Probably helps 3dfx had 3dnow! optimizations in their drivers. It's pretty evident here seeing it how it gets past a bit of the CPU bottleneck 😀

A Radeon 7xxx card could be part of this chart...

leileilol wrote:

A Radeon 7xxx card could be part of this chart...

Unfourtanely i never had any ATI 7xxx card. If i ever get my hands on one i might as well update the chart.

i have a few ideas:
1. did you ensure to disable vsync on all nvidia cards? the geforces don't work well on socket7 cpus as we knew but i didn't expect them to be so close to the tnt2m64.
2. what drivers did you use for the nvidia cards? in general(for nv drivers), newer drivers tends to push the card harder to its limit, but they also consumes more cpu power, so they would be slower with slow cpus.
3. i suggest you to give my quake3 benchmark demo a try, it pushes the cpu really hard and i want to see if the voodoo3 and r9200se would still perform well. its in this thread: introducing my quake3 benchmark demo

Just for fun I tried that Quake III demo on my Pentium III 800 with 512MB RAM and Matrox Parhelia. It gave me 4.4fps which seems surprisingly low, even though it was at 1280x1024 with everything at max.

I'm putting together a K6-2 soon, and I've got a small selection of different AGP graphics cards, including I think a Radeon 7000. Unfortunately the CPU is only 350MHz, so I doubt I could clock it as far as 500 to get similar results, but we'll see what I can manage.

Somethings definitely wrong there LunarG!

Also, like leileilol said, K6-2 and K6-3 CPUs love 3Dfx, due to 3DNow! optimizations done in the driver. K6-400 + Voodoo 2 was a good combo back then 😀.
Not to mention, the Via MVP3 and Ali Aladdin V have a lot of issues with AGP cards, 3Dfx only manage to get away because they only use the extra bandwidth provided by the bus.

What happened to the 9700 Pro? Doesn't UT run on it?

sliderider wrote:

What happened to the 9700 Pro? Doesn't UT run on it?

I'm not entirely sure sliderider, whenever i ran either UT or any 3dmark it would exit immediately to Windows. I've tried a few different drivers and even removed the cpu overclock. Also it felt a bit unstable on that board, random freezing while windows was booting, weird stuff. That ATI was reflowed once due to artifacts but it never get me problems ever since. I guess there could be some incompatibility between this card and the boards chipset?

noshutdown wrote:

i have a few ideas: 1. did you ensure to disable vsync on all nvidia cards? the geforces don't work well on socket7 cpus as we k […]
Show full quote

i have a few ideas:
1. did you ensure to disable vsync on all nvidia cards? the geforces don't work well on socket7 cpus as we knew but i didn't expect them to be so close to the tnt2m64.
2. what drivers did you use for the nvidia cards? in general(for nv drivers), newer drivers tends to push the card harder to its limit, but they also consumes more cpu power, so they would be slower with slow cpus.
3. i suggest you to give my quake3 benchmark demo a try, it pushes the cpu really hard and i want to see if the voodoo3 and r9200se would still perform well. its in this thread: introducing my quake3 benchmark demo

You mean disabling vsync at driver level or in the game setup? Either way i did none and left everything at default. I understand the framerate could go a tiny bit higher with vsync off but with so much bottlenecking from the cpu that the difference would be negligible. Also all cards ran at default so it's still a good reference for all i guess..
As for the drivers i have used 44.xx (can't remember exactly from the top of my head).

I forgot to mention that for quake 3 i used "timedemo 1" and "demo four". I believe this demo came originally with quake. For Unreal Tournament i ran a full cityintro flyby.

Nice work! That must have taken you a very long time. Windows benchmarking takes forever I find. DOS is so quick in comparison 😀

rick6 wrote:

You mean disabling vsync at driver level or in the game setup? Either way i did none and left everything at default. I understand the framerate could go a tiny bit higher with vsync off but with so much bottlenecking from the cpu that the difference would be negligible. Also all cards ran at default so it's still a good reference for all i guess..
As for the drivers i have used 44.xx (can't remember exactly from the top of my head).

I forgot to mention that for quake 3 i used "timedemo 1" and "demo four". I believe this demo came originally with quake. For Unreal Tournament i ran a full cityintro flyby.

disable vsync at driver level. 3dmark does it by default too, but quake and unreal may be affected.
for your platform, i suggest using 28.32 or 30.82 for geforce cards. as i said, the 40/50 series drivers and later can push the card a bit harder to the limit, but consumes more cpu power, so they are no good for slow cpus.
the four demo did come with quake3, but is probably too lightweight for stressing the rig.

Mau1wurf1977 yes, this sort of benchmarking can be quite tedious (but also quite gratifying in the end) so to make things a bit better i've putted the hardware on my desk. By doing that i could multitask better between this hobby of mine and more serious work at my main computer.

noshutdown, i'm going to do a new chart with only one nvidia card comparing results between these drivers you mentioned and also the one i used, vsync on and off, also i'll be using the "demo four" and the one you provided in your link so stay tuned!

Oh and the video card i'm going to pick is the Geforce 256 SDR. Even though it isn't the most powerful it seems to be the most fair and contemporary for this system.
I think it will go well with my voodoo 2 that still needs repair (missing a smd cap)

Stay tuned for the new chart.

Yay! I got my K6-2 3DNow! 350MHz up and running. Currently it's equipped with SB Live! sound card and Matrox Millennium G400MAX. I've put a PC133 128MB stick of RAM in it. Going to try to do some benchmarking later.
Got a Voodoo 2 8MB and 12MB I could try out. Also got a Radeon 7000, 3650 (with GDDR5 I think) and a TNT2 Pro 16MB. We'll see what I have time for. Hopefully the results will be close enough to what rick6 has got to be comparable. At least in a general performance level of each card.

Is that Radeon 3650 keyed for AGP 3.3v? That would be a juggernaut card for such a system! Still the k6 might hold it to Geforce DDR levels. I'm curious what results you'll have since i have none of those cards (apart from my at_the_moment faulty voodoo 2).

Voodoo cards indeed work really well with the K6. Especially the mobile versions with on-die cache. Make sure you run them at 100 MHz FSB.

Mine is running at stock speeds at the moment. 3.5x 100MHz, will try overclocking it at some point though.
Just did a quick 3DMark2000 and got 1135, so I guess the clock speed really holds the G400MAX back. It scored around 3.5k in my P3.
My motherboard has a whopping 1MB cache on it, but only L1 cache on the CPU.

rick6: You're totally right about the Radeon 3650... It''s AGP8x only, so I can forget about trying that. Would be totally overkill for that system anyway. I'm going to run the G400MAX on a day to day basis, just because I like it, and because it was like a wet dream back in the late 90's, but I'll try out some other cards. I also intend to run a Voodoo 2 on a daily basis in that system.

I did this Super Socket 7 Voodoo 2 CPU scaling chart a while ago. You can see the big jump when going form Intel to the K6 chips. And then another huge jump when going from a standard K6-2 to the mobile ones with on-die cache.

That's a really nice scalling chart you got there. I've been wondering about the performance of a voodoo 2 on different systems and that helps a lot!

I wonder if i could push my AMD k6-2 450mhz up to 550mhz. In order to run everything smooth and stable at 500mhz i had to push voltage up to 2.4 (2.3 did it as well but some video cards made it crash at 2.3v, i guess these video cards are more hungry for power than others). Any ideia what would be the safer max voltage for this cpu?

Don't go over 2.4V. 500 vs 550 on a K6-2 CPU is negligible.

To be honest a Slot 1 system is much better suited for Voodoo 2. I'm waiting for another CPU but I'm close to continuing this chart upwards. The Voodoo 2 scales like crazy, especially once you add SLI to the mix.

I have a 400 MHz K6-III and it does 550 with a bit of extra volts. But the difference between 400 and 550 isn't worth the trouble IMO. The main thing is having cache and 100 MHz FSB.

There is a German eBay seller selling the 1.6v K6-III 400 and that's the chip to go for.

If you need more speed than this then you should really go straight to a Slot 1 system with a Pentium III. The Voodoo 2 just continues to scale, even at 900 MHz. But I will know more once I have that 900 MHz CPU and then I will continue with Socket 370 and, if I have to, Socket 478 😀

Anyone know how stable the MVP3 chipsets run with 112MHz bus? Tempted to try 392MHz (3.5 x 112) on mine.