Oh wow.

After some further testing I'm not really sure if this program does anything on my CPU... especially because I just found a jumper on the floor. It seems that, while making a list of all caps on the mainboard yesterday, the AGP clock divider jumper fell off. Yes, I was running those benchmarks at 120 MHz AGP clock! 😳

I already thought a perfectly stable system with zero latency L2 cache was a bit weird, anyway... But then, a perfectly stable system at 120MHz AGP might be even more unusual. Needs more benchmarks.

My mind is so gone. I forgot which forum I read about this... I thought I found it randomly while searching the net and then posted about it as well. Haha. Darn that's sad.

I saw that, and when I posted a reply I got "topic does not exist" 🤣

Anyway this program doesn't seem to do much for me. Speedsys reports 846.56MB/s L2 throughput on latency 15, and 850.50MB/s on latency 0. Wouldn't surprise me if Intel locked or limited this setting at some point, so you might need an older CPU for this to work.

You're on a Coppermine, correct? I'm interested to see how it improves Katmai cores since they were about half the speed.

I'd like to see the AGP card that can run at 120mhz bus overclock. Most can't handle 90.

Logistics wrote:

You're on a Coppermine, correct? I'm interested to see how it improves Katmai cores since they were about half the speed.

Nope, this is on a Katmai 550. I have a Coppermine 1000 with slocket I'd like to test, but no S370 heatsinks. Will dig around in my CPU box to see if I can find any Deschutes cpus, however I fear I might have given my last one away to a friend.

sliderider wrote:

I'd like to see the AGP card that can run at 120mhz bus overclock. Most can't handle 90.

Tadaah.
For some reason it won't run at 133 with 2/3 divider (89MHz), however. It used to do that with no problems (except in Windows). I hope a recap will fix that.

Logistics wrote:

You're on a Coppermine, correct? I'm interested to see how it improves Katmai cores since they were about half the speed.

L2 cache speed you mean? For gaming it does not matter much. At least, not as much as usually suggested. See My small addition to the 133MHz challenge.

Regarding the voodoo 3 AGP/PCI. That picture is the usual STB model '2000', very common. I made a small benchmark before at different bus speeds: 50 to 133MHz FSB on a BX Mainboard. The AGP divider is left at 2/3 all the time, so one can calculate that for each FSB speed listed. Note that I clocked down all my voodoo cards, to either 95 or 100 MHz. Since I found them rather hot at stock speeds, and I don't run anything demanding anyways. I remember clocking down even lower, to like 75 MHz, caused visual glitches.

Looking forward to the recapping result. Though my intuition tells me something else is causing the visual gliches. Do not really have time to investigate this now, as I am already occupied with my Ultrasound clone and a CPU underclocking test.

Found some more CPUs, a P2-350 Deschutes and P3-600 Katmai.
With the P2-350, changing the cache latency didn't seem to do anything either. Speedsys reported about 508MB/s for every setting. Next I tried turning up the FSB speed, and was able to boot at 140MHz with AGP divider 2/3. Pretty unstable though and the P3-550 is faster at 120MHz with 1/1 divider anyway, so not worth further investigation.
On the P3-600 I didn't try changing the cache latency, only FSB speed. I was able to boot at 115MHz with AGP 1/1 and run the Doom timedemo, which did finish, but returned some error instead of the realtics number.

I also noticed that the Voodoo3 almost went up to 50°C with the P2-350 at 100MHz, but didn't go much above 35°C with P3-550 at 120MHz FSB and 1/1 divider! Doesn't make any sense to me.

@gerwin, how do you change the GPU clock in DOS, do you have some program for that?
Also, have you ever tried running the V3 at 1/1 divider? It seems to work pretty well... but then I haven't tried any Glide games yet.

The Voodoo3-2000 on that picture is slightly different from mine, though. This card has 12 electrolytic caps while mine only has 5. The others are solid tantalum which I had mistaken for diodes at first...

So your card looks like the one shown here: 3dfx STB Voodoo3 2000 AGP video card, basic info. Actually I never really noticed the difference in capacitor style before.
Coincidentally, page two of that topic shows my notes on lowering the default clock of a Voodoo 3. It is a video BIOS mod. I have more notes at home, for settings below 125MHz.

I don't think leaving the jumper off will allow the AGP bus to overclock that high. It must default to one of the available dividers. Knowing how easy it is to leave a jumper off, I doubt the manufacturer would setup a situation where one of their customers could potentially burn out their video card.

gerwin wrote:

So your card looks like the one shown here: 3dfx STB Voodoo3 2000 AGP video card, basic info. Actually I never really noticed the difference in capacitor style before.
Coincidentally, page two of that topic shows my notes on lowering the default clock of a Voodoo 3. It is a video BIOS mod. I have more notes at home, for settings below 125MHz.

Alright, thanks. I'll try that sometime. Maybe a lower GPU clock would allow even higher AGP clocks! 🤣
I don't trust my system enough to perform a BIOS flash at the moment so that'll have to wait until after the recap.

sliderider wrote:

I don't think leaving the jumper off will allow the AGP bus to overclock that high. It must default to one of the available dividers. Knowing how easy it is to leave a jumper off, I doubt the manufacturer would setup a situation where one of their customers could potentially burn out their video card.

I get the same result when I set the jumper explicitly to 1/1... And how else would you explain the ridiculously high framerate in Doom? Somehow this just works, even though I have no idea how. As I said, at 133 with 2/3 divider the BIOS will beep about a graphics card fault (long-short-short).

Has anyone ever even tried running above 100MHz with 1/1 divider? Or maybe everyone always assumed that, if 133 at 2/3 doesn't work, 100+ at 1/1 CERTAINLY won't.
Pretty sure it won't be able to boot into Windows on this setting but I don't really care about that.

jwt27 wrote:

Has anyone ever even tried running above 100MHz with 1/1 divider?

Unintentionally, yes: 50 to 133MHz FSB on a BX Mainboard. ....At first I thought AGP was at 89MHz. It did not go very well.

gerwin wrote:

jwt27 wrote:

Has anyone ever even tried running above 100MHz with 1/1 divider?

Unintentionally, yes: 50 to 133MHz FSB on a BX Mainboard. ....At first I thought AGP was at 89MHz. It did not go very well.

I haven't fully tested the stability of this setting, of course, but so far it seems to go quite well. Can you recommend any programs or games to test that?

Also I just found a 800MHz Coppermine. Didn't even know I had one of these in slot form. It's supposed to run on 1.65V which my board does not support, but I gave it a try anyway. So far it seems to boot at 150MHz FSB, with 2/3 AGP divider. This scores 336 realtics in the Doom timedemo, and 190.5fps in Quake! 😳

edit: oh it just crashed a few times. Added a CPU fan, this seems to help 🤣
I'll let GTA run in Glide mode (without frame limiter) for a while and see what happens.

edit 2:
GTA ran fine for 45 minutes or so. That was a bit boring so I tried to find the highest FSB where it would boot with 1/1 AGP divider. This turned out to be 133MHz. Interesting was that all scores in Maulwurf's VGA benchmark suffered except Doom, which went up from 336 to 316 realtics!

Oops.

I got this message during PCPBench. ECC was enabled, so I guess it's serious.