AlexZ wrote on 2025-11-07, 09:06:

Does nForce3 run in AGP 8x mode and 66Mhz? See computer->overclock tab in Everest if there is anything unusual. It would be great to run Everest/Aida64 memory/cache test to see memory latency. This would rule out poor configuration by BIOS.

Yes it does! I put up the BIOS and Everest pictures and I will assemble the nforce3 system again for the 4 repeat of the tests.

26. The last synthetic test is Tom2D. It is the only GUI test in the entire suite. The user swaaye to whom I thank recommended this back in 2018 when i started testing socket A.
In fact he started using and recommending this benchmark back in 2011: VIA vs. 2D speed when he noticed abnormal (low) speed in Windows XP 2D desktop.
The fact that swaaye did noticed the problems with a VIA K8T800 board makes this all the more interesting. He actually got a few numbers: Re: VIA vs. 2D speed
Back on socket A nforce2 CRUSHED VIA in Tom2D:
A. AN7/nforce2 ultra
B. KV7/KT600

The score/speed of Tom2D is dependent on the particular CPU model/revision, motherboard and chipset driver version, video card and driver version. So you can actually have the same system, change the processor from clawhammer to Newcastle for example and the score will be different. Or you can install an older version of VIA 4 in 1 and presto, you get a completely different outcome!
I think by now you realize that I am writing all this because i am going again to contradict the whole world!

So here are the scores for all AGP socket 754 systems:
A. NF8
B. KU8
C. KV8

User 1: Whaaaat??
User 2: that's impossible!!!
User 3: get a moderator in here!!!

After the humiliation it got on socket A, VIA got its revenge! I repeated the test dozens of time and the scores remain. I tried multiple drivers on both nforce3 and m1689 and every single time the score remain just as bad. Also tried multiple forceware version and no luck! NF8 and KU8 gets killed by KV8 in this test. I got over 10 screenshots and all are bad for ULI and NVIDIA nforce3.
On VIA you need chipset driver version 4.56v in order to make Tom2D "work". You will get excellent scores regardless of the driver version you use for the geforce 7800gs.

Because of the gargantuan difference between K8T800 and its competitors this single test will weight heavily in the final score!
For NVIDIA there is still hope - hint: nforce4 - however for ULI there is none!

So it would seem VIA eventually did produce a high quality chipset and drivers. Based on the results so far VIA K8T800 seems to be the best for s754 with AGP. On my nForce4 I get 2754 points.

When I said that a simple CPU change from one revision to another "breaks"Tom2D I was not kidding! In anticipation of part 3 here are the scores for KV8 PRO with radeon X1950 PRO:
A. CLAWHAMMER
B. NEWCASTLE

You need a perfect combination of motherboard make and model + chipset driver + video card + graphic card driver + specific CPU revision in order to make Tom2D "work".

I suspect this is caused by if/else code in graphics card driver. If you have the right CPU and chipset driver then acceleration gets enabled. To be fair tested GPUs are from the end of XP/beginning of Vista era when GDI was no longer hardware accelerated. Tested CPU is weak so it isn't that great.

I also suspect this. The problem I got is that back on socket A everything was clear cut: on Abit KV7/KT600 with every single CPU and regardless of the video card used Tom2D was broken and the score was abysmal! On Abit AN7/nforce2 with every single Barton CPU - I need to correct myself as I searched the screenshots to be sure! - and all video cards Tom2D was fine and the score was just like in the screenshot: 40000+ points.
So VIA = bad; nforce2 = good; by the way nforce1 was bad!
On socket 754 you need a very particular combination of CPU revision; motherboard make and model; chipset driver version; specific video card and driver to get Tom2D to "work".
I exaggerated when I said that you can use any forceware with 7800gs on KV8: you have to use version 197.77 or earlier; any 200 series or later brakes Tom2D! because the strongest NVIDIA AGP card is from the 7000 series using 175.19 like I did poses no problem.

I'm playing Need for Speed: Most Wanted on s754 PCIe right now. It's a big upgrade over Need for Speed: Hot Pursuit 2.

Need for Speed: Hot Pursuit 2 felt great compared to Need for Speed III: Hot Pursuit until you find out there are only like 3 maps with variations like reverse or different path on the same map and the campaign is quite static.

On AGP you will likely be limited to Need for Speed: Underground 2 which features only night driving in city which is quite small. It feels cheap compared to successor Need for Speed: Most Wanted.

Need for Speed: Most Wanted is quite good compared to all predecessors. Graphics looks great (weather is fantastic), map is like 3x larger than in Need for Speed: Underground 2, has both open country (you can even drive in golf course) and city, lot of objects can be destroyed while your car is indestructible, fun with police chase (where like 10 police cars can be after you) and more interesting challenges with chance to win opponent's car. I got Toyota Supra from opponent, great RWD car, won me many races, I got Mazda RX-8 from opponent too but it is useless, bought Mitsubishi Lancer as being AWD is great for police chase later in the game when WW Golf is no longer enough, I got Porsche Cayman S from opponent, being RWD is great for racing against opponents. In police chases RWDs are bad as it's easy to get stuck, FWD or AWD works best. When racing against opponents RWD is best as they tend to oversteer which can be used to advantage.

Need for Speed: Carbon can no longer be played comfortably as it stutters too much.

I do not like too much MW despite gorgeous graphics and aspect because of the rubber band AI: early on AI controlled opponents simply brake in order to let you win and later in the game they "adjust" their skills in order to catch up with you no matter how good you are driving. For me the favorite NFS remains Porsche unleashed.

Now that we are finished with synthetics it is time for the games.
The first one is Unreal tournament 2004 one of the last DirectX8 games, a really good arena shooter. Umark benchmark utility refuses to run with UT 2k4 GOG edition that I have so I used fraps to test the Death match Antalus map in spectate mode. After loading the map I waited 1 min before pressing F11 for benchmark. After 1 minute of testing I changed the resolution and waited again 1 minute before benchmarking and so on. I repeated 3 times and kept the final 3 runs at 1024*768, 1280*1024 and 1600*1200. Please note that I used maximum details and also sound!

Because Fraps outputs the results to excel and .TXT files I do not have any screenshots!
38. UT 2k4 - 1024*768
A. NF8:
2025-07-29 14:52:13 - UT2004
Frames: 7633 - Time: 60000ms - Avg: 127.217 - Min: 86 - Max: 190

B. KU8:
2025-10-30 13:12:55 - UT2004
Frames: 7604 - Time: 60000ms - Avg: 126.733 - Min: 77 - Max: 181

C. KV8:
2025-08-20 15:53:41 - UT2004
Frames: 8069 - Time: 60000ms - Avg: 134.483 - Min: 90 - Max: 208

VIA establishes an early lead...
Please note that 7800gs is more than capable of running UT at max details so we are not GPU limited!

39. UT2K4 - 1280*1024
A. NF8:
2025-07-29 14:51:00 - UT2004
Frames: 7367 - Time: 60000ms - Avg: 122.783 - Min: 81 - Max: 180

B. KU8
2025-10-30 13:14:33 - UT2004
Frames: 6483 - Time: 60000ms - Avg: 108.050 - Min: 79 - Max: 161

C. KV8
2025-08-20 15:55:04 - UT2004
Frames: 7913 - Time: 60000ms - Avg: 131.883 - Min: 87 - Max: 173

VIA is still keeping its lead over the competition which in the case of KU8 suffers quite a lot as we increase the resolution!

40. UT 2K4 - 1600*1200
A. NF 8
2025-07-29 14:49:29 - UT2004
Frames: 6887 - Time: 60000ms - Avg: 114.783 - Min: 73 - Max: 185

B. KU8
2025-10-30 13:15:58 - UT2004
Frames: 5996 - Time: 60000ms - Avg: 99.933 - Min: 71 - Max: 135

C. KV8
2025-08-20 15:52:23 - UT2004
Frames: 7618 - Time: 60000ms - Avg: 126.967 - Min: 83 - Max: 196

Excellent AGP implementation enables VIA to score the first victory in the gaming tests with UT 2004!
ULI scores are worse as the resolution is increased. This is repeatable and will not be the only time!

41 & 42 & 43. Farcry is the first Directx9 game to be tested. Usually highly CPU bound; geforce 7800gs is more than capable of running this game with all the bells and whistles so I tested at max details. I used hardware oc farcry becnhmark tool that works perfectly!
A. nforce3 is first:

B.M1689 is second:

C. K8T800 is last: