Did you do any memory performance testing with 512MB and 768MB?

I think nothing is wrong here. KT133A and SDRAM is just not fast enough to keep up with AXP 2100+ performance.

Here is couple of mine 3DM01 scores:

Athlon 1000 Tbird + KT133A + 2x256 PC 133 CL2 + GF 6800 GT = 8200 points
Athlon 1400 Tbird + KT133A + 2x256 PC 133 CL2 + GF 6800 GT = 10300 points
Athlon XP 1600+ + AMD 760MPX + 2x512 DDR 266 CL2 + GF 7800 GS = 9430 points
Athlon XP 2200+ + NF2 U400 + 2x512 DDR 266 CL2 + GF 7800 GS = 14400 points

The KT133A results are similar to yours (6800GT doesn't make much difference, only small boost in Nature test). AMD 760MPX is just bad (even with DDR memory)... nForce2 really shines and is capable to unleash the full potential of K7.

VIA chipsets usually run slower with more RAM sticks - performance drop is minimal but can be measured in benchmarks. Also some boards can be slower than others, even if most of the hardware is identical. For example these two - http://hw-museum.cz/view-mb.php?mbID=17 and http://hw-museum.cz/view-mb.php?mbID=15 The MSI is 5-10% slower in most benchmarks with exactly same components used and same chipset/memory settings in BIOS. I have no idea why is that.

Yeah that's why I asked if he'd compared memory scores at 512 and 768MB. It sounds like memory performance is dropping for some reason. And with a fast K7 on SDRAM, the CPU is quite memory bandwidth bottlenecked. 3DMark2001 combined with a fast GPU causes the score to be determined significantly by CPU performance (and memory bandwidth affects the CPU).

Swaaye:

Both systems are using the same brand/model of RAM. Although I have not benchmarked it, the memory performance should be identical for both PCs.

Havli:
Thanks for the reassurance! After reading the above comments by you and Swaaye, I am starting to agree that the RAM is the bottleneck. If that's the case, then there probably isn't much point in troubleshooting this any further. I'll just have to accept that the scores are as good as I can get them.

Thank you both for your comments/suggestions!

(Still, shouldn't the Voodoo 5 5500 scores be a bit higher?)

Dou you have the via hyperion Drivers installed?

I think i used 4.32 or 4.41... and the AGP Update.

havli said; "VIA chipsets usually run slower with more RAM sticks - performance drop
is minimal but can be measured in benchmarks. Also some boards can be slower than
others, even if most of the hardware is identical. For example these two -
http://hw-museum.cz/view-mb.php?mbID=17 and
http://hw-museum.cz/view-mb.php?mbID=15 . The MSI is 5-10% slower in most
benchmarks with exactly same components used and same chipset/memory settings
in BIOS. I have no idea why is that."
Although I make no comment in agreement or disagreement with your statement above,
it may be more helpful for you to use a different example in future, as that motherboard
MSI K7T Turbo-R Limited Edition is visibly in a degraded condition - the caps are faulty.

Yes, caps are bad... this can decrease board stability but lower performance is very unlikely.

After reducing RAM to 512mb and tweaking the BIOS memory settings a bit, I have been able to get the system to 9,715 in 3DMark. When using the Voodoo 5, I now get a score of 1653.

If I set the CAS latency in the BIOS to 2, I get a score of about 9,900 for the FX5950 and no change for the Voodoo 5. However, the RAM I'm using is CAS3.

Considering that the improvement of running RAM at CAS2 is only a 2% gain, I don't consider it worthwhile to purchase new RAM. I also feel that 9,700 is about where a FX5950 Ultra should be benching with this system. The Voodoo 5 scores still feel low, but I'll just deal with it. I may also remove the V5 and replace it with a pair of Voodoo 2 cards in SLI. It's awful handy to have a PCI V5 around, just in case.

Thanks again to everybody! I really appreciate your comments and suggestions!

My two cents;

> The V5 is a GeForce 2 era card and is slightly slower than a GeForce 2 Ti, don't expect much from it.
> The GeForce 4 Ti was often a good match for the FX series and regularly beat them in DX8.
> You dont have some weird OEM version of the Athlon 2100 do you? Because some had slower FSB speeds, notably the only way I can run my 7VJL on a 2600+ is by using a more obscure AXDA2600DKV4C, but it does take a hit in performance compared to the AXDA2600DKV4E when used in a board which supports both because the RAM and AGP slow down dramatically.