We finally reached the games and the first to be tested is Farcry 1 from 2004, a single threaded game that benefits greatly from a powerful CPU!
1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

While the first 3 generations of Athlon 64 - Clawhammer, newcastle and Winchester - perform neck in neck, Venice and Manchester have 6% better performance! Again there is a significant delta between early and late Athlon's!

The second game is Half life 2 original release. I choose the original because the Steam version is practically Half life 2.5 because it is heavily modified and much tougher on the CPU and GPU. 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 using WASD and no mouse look. When we will get to the dual cores I think I will use the Steam version as it has an integrated benchmark.

1. Clawhammer
1024*768:
Frames: 4193, Time (ms): 60000, Min: 47, Max:83, Avg:69.883
1280*1024:
Frames: 4055, Time (ms): 60000, Min: 51, Max: 88, Avg: 67.583

2.Newcastle
1024*768:
Frames: 4293, Time (ms): 60000, Min: 47, Max: 99, Avg: 71.550
4293, 60000, 47, 99, 71.550
1280*1024:
Frames: 4029, Time (ms): 60000, Min: 46, Max: 87, Avg: 67.150

3.Winchester
1024*768:
Frames: 4396, Time (ms): 60000, Min: 57, Max: 99, Avg: 73.267
1280*1024:
Frames: 3969, Time (ms): 60000, Min: 52, Max: 85, Avg: 66.150

4. Venice:
1024*768:
Frames: 4506, Time (ms): 60000, Min: 58, Max: 91, Avg: 75.100
1280*1024:
Frames: 4160, Time (ms): 60000, Min: 49, Max: 95, Avg: 69.333

5. Manchester:
1024*768:
Frames: 4739, Time (ms): 60000, Min: 61, Max: 100, Avg: 77.983
1280*1024:
Frames: 4145, Time (ms): 60000, Min: 54, Max:96, Avg: 69.083
4145, 60000, 14, 96, 69.083

Again we see that Venice and Manchester manage to distance themselves at 1024*768 with 8%, respectively 10% better averages. At 1280*1024 all are within 2% so they perform equal. With GTS 250 you would get 100FPS with a stronger CPU - with Venice 3800 I got 82 average FSP at 1024*768. The game is single threaded so all you need is a high clock Athlon - I estimate that an Athlon 64 4000 would get you around 85-86FPS and a FX-57 around 100 based on performance scaling of all Athlon's that I tested.

The third game is Splinter Cell Chaos Theory. It is the first game from 2005 under testing and the first time I used it in the benchmarks. I used HardwareOC Splinter Cell Chaos Theory Benchmarking Tool and tested the first demo: Lighthouse.
1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

nd22 wrote on 2026-08-09, 11:59:

I used HardwareOC Splinter Cell Chaos Theory Benchmarking Tool and tested the first demo: Lighthouse.

Out of curiosity, why not use the built-in benchmark from Chaos Theory? It shows minimum, maximum and average frame rates, which I find much more useful. The file you want is Timedemo.bat located in the game's System subfolder. After running that, the result is recorded in the TimeDemoResult.xls file.

I wonder if CPUs with more cache would have better minimum FPS values.

Newcastle suffer a serious setback, Clawhammer and Winchester are tied, while Venice and Manchester are in front. Please note the we are completely CPU limited, with Venice 3800 the scores improves but still the game needs something along the line of Athlon 64 4000/FX-55 to balance the GTS 250.
Thank you Joseph I did not even knew that Chaos theory has an integrated benchmark - I never played the game because I do not play third person shooters, only FPS. I also used the HardwareOC Splinter Cell Chaos Theory Benchmarking Tool because with that one you get a HTML file and I can take a screenshot with the system details. With that Excel is like testing Half life 2 - no screenshots. I will try that to see what scores I get with that demo in the next chapter.

After Splinter cell which is considered one of the best stealth games ever made we advance to Serious Sam 2 - a shooter that does not require any intellectual effort but provides tons of fun after a hard day at work. Also from 2005, also tested at maximum settings:
1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

In this game the newer revisions of Athlon 64's do not provide an improved experience, Winchester is the one who manages to stay on top but - from the performance point of view - they are all the same.

I noticed we have about 11-14% performance drop over Clawhammer 3400+ in s754 with Radeon X1950 Pro on nForce4. GeForce GTS 250 is supposed to be much faster than Radeon X1950 Pro. Probably half of L2 cache and Command Rate 2T is responsible.

Re: Battle of the platforms: socket 754!

HALF LIFE 2 : 1280*1024
Clawhammer 3400+ (s754)
Frames: 4739 - Time: 60000ms - Avg: 78.983 - Min: 59 - Max: 110

Clawhammer 3500+ (s939)
Frames: 4055, Time (ms): 60000, Min: 51, Max: 88, Avg: 67.583

Far Cry: 1280*1024
Clawhammer 3400+ (s754)
Average Score: 89.57 FPS

Clawhammer 3500+ (s939)
Average Score: 79.68 FPS

I saw it too when I put the results in Excel. Despite other memory timings making no difference whatsoever, command rate has an extraordinary influence on the results - performance drops 10%!! I think that it is imperative to try to set CR to 1T!
In the next chapter I will use corsair 3500LL pro memory to see what it is possible with very high end memory.
This kind of performance drop is also visible if I test with RAM set at 333 MHz - so it is out of sync with the CPU.

Corsair 3500LL pro should do much better. With 2x 1GB it can do:

5-2-2-2 2.8V 200MHz 1T
6-2-3-2 2.7V 230MHz 1T
6-2-3-2.5 2.7V 250MHz 1T

Basically it can run with CL2 at DDR433 or CL2.5 at DDR500.

It utilizes Infineon chips, later known as Qimonda (since 2006, after spin-off).

I doubt you will be able to use Command Rate 1T with 4 of them, you will have to use 3GB RAM with very fast 1GB and 512MB modules.

Next game is Fear - one of the best first person shooters ever made. You really need a good graphics card and an audio card to experience this game in all its glory!

1024*768 first:

1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

1280*1024 second:
1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

Performance is split according to manufacturing process: older 130nm parts on one side and newer 90nm parts leading by ~5-6%!
You really need a stronger CPU, Athlon 64 4000 would get you around 10-12% better performance and would constitute a good balance with the GTS 250.

We reached the final game: Quake 4. This game is exceptionally tough on the video card (and all games built on the IDTECH4 ENGINE) so I chose only 8X AA (the rest of the settings turned all the way to 11) but even under this conditions we are GPU-limited. GTX 280/285 or better needed to break past the video card limitation.

1. Clawhammer
2. Newcastle
3. Winchester
4. Venice
5. Manchester

All scores are the same, there is no difference!
You really need a dual core for this one and also a GTX 285 if you can get one (I got one GTX 280 that I can not use because it stays at ~80 degrees under load all the time, I hope I can get a 285 to solve the temperature problem). Frame rate improves greatly with a dual core CPU, even a slower clocked one, than with a high clocked single core CPU.

You may be able to get temperatures down on GTX 280 by adjusting fan curve in MSI Afterburner or in BIOS. I always update the fan curve in BIOS in all NVidia GPUs up to GTX 580 - using standard curve, targeting about 60-65'C while gaming and about 72'C in Sanctuary Benchmark (most heat production, but not as dangerous as Furmark). For simplicity you could even set fan speed to 100% for benchmarking. Assuming thermal paste is in good state.

I have one GTX 285, but it crashes the driver about every 20-30 minutes and resumes after that. Good enough for benchmarking but not for real usage. Perhaps the GTX 285 I have just needs recapping. Only worth getting if the price is really low, they may be faulty even if seller thinks they are ok. GTX 260/275 are easier to find.

See screenshot for the fanspeed ic settings on my GTX 285 in NiBitor and I also like to enable 2s boot message like old AGP/PCI cards used to have. Only flash BIOS in DOS, Windows will brick the GPU and recovery using PCI card will be needed.

I did use MSI afterburner to set the fan at 100% and I re-pasted it late last night with arctic silver 5. I will test it ASAP in 3dmark.

Time for the rankings! Clawhammer, as the first generation of Athlon64, is awarded a 100% score and all subsequent generations are compared to it:

1. Clawhammer: 100%
2. Newcastle: 99.33%
3. Winchester: 101.61%
4. Venice: 103.01%
5. Manchester: 103.00%

So the later Athlon 64 revisions provide a very small performance boost, insufficient to be visible during gameplay. You can see it in the benchmarks but otherwise all Athlon 64's perform the same.

The main advantage is the TDP: 67W for all versions up to 3500 which is plenty for a socket 939 system with AGP and games up to 2004. For 2005 higher clocked Athlon are a must: Venice 3800 or even better San Diego 4000 with 1Mb of L2 cache. Those have a 89W TDP but I found the Venice 3800 (the one I got) to be very easy to cool even with the stock aluminium cooler from AMD. Price for the lower models are ridiculous: I got Newcastle 3800 and Venice 3800 for free.