21. 3dmark 2003 is the first version with one directx9 test: mother nature! It is also the only 3dmark with 3 sound tests: no sound and 24 sounds which can be performed by anything including the AC97 codec found on all 3 boards and 60 sounds for which you need at a minimum creative live!
1024.768
A. NF8
B. KU8
C. KV8

22. 1280*1024: we are already GPU limited!

23. 1600*1200: ULI manages 2 prestige victories at 1280 and 1600 with a small 2% lead!
A. NF8
B. KU8
C. KV8

24 & 25. All the general scores are pretty close so no clear winner here!
The sound tests are far more strenuous in 3dmark than in previous benchmarks. We see a big performance hit on the ULI board: 10% behind nforce3 and 12% behind K8T800! While this may seem irrelevant now in all games tested with sound enabled the impact will be very high!

26 & 27 & 28. Synthetic 3d tests are over; it is time to test system performance with PCMARK 2002. There are 3 scores: CPU, memory and HDD:
A. NF8
B. KU8
C. KV8

Less than 1% separates processor and memory score as it should be! nforce3 storage controller does not perform well being slower than the other 2!

29 & 30 & 31 & 32. PCMARK 2004 is next and it adds a graphic test:
A. NF8
B. KU8
C. KV8

Again AGP performance on nforce3 is not great as it lags behind by quite a bit! ULI has excellent AGP implementation as it wins by a large margin!
CPU and RAM scores: nothing to see!
The hard drive controller of nforce3 is really poor while on ULI it is absolutely excellent!

33 & 34 & 35 & 36. The final PCMARK version is the 2005 one.
A. NF8
B. KU8
C. KV8

And there is why NF8 scores lower in games: weak AGP on NVIDIA's nforce3 leads to lower performance and because all but 12 of the benchmarks test 3D performance there is no way NVIDIA can win anymore!

3400+ 2.4Ghz seems to be doing fine. Synthetic benchmarks have higher score, games should be about the same as 2.2Ghz Clawhammer.

There are still some on sale https://www.ebay.de/itm/236372790447 . Price is ok, still not too expensive. Someone in Canada is selling it with VIA board https://www.ebay.de/itm/257099934428 for even better price.

Yes, that is why I chose not to test with top of the line clawhammer 3700. The price is just ridiculous. Second best always has the best price to performance ratio! And Choosing Athlon 34 3400 has the additional benefit of testing the old cache versus frequency problem!

37 & 38 & 39. We finally reached the games and the first to be tested is Unreal tournament 2004 GOG version. Because I could not make Umark to run I used fraps for 1 minute benchmark across all the resolutions on the death match map Antalus which was a very popular map to test with back in the day!
I put all the results for the 3 resolutions for each system, one of a time:

A. NF8
2025-07-30 18:56:25 - UT2004
Frames: 7297 - Time: 60000ms - Avg: 121.617 - Min: 78 - Max: 169

2025-07-30 18:57:46 - UT2004
Frames: 7390 - Time: 60000ms - Avg: 123.167 - Min: 79 - Max: 195

2025-07-30 18:59:05 - UT2004
Frames: 7751 - Time: 60000ms - Avg: 129.183 - Min: 79 - Max: 196

B. KU8: 1600 first, 1280 second, 1024 third:
2025-11-10 17:28:27 - UT2004
Frames: 5331 - Time: 60000ms - Avg: 88.850 - Min: 71 - Max: 119

2025-11-10 17:29:44 - UT2004
Frames: 6410 - Time: 60000ms - Avg: 106.833 - Min: 72 - Max: 149

2025-11-10 17:31:05 - UT2004
Frames: 7475 - Time: 60000ms - Avg: 124.583 - Min: 84 - Max: 181

C. KV8
1600*1200
2025-08-22 14:46:46 - UT2004
Frames: 7333 - Time: 60000ms - Avg: 122.217 - Min: 86 - Max: 180

1280*1024
2025-08-22 14:50:03 - UT2004
Frames: 7628 - Time: 60000ms - Avg: 127.133 - Min: 78 - Max: 196

1024*768
2025-08-22 14:48:25 - UT2004
Frames: 8237 - Time: 60000ms - Avg: 137.283 - Min: 97 - Max: 186

Because I used fraps - so I tested with sound enabled ULI takes a hit and drops to last place! VIA is first with NVIDIA close behind.

40 & 41 & 42. FARCRY is next and is the first DirectX9 game to be tested with Newcastle:
1024*768
A. NF8
B. KU8
C. KV8

Because I used HardwareOC farcry tool - so sound was disabled - all 3 systems perform neck in neck here!

For UT2004 results please add resolution to every result. You are posting results per board instead of per resolution as in Re: Battle of the platforms: socket 754! .