Software environment:
1. operating system: Windows XP SP3 with no updates. If required some extra components are installed on a need for testing basis. Example Catalyst control center requires net framework 2.0. DirectX 9.0c latest SDK from June 2010.
2. monitoring software: everest 5.50 - old version that work just fine on any board I used it on starting from Abit AX5 up to Abit IX48 GT3 and any Windows version: 95 - XP; MSI afterburner 4.30 that works on any CPU including those without SSE such as Athlon 1400C; fraps 3.4.7 for backwards compatibility as it is the latest version that works on processors without SSE2; fraps 3.5.99 for pci-express platforms.
Please note that both afterburner and fraps are running at all times with fraps logging the frame-rate; after each test run I check the log for outliers - 0 FPS.
There is no automation - no Bench'emAll! - all tests are done manually, and afterwards afterburner and fraps log are checked against each other for inconsistencies.
3. all tests are repeated on secondary systems - that is, for example, a second Abit AV8 with another raptor, XP and software installed all over again. There must be no result between the "primary" and "secondary" system with a difference greater than 5% otherwise I scrap all results and start all over again like I did when I started with socket A back in 2018 and screwed up the results with KW7/KT880.
4. and I forgot to mention that all systems are disconnected from the intranet during testing. Because my retro corner is only 4sqm I have to be creative and I use a TP-Link 851ND wireless adapter with an Atheros chip on all Windows XP/2000 machines.

List of tests and settings for AGP platforms:
1. Code creatures 1024*768 default
2. Code creatures 1024*768 default + sound
3. Code creatures 1024*768 max
4. Code creatures 1024*768 max + sound
5. Code creatures 1280*1024 max
6. Code creatures 1280*1024 max + sound
7. Code creatures 1600*1200 max
8. Code creatures 1600*1200 max + sound
9. Aqua mark 3 1024*768 default
10. Aqua mark 3 1024*768 max + sound
11. Aqua mark 3 1280*1024 max
12. Aqua mark 3 1280*1024 max + sound
13. Aqua mark 3 1600*1200 max
14. Aqua mark 3 1600*1200 max + sound
15. 3d mark 2000 1024*768 – default
16. 3d mark 2000 1280*1024 – max
17. 3d mark 2000 1600*1200 – max
18. 3d mark 2001 1024*768 – default
19. 3d mark 2001 1280*1024 – max
20. 3d mark 2001 1600*1200 – max
21. 3d mark 2003 1024*768 – default
22. 3d mark 2003 1280*1024 – max
23. 3d mark 2003 1600*1200 – max
24. 3d mark 2003 1600*1200 – max; no sound
25. 3d mark 2003 1600*1200 – max; 24 sounds
26. Unreal tournament 2004 1024-768 – max
27. Unreal tournament 2004 1280*1024 – max
28. Unreal tournament 2004 1600*1200 – max
29. Tom 2d
30. Pc mark 2002 – CPU
31. Pc mark 2002 – RAM
32. Pc mark 2002 - HDD
33. Pc mark 2004 – CPU
34. Pc mark 2004 – RAM
35. Pc mark 2004 – GPU
36. Pc mark 2004 - HDD
37. Pc mark 2005 – CPU
38. Pc mark 2005 – RAM
39. Pc mark 2005 – GPU
40. Pc mark 2005 - HDD
41. Far cry 1 1024*768 – max
42. Far cry 1 1280*1024 – max
43. Far cry 1 1600*1200 – max
44. Half life 2 original release 1024*768 – max
45. Half life 2 original release 1280*1024 – max
46. Half life 2 original release 1600*1200 – max
47. Doom 3 no AA 1024*768
48. Doom 3 no AA 1280*1024
49. Doom 3 no AA 1600*1200
50. Super pi 1m
51. Cine bench 2003

List of tests and settings for pci-express platforms is not finalized yet; all directx 7 & 8 tests are dropped - I am considering to keep 3dmark 2001 as the only legacy test because it is so loved by the community. As a general rule I consider that late Windows XP era (2005 - 2006) should focus on DirectX 9.
Tests intended for pci-express platforms:
- synthetic tests ( these are very easy to do, easy to check by anyone): Aquamark3 - first Directx 9 benchmark; 3dmark 2003 & 2005 & 2006; PC mark 2004 & 2005; super pi 1m; cine bench 2003; Tom 2D.
- Direct3D games: Farcry 1; Half life 2 steam release; FEAR; Call of duty 2; Serious sam 2; company of heroes 1; The Elder Scrolls IV: Oblivion.
- Opengl games: Doom 3; Quake 4; Prey.
Another thing to note is that by 2006 a transition from square aspect ratio to wide screen started to take place; some games refuse to support anything but 4:3; some, such as FEAR, support widescreen in an unofficial form; some support it natively. In the interest of time I think I will keep the current resolutions that I benchmarked with: 1024*768, 1280*1024, 1600*1200.

Where can I find Tom2D and the games' benchmark utilities?
SuperPi version?

1. Super pi
2. Tom 2D
3. I think for the rest a google drive link would be better!

Some of it is commercial software that has long since been abandoned - example Aquamark3. If moderators/colleagues consider it would be illegal I will not post the link to google drive!
For Mad Onion - as old boys know it - Futuremark software there is an official link to the legacy page of 3dmark/PC mark where you can download it legally with official licenses: https://benchmarks.ul.com/legacy-benchmarks

Here it is: https://drive.google.com/file/d/1lK15a9huTFti … iew?usp=sharing
I am sorry if I broke any rules!
For screenshots I use irfan view because it is so lightweight and compatible but you can use whatever you like!
Page file is always disabled -this is highly important - so 2gb of RAM are absolutely required! I have already encountered games that need more than that - FEAR and QUAKE 4 - so for late Windows XP era I think 3gb should be the minimum.

29 & 30 & 31 & 32. We continue with the results and the next test is PCmark 2004, a completely overhaul suite:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

The first time I tested by mistake the SSD in the socket 939 ULI system. Please note that even with the first generation SATA 150 there is an enormous difference between the cheapest available SSD and the most expensive hard drive at the time of its release.
Memory score is 44% better on socket 939 - a sign that dual channel is working as it should be.

33 & 34 & 35 & 36. PCMARK 2005 is the last version to be tested:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

CPU scores are equals across the board, after all we are talking about the same 2.2 GHz Athlon 64 with 512kb of L2 cache.
29% better memory score reflects the dual channel advantage that socket 939 has.
This time around there is a 10% better graphics score, a sign that dual channel does indeed help with 3d performance and that socket 754 is limiting the radeon x1950 pro.
One more time hard drive performance on Abit AV8 leaves a lot to be desired!

40 & 41 & 42. I misplaced the fraps log for Unreal tournament 2004 for the KV8 pro system so let's skip to the next game while I assemble the system and retest.
The first directx9 game is FARCRY from 2004. Tested with hardware OC benchmark tool at max details:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

Finally we got a greater than 5% advantage for the 2 socket 939 systems. With 6% lead over its single channel brother we are still below the 9.4% required by the PR rating of the Athlon 64. It seems that AMD inflated the ratings of the higher specs Athlon 64 for socket 939. Low end speed grades such as Athlon 64 3000 & 3200 have only a 200 extra PR ratings against the socket 754 counterparts.
VIA has a lead that increases as the resolution increases against the ULI based systems in this game.

43. Next game is HALF LIFE 2. I resorted to fraps to benchmark a 60 seconds interval in front of the train station on the first level, where doctor Breen is making his infamous speech. I walked clockwise around the monument looking forward all the time. I used the original release of the game at maxed out details.
1024*768 first:
A. Abit KU8
2026-02-15 05:26:35 - hl2
Frames: 4360 - Time: 60000ms - Avg: 72.667 - Min: 53 - Max: 92

B. Abit KV8 PRO
2026-03-07 03:44:00 - hl2
Frames: 4930 - Time: 60000ms - Avg: 82.167 - Min: 38 - Max: 113

C. Abit UL8
2026-02-21 04:19:31 - hl2
Frames: 4593 - Time: 60000ms - Avg: 76.550 - Min: 61 - Max: 101

D. Abit AV8
2026-03-03 03:58:16 - hl2
Frames: 5249 - Time: 60000ms - Avg: 87.483 - Min: 67 - Max: 119

In Half Life 2 VIA had an advantage since i started testing 8 years ago on socket A. It seems that this game perform very well on VIA chipsets as evidenced by the huge lead of both VIA systems against the 2 ULI systems.

44. Half life 2 1280*1024:
A. Abit KU8
2026-02-15 05:28:39 - hl2
Frames: 4332 - Time: 60000ms - Avg: 72.200 - Min: 56 - Max: 87

B. Abit KV8 PRO
2026-03-07 03:44:00 - hl2
Frames: 4930 - Time: 60000ms - Avg: 82.167 - Min: 38 - Max: 113

C. Abit UL8
2026-02-21 04:17:58 - hl2
Frames: 4557 - Time: 60000ms - Avg: 75.950 - Min: 57 - Max: 103

D. Abit AV8
2026-03-03 03:56:37 - hl2
Frames: 4773 - Time: 60000ms - Avg: 79.550 - Min: 58 - Max: 107

VIA continues to lead as the resolution grows; again socket 939 has a 5% advantage over socket 754 like in many other tests, still far below what the PR of the Athlon 64 3500 indicates!

45. Half life 2 1600*1200:
A. Abit KU8
2026-02-15 05:24:52 - hl2
Frames: 3934 - Time: 60000ms - Avg: 65.567 - Min: 47 - Max: 86

B. Abit KV8 PRO
2026-03-06 04:43:25 - hl2
Frames: 4188 - Time: 60000ms - Avg: 69.800 - Min: 51 - Max: 92

C. Abit UL8
2026-02-21 04:16:17 - hl2
Frames: 4340 - Time: 60000ms - Avg: 72.333 - Min: 54 - Max: 99

D. Abit AV8
2026-03-07 08:25:54 - hl2
Frames: 4435 - Time: 60000ms - Avg: 73.917 - Min: 57 - Max: 104

At 1600*1200 the socket 754 ULI platform suffers a serious setback, being 10% behind the rest; this is repeatable on both Abit KU8 boards that I have with a fresh install of XP and the game. As you can see, radeon x1950pro is bottlenecked by the CPU in this game as it is capable of nearly 100 FPS. Athlon 64 4000 - A FX-53 in disguise - should unlock the full potential of this card!

49. We reached the only OpenGL game: Doom 3. Extremely tough on the video card, I tested with AA disabled, but even so, the inherent weakness of Ati hardware in opengl games leads to not so great performance!
1024*768 first:
A. KU8
B. KV8 PRO
C. UL8
D. AV8

50. Doom 3 1280*1024
A. KU8
B. KV8 PRO
C. UL8
D. AV8

51. Doom 3 - 1600*1200:

No, I did not screwed up the order!! Under Opengl, ULI M1689 chipset no longer has to use specific BIOS settings and Omega drivers to allow compatibility with bridged AGP Radeons and you can set both KU8 and UL8 to the maximum performance settings in the BIOS - AGP secondary latency timer is the setting to tweak - and install official 10.2 drivers. ULI is faster than VIA in this game thanks to the opengl nature of Doom 3.