D. NFORCE4

It's a tossup. 8800gtx is actually the minimum video card required to play Quake 4 at ultra quality and even that card can not play it with AA enabled!

I can confirm that. You would be enabling the lowest AA perhaps, not the highest. It's good to have at least 896 MB GPU RAM. Some slim GeForce 9800 GT do have that. If you have any good slim GPU that doesn't block all the SATA ports perhaps you could re-test just Quake 4 on nForce 4.

I do have a geforce 9800gt 512mb 1 slot card and still it is ram limited- the game needs more than 512mb of VRAM at max details! On a 8800GTX it eats all the VRAM even with AA disabled. Performance is of course much better than with the radeon x1950 pro but still unplayable. I have to test GTX 285 - the most powerful "classic era" card to see if I can play Quake 4 at ultra quality, advanced settings all on and AA =16x comfortably.

52. Final game to be tested is FEAR. Extremely tough on the video card, i tested with soft shadows - SS - disabled and is still limited by the radeon! In fact I tested this game on the Athlon XP 3200 and I got the exact same averages! In retrospect I consider that I made a mistake by including it in the test suite. PCI-express platform with minimum 8800gtx is required for FEAR.

1024*768
A. nforce3
B. VIA
C. ULI
D. nforce4

I tested on the Abit KU8 without MSI afterburner enabled. I have to retest and take the screenshot!

53. FEAR - 1280*1024
A. nforce3
B. VIA
D. nforce4

54. FEAR - 1600*1200
A. nforce3
B. VIA
D. nforce4

nd22 wrote on 2026-01-16, 09:26:

I have to test GTX 285 - the most powerful "classic era" card to see if I can play Quake 4 at ultra quality, advanced settings all on and AA =16x comfortably.

I don't think 16x AA will work well, but 4x should be fine. We probably need GTX 680 for 16x AA. GTX 580 is just a refinement of GTX 280.

GTX 285 should also allow soft shadows in FEAR.

GTX 580 and 480 share the same architecture and share very few in common with the Tesla architecture. I already testes GTX 285 in FEAR and I can play it well above 60 FPS at 1600*1200, AA= 4x, SS on.

52 & 53 & 54.FEAR on the ULI M1689 system:
I. 1024*768
II. 1280*1024
III. 1600*1200

As expected, there is absolutely no difference whatsoever!

And there you have it! All socket 754 Athlon 64 revisions have been tested on all available Abit manufactured boards. Based on the results radeon x1950 pro is too strong for Athlon 64 3400 - you are leaving a lot of performance on the table - while the geforce 7800gs proves to be a perfect match with no processor or video card bottleneck!
Let's see if with a stronger card we got some differentiation between clock speed and amount of Level 2 cache. Clawhammer gets a 100% base result:

Abit NF8 PRO/NFORCE3
CH: 100%
NC: 99.813%

Abit KV8 PRO/K8T800 PRO
CH: 100%
NC: 98.442% - the difference is based entirely on the astronomical difference in Tom2D!

Abit KU8/ULI M1689
CH: 100%
NC: 100.52%

Abit NV8/NFORCE4
CH: 100%
NC: 99.22%

So, again, AMD performance rating is spot on! 22 years later, I, nd22, a nobody, validated the PR rating of the socket 754 Athlon's and found that 512kb of L2 cache = 200mhz 😀. Athlon 64 3700 has an artificial inflated PR most probably because it is the top dog for the socket!
Now let's see the standings when comparing the 4 platforms.

In order to not skew the results because of the "anomaly" of the Tom2D score with the Clawhammer installed in the VIA system the reference point is going to be Abit KV8 PRO with K8T800 PRO chipset - otherwise we would have something like a 25% lead of the VIA platform against the competition and itself!

The first generation of Athlon 64 - CLAWHAMMER
1. NVIDIA nforce4 - 100.71%
2. VIA K8T800 pro - 100%
3. ULI M1689 - 96.60%
4. NVIDIA nforce3 - 96.18%

The second generation of Athlon 64- NEWCASTLE
1. NVIDIA nforce4 - 102.49%
2. VIA K8T800 pro - 100%
3. ULI M1689 - 99.77%
4. NVIDIA nforce3 - 98.42

Interesting outcome in my opinion agp wise looks like i need a K8T800 Pro board that supports Newark and im golden. so would these results transfer over to 939 seeing how its a difference of single channel memory vs dual channel? looks like i should be able to test that. thanks again.

*Edit kv8 pro does with bios 23 E6 revision support.

Newark makes sense if you already have the CPU, especially the 2.4 or 2.6 Ghz version. It's a slightly better alternative to Clawhammer, both with 1MB L2 cache. Price of this CPU is prohibitive.

AlexZ wrote on 2026-01-19, 22:28:

Newark makes sense if you already have the CPU, especially the 2.4 or 2.6 Ghz version. It's a slightly better alternative to Clawhammer, both with 1MB L2 cache. Price of this CPU is prohibitive.

yes i do have a 4000+ newark and a cg 3700+ but only nf3 250gb boards. only thing holding me back on a via board is the w98se bug, possibly. It shouldn't matter really to be honest .

greasemonkey90s wrote on 2026-01-19, 18:24:

so would these results transfer over to 939 seeing how its a difference of single channel memory vs dual channel?
*Edit kv8 pro does with bios 23 E6 revision support.

Hold that thought! Because the K8 micro architecture is the same regardless of the socket and chipset used I think I got an idea 😀

Now that socket 754 has been investigated with every single CPU generation and every chipset (minus SIS) available I will make, based on the results that you have seen, a few recommendations and a practical example of a socket 754 system. I am aware that there are many people who have done more extensive tests that I have done and are far more knowledgeable about socket 754 so I welcome any opinion or critique.
The "base assumption"is: you want to build a system to enjoy 2000-2004 games without too much money and without too much hassle - so it has to be as problem free as possible. I strongly consider that going beyond 2004 is a mistake, if you want to play 2005 or later games you really should consider socket 939. And yes, there are later games that can be played at 60+FPS with max details on a socket 754 PC however I consider all games released within a particular year as good performance with an average of 60 FPS as possible candidates for such a system and that rules out many titles.
Unlike socket A when you have 2 possibilities: A: a best price to performance option and B: an ultimate Athlon XP build, on socket 754 you really only have 1 option: a combination of A and B!
Prices are low so you should not spend lots of cash to build such a system, also one very important advantage of a socket 754 system is: 12V heavy so fully compatible with modern power supplies!

I should mention from the outset that everything I will say is strictly about what I - nd22- have personally worked with and know it to be good! It's obvious that the motherboard suggested is going to be an Abit 😀.