pretty well, games run much much slower. Use 95, 98, ME. I use ME myself.

I have had the most luck with the short green radeon 7500 cards on pci socket 7 systems.

On agp systems if the motherboard likes it, I have had good results with 9800 pro (agp is very picky on ss7)

Your performance looks alright to me. For comparison, here are my results with a Celeron 466 and a Voodoo3 2000 AGP.

BTW, don't expect great results in 3D Mark 2000 and 2001 on the Voodoo3. Those benchmarks expect T&L (2000) and pixel shader support (2001) which that card doesn't have. Stick with 3D Mark 99 and period correct games (Unreal Tournament, Quake 2&3 etc.) for your tests. Lastly, to get the most out of Quake 2 on that system, you need both the 3DNow! patch and MiniGL.

I'm working on a similar system and could help compare. I am running at 600MHz but I can clock down to 400MHz and go from there.

• Motherboard: Shuttle HOT-591p with 512k L3
• CPU: K6-3+ @ 600MHz (100 x 6)
• RAM: 512MB PC133 CL2
• GPU: Voodoo3 3000 AGP

I expect to see higher 3D scores because this system is using AGP, but maybe we can find similar CPU and memory scores.

A couple ideas:

• Could you try running some 3DMark 99 benches? Like Joseph_Joestar, I think 99 and 2000 are most appropriate for these systems.
• Instead of CPU-Z v1.2, try CPU-Z Vintage Edition v1.03 (https://www.cpuid.com/softwares/cpu-z.html). Could you show me your scores under the Bench tab?

Cosmic wrote on 2022-08-31, 04:12:

I'm working on a similar system and could help compare. I am running at 600MHz but I can clock down to 400MHz and go from there. […]
Show full quote

I'm working on a similar system and could help compare. I am running at 600MHz but I can clock down to 400MHz and go from there.

• Motherboard: Shuttle HOT-591p with 512k L3
• CPU: K6-3+ @ 600MHz (100 x 6)
• RAM: 512MB PC133 CL2
• GPU: Voodoo3 3000 AGP

I expect to see higher 3D scores because this system is using AGP, but maybe we can find similar CPU and memory scores.

A couple ideas:

• Could you try running some 3DMark 99 benches? Like Joseph_Joestar, I think 99 and 2000 are most appropriate for these systems.
• Instead of CPU-Z v1.2, try CPU-Z Vintage Edition v1.03 (https://www.cpuid.com/softwares/cpu-z.html). Could you show me your scores under the Bench tab?

Ok thanks I ran the 3D Bench 99 and CPU-Z Vintage benchmarks. I thought there wouldn't be much difference between AGP and PCI of the same/similar models of graphics card.

Not sure if the reference CPU in the CPU-Z screen shot is a plus model or not. No idea how much difference it would make if wasn't a plus model.

Ok after running 3DMark 99 I searched around and found a very similar setup to mine and they had quite a bit higher scores in both CPU and GPU:

Courtesy blank001 - 3dmark99 MegaThread

I guess the reason I started this thread is I have some hardware I've accumulated and I went with the shuttle because it worked right from the get go when I started. So if my numbers are low I'd like to make sure it isn't a bios/software issue because if its hardware maybe I can change things around.

If your board has an option in CMOS for memory bank interleaving be sure it's enabled, four way if your memory works with it. If there isn't an option in CMOS, you can use the George Breese patch.

Shogun wrote on 2022-08-31, 22:22:

Ok after running 3DMark 99 I searched around and found a very similar setup to mine and they had quite a bit higher scores in bo […]
Show full quote

Ok after running 3DMark 99 I searched around and found a very similar setup to mine and they had quite a bit higher scores in both CPU and GPU:

Courtesy blank001 - 3dmark99 MegaThread

I guess the reason I started this thread is I have some hardware I've accumulated and I went with the shuttle because it worked right from the get go when I started. So if my numbers are low I'd like to make sure it isn't a bios/software issue because if its hardware maybe I can change things around.

That score was achieved with an Ali Aladdin chipset motherboard - that chipset has superior memory bandwidth compared to the MVP3.

I have various K6-3 benchmarks with different motherboards that might help https://www.vogonswiki.com/index.php/Super_So … et_7_Benchmarks

I'm glad you made this thread, after comparing my setup I discovered my L1 and L2 cache speeds are super slow for some reason. :(

For background in case it's the issue, I am using a K6-2+ 570 with the 0-ohm mod to unlock the rest of the L2 cache and to make it a K6-3+, then clocked to 600MHz with the 2x multiplier jumper and 2.1 volts.

CPU-Z v1.03.0.w9x

I used SetMul to bring my CPU down to 400MHz, so we're both at 400MHz.

Your CPU-Z: 707 CPU and 2538 FPU
My CPU-Z: 589 CPU and 2444 FPU.

At 600MHz I get 804 CPU and 3706 FPU, so I have to be at 600MHz to surpass your 400MHz.

On the K6 benchmark wiki, Machine 5 - K6-3+/500 gets 887.1 CPU and 3197.0 FPU, so the main CPU score is still higher despite being 100MHz slower.

AIDA64 v2.20.1800 Cache and Memory

Our memory speeds are comparable:
Yours: 158MB/s, 82 MB/s, 96 MB/s, 227.0 ns.
Mine: 156 MB/s, 324 MB/s, 106 MB/s, 281.4 ns.

But in L1 and L2 I get weird results:

Your L1: 3145 MB/s, 2986 MB/s, 5391 MB/s, 5.1ns.
My L1: 3148 MB/s, 81 MB/s, 171 MB/s, 5.1ns .

Your L2: 379 MB/s, 179 MB/s, 231 MB/s, 45.1 ns.
My L2: 373 MB/s, 82 MB/s, 157 MB/s, 48.2ns.

How is my L1 and L2 slower than main memory? Could that explain why my other scores are lower?

3DMark 99

* Note, screenshot is from a 600MHz run, I didn't screenshot the results below when running at 400MHz.

Both at 400MHz, 800x600x16:

Your score (Voodoo3 2000 PCI): 2347 3D, 5369 CPU
My score (Voodoo3 3000 AGP): 1665 3D, 4203 CPU

That's a big difference! It's got to be the cache. Even when I go back up to 600MHz I still can't reach your score.

Memtest86 v.4.3.7

Memtest shows my main memory at 114 MB/s and both L1 and L2 showed 154MB/s. That's so slow... I never noticed it until now. According to a post in the K6-3+ mod thread I should be getting closer to 4378 MB/s L1 and 1761 MB/s L2.

Speedsys 4.78

Under DOS 7 (Windows 98 built-in DOS) Speedsys shows better speeds, but it doesn't explain why it's so bad everywhere else.

Main memory: 175 MB/s
L1 (32KB): 1749.07 MB/s
L2 (256KB): 1282.32 MB/s
L3 (512KB on-board): 281.71 MB/s

Next steps for me

Has anyone ever seen this weird discrepancy in L1/L2 speeds on a K6-3+ CPU before?

I don't want to hijack your thread so I can make another one if that would be better. :)

I think it’s a flase read. To benchmark with memtest you have to use the same mobo and same memtest version it seems

One version to the next will see different memory/cache speed results.

But, just in case, download central tweaking unit (win 9x only) and play with those settings they can help a fair amount. If you actually do have an issue (bios not supporting write combining?) this may fix it.

DIY Bios Modding guide Jan Steunebrink k6-2+/3+ 128gb

Repo Man11 wrote on 2022-09-01, 00:19:

If your board has an option in CMOS for memory bank interleaving be sure it's enabled, four way if your memory works with it. If there isn't an option in CMOS, you can use the George Breese patch.

Thanks for the tip! After enabling that setting I saw a very minor improvement:

3DMark 99 before: 2347 / 5369
3DMark 99 after enabling interleaving: 2439 / 5700

CPU-Z Vintage 1.030 before: 707.7 / 2538.7
CPU-Z Vintage 1.030 after: 708 / 2540

Quake 3 Arena timedemo 1 before: 37.2fps
Quake 3 Arena timedemo 1 after: 38.3fps
(640x480, normal)

Before:

After:

Also thanks @mwdmeyer & @cosmic those results helped a lot. I'll poke around a bit more as theres a lot to comb through here.

Shogun wrote on 2022-08-30, 19:08:

I'm having trouble verifying if my setup is performing correctly by finding benchmarks with similar hardware ... From what I've […]
Show full quote

I'm having trouble verifying if my setup is performing correctly by finding benchmarks with similar hardware ... From what I've seen in systems that are kind of close to my specs these numbers seem a bit off.

Build
Motherboard: Shuttle Hot-597
CPU: AMD K6-III+/400ATZ running at 400mhz (100fsb x 4)
RAM: 256mb Pc133 ram (CAS 2)
Video card: Voodoo3 2000 16mb PCI

Benchmarks
3DMark 2000: 1310 (1024x768)

From that 3DMark2000 score it would seem that you're in line. My system initially scored 1360 points at 1024x768x16 as well. Although, you're running a higher FSB, so in that sense it seems odd that the results would be slightly in my favor. My system utilizes an ASUS TX97-XE and a K6-III+ 400 MHz (6x66).

edit: Nevermind, you've got a V3 2000 - I've got a V3 3000. That probably explains the disparity.

I have a pretty much identical set up at home, but mine is running at 600mhz.
Edit: I do have a K6-2+ system running at 550mhz as well , but both are VIA MVP3 systems.

DEFINITELY stick with the 3dfx card if you can. The driver overhead is a massive relief on the K6 platform.

The only benchmark that I can remember my score from is Final Reality, I think I got like 5.6
I do have 512MB of ram in that system, although that shouldn't really matter.

If you haven't figured it out by this evening I can run some benches and post up the results for you.

So I enabled bank interleaving and that helped a little. I then took a look at CTU and tried to enable write combining using the method shown at My oldest rig, Hyundai HD-P5233M and performance didn't change. I tried both 32mb and 8mb in mtrr0. Also I'm unable to enable mtrr1. When I type something in it just changes back to all zeroes. Not sure why nothing changed with write combining on.

I think something is wrong I tried increasing the cpu to 500mhz via the multiplier in CTU which worked but now my 3DMark99 score dropped to 2057 / 4802?

Image of how the write combining settings were applied from the thread referenced above but using the memory address shown on my system: