That is a motherboard specific problem. There are vogons members who run 4GB RAM at DDR400 in s939, probably even at 1T, but with ECC memory ( Re: Socket 939 dual core build. Decisions, decisions.... ). Asus K8N-E (s754) also allows max DDR333 with 2x 1GB modules, but it is an AGP board so it doesn't matter, if needed there is a workaround.

On Athlon 64 memory always runs in sync with the CPU. Memory divisor is chosen indirectly and typically not visible in BIOS. The memory setting in BIOS influences the divisor, but CPU frequency multiplier influences it as well. You could get around this problem by increasing base frequency.

2209Mhz CPU, 260Mhz base, 8.5x multiplier, HT 1040 (4x), memory divisor 11 (select DDR333 in BIOS, real clock will be 400) - this is the closest you can get to stock clocks, very mild HT OC - should have no effect. In this way you should be able to run 4x 1GB at DDR400. These settings were identified in Re: Battle of the platforms: socket 754!

I have read that topic and nowhere is mention the first generation of Athlon 64! I have not reached dual cores yet; I am interested in getting Clawhammer 3500 to work with 4 dual sided RAM modules at 400 MHz!
I tested with Newcastle and the problem remains: with 2 modules I got DDR-400 at 3-3-3-8 CR 1T, with 4 modules I got DDR-333 at 3-3-3-8 CR 2T!

Abit KN8 SLI getting some TLC; there is no TIM just that rubber like material that you see in the pictures!

First tests done with Athlon 64 3500 Manchester revision with 4gb of RAM and geforce 9800gtx. As I already know that 2004 games run OK I started testing 2005 games.
First was Age of empires 3 that I thought should run smooth on such a system; I started a skirmish with the maximum number of AI players and the experience was not good: after half of hour the game slowed down to a crawl! Well, I thought AoE3 is badly optimised let's run Quake 4. To my surprise the experience is very bad with lows in the 20's and average in the 30-40's! And that is without AA - just ultra quality with advanced settings turned on to compare with the radeon x1950 pro! 9800GTX should run at 60FPS locked with these settings but it seems that CPU matters a lot in this game!
Also in FEAR the in-game test shows practically no difference between 1024*768 ultra quality, AA = 4x, SS on and 1600*1200 with the same settings indicating a strong CPU bottleneck!
It seems that the second generation of directx9 games require besides humongous amount of RAM and a powerful video card a very good processor!

Games from 2005 tend to be CPU bottlenecked, but a powerful GPU is still able to save the day.

s754 at 2.2Ghz with GTX 275 - Re: Battle of the platforms: socket 754!

Maybe try a few other GPUs before committing to one. I originally had GeForce 9800 GT, tried also 9800 GTX, GTX 260 and ended up with GTX 275. SATA ports must not be completely blocked. At least 2 must be usable.

Abit boards for socket 939 have a very good layout, no SATA port is blocked off!
I used a brand new Kingston SSD for a fresh install of XP, no radeon drivers were ever installed, geforce 197.77 installed form the get go!
I suspect that msi afterburner, fraps and background Windows processes take a lot of CPU power; in period correct reviews I read that a single frame rate monitoring program was used usually fraps. I will test with Athlon 64 X2 4200 which is a 3500*2 to see what results I will get!

Installed Athlon 64 4200 which is exactly a 3500 X 2. The results are unbelievable!
Athlon 34 3500:
1. Quake 4
2. Fear

Athlon 64 X2 4200
3. Quake 4
4. Fear

Quake 4 is one of the earliest games that support dual cores processors: it shows a 24% (!!!) improvement over the single core. Fear is a single threaded game yet it still shows a 14% better result with the dual core! No period correct review ever showed such an improvement when they reviewed Athlon 64 X2 back in 2005! Age of empires 3 became playable with the dual core. It is impossible that just one year later games would show such a drastic uplift in performance with the first consumer dual cores!
Actually Quake 4 has become very playable with averages around 50 -60 FPS and it does not drop below 30 anymore! Also Fear feels smoother!
Waiting for others for confirmation!
Athlon 64 3500 - wrong screenshot selected the first time around, this are the results for 1024*768 max details, SS on, AA 4x:

It doesn't surprise me.

Athlon 34 3500+ has only 512KB L2 cache, so it will be a bit slower than CPU with 1MB L2 cache. AMD gave us 2 glued memory controllers and took away cache to save cost. You also run at CL3 due to using 4 memory sticks. I don't have identical GTX+, just GTX so I can't verify it but the results make sense. Slightly worse than GTX 275. If you put in a better GPU, you will get slightly better results.

Athlon 64 X2 4200 here managed to match Turion at 2.4Ghz, while suffering from L2 cache size penalty, cas latency penalty and worse GPU. You should be having very similar experience in games. X2 CPU really deserves 4GB RAM.

Performance wise, GeForce 9800 GTX+ sits between GTX 260 and GTX 275. GTX 275 has significantly higher memory bandwidth, pixel rate and more memory. It isn't a bad card at all.

X2 CPU helps a bit, but these early games don't benefit from dual core that much.

In Quake 4 I normally use 1280x1024 with 4x AA (8x overloads the GPU) and ultra quality. Frame rate is practically locked at 60fps. I don't see fps drops.

In FEAR I didn't use SS except until the end as you didn't use it with Radeon X1950 Pro either. I also enabled HW audio mixing and EAX 2.0, not sure about the effect on FPS.

In Age of Empires III I use 1280x960 with maximum details. After one hour of easy game and defeating 2 enemies (1 ally is alive) I get 30-60 fps. Lows are reached when looking at my own army of 50 units. Normally it is about 40 fps.

So you want to tell me that games in 2005 are already started to take advantage of dual cores processors? It seems surreal; I am expecting games from 2007-2008 (Vista era) to utilise more than 1 core.
I tested 2 games from 2004: Farcry (directx9) and Doom 3 (opengl). Both show practically the same frame rate with the Athlon 3500 and with the Athlon 4200.
Just one year later Quake 4 which is built on the same engine as Doom 3 show 24% better performance with a dual core! Fear is known to be single threaded yet across all 3 resolution performance jumps 10-14% with Athlon X2 4200.
I installed late last night Serious Sam 2; will test it this afternoon.

The small boost you see in FEAR is probably just all non game processes running on another core, perhaps even game sound. Games that do not benefit probably utilize just one thread. I would expect like you, mostly games from Vista era to benefit substantially. 2.2Ghz X2 is too slow for that though, the situation is similar to 2.6Ghz X4 - they are too slow for era when they become relevant. It helps you cover 2005/2006, that's about it.

If you have to use CL3 for 4GB RAM, then it's better to use ECC memory so that you can run with CR 1T.

I didn't find even Windsor 3.2Ghz X2 sufficient for Vista era. GTA 4, Far Cry 2 and Need for Speed Undercover ran really poorly on it.

Serious Sam 2 tested with Athlon 64 X2 4200 and geforce 9800gtx. The frame rate goes down as the resolution goes up indicating a video card limitation; still 77 fps at 1600*1200 max details is very good!
The jump in hardware requirements in just 1 year is mind blowing: in 2004 with an Athlon 64 any model and geforce 7800gs from nvidia or radeon x1950 pro from ati you could play any game at max details with the sole exception of Doom 3. After just 1 year I already discovered 2 games that are CPU limited with an Athlon 64 3500 - AoE3 and Quake 4 - and already require a dual core. Fear built-in benchmark at 1024*768 and 1600*1200 indicate also a processor bottleneck: the frame rate barely goes down by 3 FPS despite increasing the resolution from 1024*768 to 1600*1200 (nearly 2.5X more pixels) and that is with an Athlon 64 3500!
Also the video card necessary to push the 2005 titles to the max is far beyond anything I ever imagined: geforce 9800gtx+ is actually bottlenecking the system; I installed a 8800gtx and that card performs better in Quake 4 at the same ultra quality settings simply because it has a 768mb frame-buffer! I think GTX 275 used by AlexZ is the minimum required to break past the GPU limitation in 2005 games.

By comparison Fear benchmark with Athlon 64 X2 4200 at 1024*768 where you can clearly see that the dual core distances itself from the single core and 1600*1200 where the video card is the limiting factor:

What is astonishing is how soon these CPUs became obsolete. They had lifespan of about 2 years.

If you bought 3500+ (E4) San Diego in 2005, it was obsolete in 2007 as it got outclassed in new Vista era games and by cheap AM2 X2 CPUs. 2.2Ghz runs even NFS Carbon (2006) poorly. You would not have bought X2 in s939 as they were very expensive.

You would have done much better to buy 4400+-4600+ EE (F2) X2 in 2006, but since they ran only at 2.2-2.4Ghz, they would have been obsolete by 2008.

The hardware and software was advancing so fast back in the day that in 3 years, regardless of what you bought, your system was obsolete! I had for 13 years a 2600k and I only changed it because it died! It was unthinkable to keep a system for so long during the 90's and the 2000's and still do something productive with it - not gaming - office work, browsing, etc. You could not use in 2000 a system from 1990 for anything useful, in 2005 a system from 1995, neither in 2010 a PC from 2000, nor in 2015 a system from 2005. Core 2 duo changed everything - the hardware finally caught up with the software - actually I am talking about core 2 quad here; and than sandy bridge was the last time a huge leap appeared. My best friend has a 3770k which is basically a die shrinked sandy bridge as his main computer and that works just fine for excel, web browsing, as HTPC connected to a TV. A 14 years old system that still runs original hardware with the exception of the storage drives!

I installed last night Ghost recon advanced warfighter from 2006. As always the latest patch which is version 1.35 is installed. Tested first time with the Athlon 64 X2 4200 and the performance is very good with averages in the 80-90's at max details at 1024*768. This game has an engine which uses deferred lighting, a particular type of rendering different than "conventional methods"such as per-pixel lighting. The problem is very complex to be detailed here.
Because of that some options that are usually found in other games are missing here, for example antialising. 9800GTX+ is more than capable of mastering the graphics in this game. However the moment I switched to the Athlon 64 3500 the frame rate dropped in the 50-60's at 1024*768. Around 50% better performance with the dual core! What is worse are the extreme swings in frame rate with the Athlon 64 3500: one moment it drops in the low 40's, the next jumps to the high 70's!

This is getting ridiculous! Serious sam 2, a single threaded game has 25% (!!!) better averages with Athlon 64 4200! By now it is clear for me that you actually require a dual core CPU far earlier than previously thought.

I think I know why Ghost recon advanced warfighter perform so poorly on the single core Athlon 64 3500. The game has support for Ageia physx. If one does not have it than all calculations are done by the CPU. There is no support for GPU based physx regardless of the nvidia card you got in your system.
This is what practically kills the poor Athlon.
It would be interesting to test some high clocked FX such as FX-55 to see if the extra clock speed and extra cache could make up for the "loss" of the second core.

I can confirm Ghost Recon Advanced Warfighter (2006) runs at about 50-60fps on single core system. It is a slow paced stealth game so it should not be a problem, unlike in Unreal Tournament 3. It can briefly drop to 30 fps in helicopter scene in the 1st mission when missiles are fired at it. I didn't observe sudden swings, it rarely drops below 45 fps. I forced activated v-sync through driver as the game doesn't have it.

I have legacy PhysX 9.10.0222 installed so PhysX should work through the GPU. It shouldn't matter that I don't have Ageia card. On AMD GPU physics should run on CPU with reduced accuracy.

The last driver from NVidia that should work is https://www.nvidia.com/en-us/drivers/physx/ph … -legacy-driver/

It specifically states:

Games that require this installer include:
Ghost Recon Advanced Warfighter

I do not think it runs through your GTX 275! The release notes even state: "These SDKs run on the CPU only." So all physics are calculated by the CPU.

You are correct. NVidia played us. The legacy PhysX driver supports SDK versions 2.7.0 and prior as well as 2.7.2. It utilizes CPU only. They were lazy to implement the original Ageia PhysX in GPU. Only newer games from about 2008 (or those patched) support newer SDK which utilizes GPU.

After adding PhysX mod to Unreal Tournament 3 and playing PhysX maps I get like 5-9 fps instead of 50. So there is no chance it will work well on s939 either.

It seems PhysX was never about 100% GPU acceleration, some of it runs on CPU.

If you install Sysinternals Process Explorer, you can open the process exe and go to threads tab. You will be able to see per thread CPU utilization. That way you can see how many threads is the game effectively using. It would be interesting what you see for those games where you get significant performance boost. It would be best to run the game in window or have 2 monitors while benchmark is running.