On the upside, using timings like 3-3-3-8 would make the system more easy to compare to single core s939 you tested before to answer what do we gain from an additional core. This would require GeForce GTS 250 instead of GTX 460.

We know that Command Rate 2T causes significant underperformance and GTX 460 will not be ideal either. Every little piece of optimization helps - it is up to readers to try to squeeze maximum possible performance out of s939 if they so desire and show us the results how far it can go without any dangerous overvolt or risking lifespan of components.

Not only that but also the fact that I am forced to use newer drivers! Anything after 197.77 causes a significant driver overhead - yes, even with a dual core X2 4800 - so this is another reason why older drivers are required on a socket 939 system. With the same GTX 285 I got better results with 197.77 across all the benchmarks!
Different drivers makes any comparison with the previous results invalid. Also the fact that we moved from single core to dual core and from GTS 250 to GTX 460 makes any kind of comparison with the results from chapter 1 where I tested single cores CPU's like comparing apples to oranges.
I finally found a driver that is compatible with all the games and benchmarks on both systems: 275.33! I will use it for testing.

Software used is also different from the previous tests:
1. Fraps 3.5.99 - latest version
2. MSI afterburner 4.50
3. irfanview for all screenshots
4. everest 5.50
5. GPUZ 2.5.0

Synthetic benchmarks:
1. Aquamark3
2. 3dmark 2001
3. 3dmark 2003
4. 3dmark 2005
5. 3dmark 2006
6. Tom 2D
7. WPrime 2.10
8. Cinebench 10.0
9. PCMARK 2004
10. PCMARK 2005

Games:
1. Farcry
2. Half life 2 steam version
3. Doom 3
4. Serious Sam 2
5. FEAR
6. Call of duty 2
7. Quake 4
8. Company of heroes
9. Microsoft flight simulator X original release
10. Prey

Games that didn't make the cut: Chronicles of Riddick: escape from Butcher bay - impossible to get repeatable results! NFS most wanted - the same.
Splinter cell chaos theory refuses to install on VIA system - tested with a secondary AX8 and a fresh install of XP! Ghost recon advanced warfighter - impossible to benchmark, results are all over the place!

All synthetic benchmarks are tested first at the default settings and maxed out details in subsequent runs. All games are tests at max details with the exception of Doom 3 and Quake 4 that are tested with AA 8X only. Note that the Achilles heal of GTX 460 is the 256 bit memory bus. GTX 285 has a 512 bit bus that, at very high AA levels such as 8X-16X, manages to overcome pretty easily the GTX 460 (and even radeon 5870 if I use AA 16x in Doom and Quake).
Monitor used is the good old Eizo S2100 with 1600*1200 max resolution. All tests done at 1024*768, 1280*1024 and 1600*1200.
All games and benchmarks are updated to the latest version - example SP2 installed on FSX, Prey updated to 1.4 and so on.

You could play around with Codecreatures benchmark for extra synth. one (it's older, but it should still show a difference if CPU is limiting max. vertices/triangle values). - default test runs three resolutions and gives a score.

CPU only : 7-Zip (for memory performance), or older Blender versions are also options.

Thank you sir for reminding me about PCMARK! I simply forgot about it! Added to the test suite.
I did used Codecreatures pro back on socket 939 with AGP, socket 754 and socket A. The benchmark is not suited for high performance dual cores like the ones I am using here IMHO.
I found 7-zip version 4.65 - one of the earliest version that is multi-threaded. I am going to test it to see what results I got.

Side note 1: the default BIOS setting for both boards is 200 FSB - it's not FSB actually but I shall use this term. However in actual practice the default FSB for AX8 is 204, while on AN8 it is 201. Neither board allows you to select a FSB lower than 200! So you are starting with a 2% overclock on both systems because I manually set the FSB to 204 on AN8 in order to match its rival, AX8! Because I can not lower the FSB to "set" it at the default 200 it means all results will show this overclock, so take it into consideration.
Side note 2: msi afterburner is running at all times - you will see it in the screenshots. This will lower the scores on all 3D tests.
Side note 3: fraps is recording the frame rate at all times so again, this will impose a penalty on the scores.

7-zip 9.20 (32-bit) doesn't work on S939 ?

Downloaded 7-zip 9.20 from www.oldversion.com to test both this version and the older 4.65.

What if you select base frequency 201-202 Mhz?

For Company of heroes you need to force disable vsync in driver overrides, the latest version of game doesn't allow to turn it off. It then affects benchmark results.

You may want to replace Doom 3 (2004) with Enemy Territory Quake Wars (2007) which is better suited for future testing with its newer engine. Quake 4 remains as representative of the old id Tech 4 engine and for comparison with single cores. The benchmark gets higher fps than Quake 4 benchmark, but the game is tougher on hardware than Quake 4 with its open maps. This game seems to have an online community. It is abandonware, thus free.

zuldan wrote on 2026-09-11, 03:21:

nd22 do you have the ABit Fatal1ty AN8 SLI with ABIT Guru Front Panel (GP-02) in your collection?

You mean this one:

AlexZ wrote on Yesterday, 14:49:

What if you select base frequency 201-202 Mhz?

For Company of heroes you need to force disable vsync in driver overrides, the latest version of game doesn't allow to turn it off. It then affects benchmark results.

You may want to replace Doom 3 (2004) with Enemy Territory Quake Wars (2007) which is better suited for future testing with its newer engine. Quake 4 remains as representative of the old id Tech 4 engine and for comparison with single cores. The benchmark gets higher fps than Quake 4 benchmark, but the game is tougher on hardware than Quake 4 with its open maps. This game seems to have an online community. It is abandonware, thus free.

200 is actually selected as you can see in the pictures below on the main page of Abit AN8 ultra. However in Windows CPU-Z correctly identifies the base frequency as 201.25. If I select 201 I get 202.25, if I select 202 I get 203.25 and so on. I can not go lower than 200 on both AN8 and AX8 boards.

I am sorry but Quake wars is a 2007 game. I stopped specifically at 2006 because I do not want to pick titles individually. As I explained in an earlier post dividing by years is far more easier: all games from a particular year run at max details than you are good to go with your socket 939 system. If I start to test game by game I am never going to finish this project. And yes, I did test Quake wars on LGA 775, however that game eats 1GB of VRAM at ultra details even with only 4X AA and keeps on demanding more!