I've reinstalled Windows 98 again on the K6, system and swapped to a K6-III instead of a K6-2. The Radeon 7000 however, won't work properly. I will try it in my PIII tomorrow, just to see if it is certified dead, but so far it doesn't look good. Error messages from DXDiag saying I got no 3D Accelerator installed, 3DMark99MAX saying it requires a DX6 compatible 3D Accelerator and so on. It's possible that it could work in a motherboard with newer revision AGP though, so we'll see how it fares in the PIII.

Going to see if I can rerun benchmarks on G400MAX and Voodoo 2 12MB with the K6-III to see if it performs noticeably better than the K6-2.

I've tried the Radeon 7000 DDR in my PIII today, and it's working just fine in that system. This leads me to believe that there's either problems with a. Running Radeon 7000 on a Via MVP3 chipset motherboard, b. Running Radeon 7000 in an AGP2x slot or c. Running this particular Radeon 7000 under Windows 98 se.
I might swap out the HD in my K6-III system and install XP on it, just to see if that would make a difference. I suspect however that a or b are the more likely alternatives.
Under Windows XP in the PIII system the Radeon 7000 scored 1934 in 3DMark 2001 and 2944 in 3DMark 2000, but these results will be heavily influenced by the faster CPU and is of course worthless as a comparison.
I would still like to run a few benchmarks on my other cards with the PIII though, just to see how they scale with a faster CPU. Hopefully I'll have some time tonight to do a few tests.
I also have a P166mmx I could try out on the MVP3 motherboard to test with a slower CPU to be able to make some graphics on how different graphics cards scale with CPUs... And if I find the inspiration, I've got Celeron 700 and Pentium III 800 coppermines I could run on the i815 motherboard. However, I don't know if I have the energy for all that at the moment 😁

Right, I've started doing some benchmarks on my K6-III and PIII. Trying out my Radeon HD3650 (512MB DDR2) in the PIII (turns out it will run on AGP4x) and it is fast! It totally puts my Parhelia to shame in 3DMark2000 and 2001, but I guess that's to be expected. The Radeon 7000 has had a runthrough on the PIII, and I'll post all the results later. Unfortunately I don't have Windows 98 on the PIII, but I might have to try out XP on the K6-III in hopes that it would run the Radeon 7000 then.

EDIT:
I've been running benchmarks all evening. I've got results for Voodoo 2 and TNT2 Pro 16MB for my K6-III 400MHz. The last one appears to be very hampered by only having 16MB as its performance plummets when you turn up the detail settings. I'll swap to the K6-2 @ 450MHz tomorrow and see if I get time to finish off my charts. I've also been running tests with a few cards on my PIII and it is incredible to see how the G400MAX scales with CPU power in Q3A, at least at low quality. At 640x480 it appears the PIII is actually the limiting factor, as it scores on par with the Parhelia and the Radeon HD3650, all of which blows the Radeon 7000 totally out of the water.

I've been having some problems with my mouse pointer disappearing in UT on the TNT2 card when running at 32bit colour, so I've had to manually change settings through the unrealtournament.ini file and relaunch the game to run benchmarks. This slows me down a bit, but only while testing on the TNT2, as this problem doesn't appear on other cards.

Back at Christmas 2000, when I got my first "real" 3D accelerator (upgraded from a Virge to a Geforce2 MX), my system was a K6-3 450MHz with a VIA MVP3 chipset.
At the time, the Geforce2 MX was a good bargain, but everybody used to say that a K6 system couldn't take advantage of anything faster than a TNT2. Luckily, back then you used to be able to browse people's 3DMark 2000 results on the 3DMark web site. You had to ignore outliers, but it was still a useful tool. You could filter based on CPU type or video card. By searching people's results this way, it was apparent that K6-3s hit their plateau with a Geforce2 MX or a Geforce 256 - both of these were significantly faster than the TNT2, despite what people (even salesmen) kept claiming. Apparently this was also true with the K6-2, as the tests in this thread have shown.

I remember being frustrated with getting my card to run right. I was about to give up and trade it for PCI. The biggest hurdle was a strange and poorly documented issue where my Tyan board required USB enabled for AGP to work. I didn't use USB back then so that bit me hard.

I saw performance issues depending on the combination of versions of VIA's AGP driver and NVidia graphics driver.
I might not remember this exactly correctly, but it was something like this:

If I used an older version of NVidia's graphics driver (v5.32 is what I think I remember), performance was significantly faster, but it had occasional graphical glitches unless I reduced the AGP mode to 1X. In order to accomplish this, I had to install a particular (I think older) version of the VIA 4-in-1 drivers and use a so-called "safe mode" install option.
The performance loss going from 2X to 1X was negligible, something like half a percent or less in 3DMark 2000, and it fixed the graphical glitches. Once I found the right combination of drivers, this was how I left it set up.

If I used the then-current version of the NVidia driver, it was stable in AGP 2X mode, but it was significantly slower. My 3DMark score dropped from something like 3000 to about 2500, or something like that. I *think* this was associated with the newer NVidia driver, not the newer VIA driver, but I might be mixing that up. I think the only reason I used an older VIA driver is because the newer ones didn't have the "safe mode" option, and at least one of them simply wouldn't install properly.
Whatever the details, I do remember that I had to use a particular combination of "obsolete" drivers, then configure my AGP speed to 1X, to make it perform the best.
Apparently somebody "fixed" something in the more recent drivers (as of Dec 2000) that made VIA AGP 2X stable but crippled the performance somehow.

Real life (niece's birthday, national day coming up and such) is slowing down my benchmarking, but I'm going to finish off the group I'm working on asap. Hopefully I'll be able to make some nice little charts and such, but I should perhaps make a new thread, as I'm moving a bit out of what the original poster started with.

Right, I've got most of my benchmarking done.
Still need to do Voodoo 2 and TNT2 on the Pentium III to get my scaling tests done, but I'll get that done this weekend I hope.
I'm amazed at how well the G400MAX scales at low detail settings under Quake III.

This is very nice 😀.
I went through the thread again and realized I haven't plugged my own thread, so I shall do so rather shamelessly 😜.

P2/P3 VS K6-3+ - A Great Battle Commences

I believe you'll find a lot of the info there quite interesting!

shamino wrote:

Apparently somebody "fixed" something in the more recent drivers (as of Dec 2000) that made VIA AGP 2X stable but crippled the performance somehow.

In my charts, in order to have "nice" performance from any of my AGP card i had to disable AGP 2X in the bios.

LunarG, thank you a lot for your charts, i really enjoy seeing these! Scary how performance scales from one cpu to another. Scary but obviously predictable.
By the way, the Unreal Tournament scores seems a bit low, did you ran the Voodoo2 in glide or direct 3D? I ask this because i just got a voodoo2 12MB yesterday to replace my faulty Voodoo2 8MB and i was able to get a average of 31,5fps at 800x600 on high, but again my cpu was running at 500Mhz, and we all know how much cpu dependant the UE1 is.
3DMark2001 seems to be a bad performance indicator for the G400 on the Pentium 3, because despite its lower 3Dmark score it did quite well on the games!

Also i'm surprised that i got better framerate with the voodoo2 on my AMD k6 than with any other video card under UnrealTournament, even better than with a Voodoo3 or a Voodoo5. Maybe a single Voodoo2 is a better pair for a AMD k6 i guess.

I did run Glide on all the UT test on the Voodoo 2 yes. I dunno why they are that low, but as you say, it could be due to the lower clock speed. I've been trying to get my CPU's to run higher speeds, but it's not working well, so I'll leave it.
The G400 scores low in 3DMark2000 and later due to lack of features. If it had supported T&L and had the same level of performance, it would automatically score higher simply for completing more tests as far as I know.

The high quality setting scores of UT under the Tualatin & HD3650 setup are more than that of low quality. I also have seen this trend occassionally but never really understood it. I can only guess that it might be due to the GPU working more and not pestering the CPU frequently thereby reducing its efficiency.

LunarG wrote:

I did run Glide on all the UT test on the Voodoo 2 yes.

Better to mark the API's in the results then, glide could be considered unfair advantage.

Putas wrote:

LunarG wrote:

I did run Glide on all the UT test on the Voodoo 2 yes.

Better to mark the API's in the results then, glide could be considered unfair advantage.

Not really, if the point is to see how different cards perform on different CPU's. You'd always run Glide on a 3dfx card where you can.

I forgot how bad the AMD k6 is even under glide on certain unreal based games. I've ran Deus EX and Undying and the performance was pretty poor when compared to Unreal and UnrealTournament. I know they're both later games from 2000\2001 but even so, textures are still 256x256 and the engine remains pretty much the same. The games themselfes are much more complex requiring a faster cpu i guess like an early AMD Athlon.

The v2 will scale but only @ 640x480 imho...

I use v2 12mb sli in my k6-3+@600 and with 1024x768 it is hardly a bottleneck. Lower resolutions, my p3 is much faster but at 1024x768 the v2`s are getting a much harder time and the k6-3 is just a little bit slower.

Same scenario for the v3 2000, only the voodoo3 3000 and upwards are really getting hold back and need a p3 to shine at any resolution.

As for the chipsets: i didn`t have a lot of luck with mvp3 boards and really prefer the ALI, preferably on the p5a but i have 2 jetway boards with ALI that run very good also (these just lack atx form factor).
Agp cards run well also if you use the 1.72 agp driver except for de TNT/TNT2 wich can be a bit unstable.
I used a lot of different agp cards on the p5a and had very few problems. It now uses the geforce2 Ti with the v2`s and runs perfectly fine (and gives me usable AA in non glide games)