17. Final resolution: 1600*1200:
A. NF8
B. KU8
C. KV8

While nforce3 and K8T800 are dueling for the first place in 3dmark 2000 with neither of them winning; ULI is performing poorly! This is repeatable; even on the reserve system with XP and 3dmark freshly installed performance of ULI M1689 remains subpar!

shevalier wrote on 2025-11-05, 07:22:

https://soggi.org/motherboards/abit.htm#S754 What BIOS versions do you use on your DUTs? Because 6% for C754 is a bit too much, […]
Show full quote

https://soggi.org/motherboards/abit.htm#S754
What BIOS versions do you use on your DUTs?
Because 6% for C754 is a bit too much, in my opinion.
Moreover, only for one motherboard.

You have seen nothing yet! As I said I am already contradicting established knowledge.

18. 3dmark 2001: "The Benchmark" - the de facto standard for everyone who wants to test his retro system!
1024*768 first without image enhancements:
A. NF8
B. KU8
C. KV8

19. We increase the resolution to 1280*1024 and set everything to maximum details:
A. NF8
B. KU8
C. KV8

20. Final and most demanding settings: 1600*1200. Being a DirectX8 benchmark geforce 7800gs is more than capable of handling it!

Please also include screenshot with detailed results of individual game fps for 3d mark 2003 and later. The score itself can be hard to interpret.

Besides the score each screenshot contains also Everest system info with the system summary: you can clearly see the motherboard name and chipset, the FSB and the multiplier, video card and hard drives/SSD used.
In 3dmark 2001 KV8 is just 5% better than nforce3/M1689 which is exactly the limit of error! Although it is not a clear victory it is still far better than all period correct reviews show!
We are going to the final version of 3dmark tested: 2003! I chose not to test 2005 because of its very heavy GPU bias.

AlexZ wrote on 2025-11-05, 11:57:

Please also include screenshot with detailed results of individual game fps for 3d mark 2003 and later. The score itself can be hard to interpret.

Yes sir! 😀 Already got them with all the details because 3dmark 2003 includes a sound test!

21. 3dmark 2003 default settings: no AA, AF = 4X
A. NF8
B. KU8
C. KV8

22. We increase the settings to the max and the resolution to 1280*1024:
A. NF8
B. KU8
C. KV8

23. Against all odds ULI manages a "prestige"victory with a small lead.
We reached the final and most demanding setting: 1600*1200 maxed out:
A. NF8
B. KU8
C. KV8

nd22 wrote on 2025-11-05, 12:27:

___

I don't know what's going on.
But a 10% difference on the same processor is abnormal.

The poor CPU score for nForce3 is worth looking into or the results could be invalid.

My Athlon 64 3400+ clawhammer gets:
CPU Test 1 109.5 fps
CPU Test 2 14.7 fps

The first value is slightly higher probably as a result of memory latency tuning. You should be able to get at least 105 fps on nForce 3.

Irrespective of weird results for nForce 3, the results of VIA/ULI show that AGP GeForce 7800 GS is totally outclassed when running with anti-aliasing. Built-in game fps is too low for real games to work well. This will likely limit it to year 2004 without AA.

The results also show that the cost of using the built-in sound card is too high - we pay 10 fps for 24 sounds. This could badly affect real game tests as well. Effect of using Sound Blaster Audigy should possibly be investigated for one of the configurations. My AM2+ system uses built-in sound card as well, I should probably put in a Sound Blaster Audigy 2.

3d mark 2003 breakdown, 1600x1200, Athlon 64 3400+, GeForce GTX 275 (anti-aliasing 4 sample, texture filtering anisotropic 16), Sound Blaster Audigy 2 ZS:

• Wings of Fury - 293 fps
• Battle of Proxycon - 131 fps
• Troll's Lair - 113 fps
• Mother Nature - 183 fps
• No sounds - 57 fps
• 24 sounds - 52 fps
• 60 sounds - 48 fps

Final result 19412 3d marks.

shevalier wrote on 2025-11-05, 16:12:

nd22 wrote on 2025-11-05, 12:27:

___

I don't know what's going on.
But a 10% difference on the same processor is abnormal.

I am going to put up results from CPU tests shortly!
Note: Clawhammer CO revision used in all the systems as seen on the previous pages when I put up the screenshots with the system details.

3dmark 2003 shows that the scores across all 3 resolutions for the 3 systems are very close to one another and also that Athlon 64 is bottlenecked by the 7800gs as AlexZ correctly put it.
However 3dmark 2003 is the only 3dmark version with 3 sound tests: because of the limitations of the ALC658 codec only the first 2 can be performed.
While nforce3 and K8T800 results for no sound and 24 sounds are extremely close - within 2% - ULI M1689 experiences a 8% respectively 11% drop in these 2 tests. These will not be the only time that ULI performs poorly when tests include sound!

56. As I said let's skip for a moment the rest of the tests and jump straight to SuperPi. Because all memory timings are the same and I used the same CPU for all 3 platforms the only difference is going to be the board:
A. NF8
B. KU8
C. KV8

With only 2% separating the nforce3 and M1689 from K8T800 the performance of all 3 platforms is practically the same.

57. Cinebench 2003 is the other processor focused test - we are interested in the CPU score only:
A. NF8
B. KU8
C. KV8

KU8 is 1% slower than NF8/KV8 - very small difference. So the processor performance across the 3 systems is really close.