I do not know of any benchmark dedicated to Call of duty 2 so i used fraps and my phone to take a fewpictures (sorry for the quality). You can clearly see CPU utilization hovers in the 70-80% with the Athlon 64 X2 4200 so the second core gets some work to do:

I also tested 3 games that are single threaded only: Splinter's cell: chaos theory; Need for speed most wanted and Battlefield 2. There are no difference in performance between the Athlon 64 3500 and Athlon 64 X2 4200.
I think that, after a few more games tested, 2005 is the very last year where you could still play anything at the highest possible details with a single core Athlon 64! It is amazing how fast single cores became obsolete! By 2006 you could not possible play most games at max details with a single core Athlon/Pentium 4.

I tested a few games on my Phenom II X4 3.2Ghz:
- Need for Speed Carbon (2006) - only utilizes about 26% of 4 CPUs, so basically just one cpu. It stutters a bit on 2.2-2.4Ghz single core, needs probably 2.6-2.8Ghz single core. You will need the strongest s939 CPUs released.
- Test Drive Unlimited (2006) - only utilizes about 35% of 4 CPUs, so it can't fully utilize 2 cores. It will not run well on 2.2ghz dual core. Highway driving is very demanding in this game. On Phenom II I see drop to about 60 fps, so you can expect about 40 fps. You will need the strongest s939 X2 CPUs released.
- Microsoft Flight Simulator X (2006) - when set to ultra high quality, it seems to run at about 15-20 fps. It doesn't seem to benefit from more than 2 cores. I would recommend at least Phenom II X4 3.6-3.7Ghz for this due to clock speed. Your best bet these days is the Steam edition, supposedly a recompiled Gold + Acceleration pack. Currently costs 4.99€.

You will have a really hard time finding 2.6Ghz X2 Opteron CPU (OSA185DAA6CD) or FX-60 (ADAFX60DAA6CD) for s939.

Need for speed carbon is the very last game in the series that I played. All later games lost their charm and never played them more than half of hour.
Test drive does not utilize the second core of the Athlon 64 x2 4200 and is also single threaded like Carbon.
FSX fully utilizes any Athlon 64 x2 including 4800. The game is not playable on a single core.
I tried the 2 fastest Athlon 64 I have: 3800 Newcastle and Venice to see if the 2.4ghz can make 2006 games that I already tested playable. It does not, reinforcing the conclusion that for 2006 games you absolutely need a dual core CPU.
PS: sorry for the mess, I did not cleaned the MX-4 off the CPU's.

The problem encountered with the Clawhammer remains with the Newcastle: with 4 double sided memory modules the DDR frequency is lowered to 333mhz and performance drops.
I think that for testing single cores the 2001-2005 period should cover early XP era very well. I will make a selection of synthetic and games benchmarks to reflect the above period and start testing.

General Guidelines for achieving DDR-400 frequency (running the RAM in sync with the CPU) with all Athlon 64 revisions.

I could not find any forum posts or any other info online that solves this problem so here it goes:

As most of you know there is a known problem with the first revisions of Athlon 64's: with 4 memory modules all you get is DDR-333 and the RAM will run out of sync with the CPU leading to a ~5% performance penalty depending on the test.
To counter this most people run only 2 dimms in their socket 939 systems equipped with an Athlon 64 revision Clawhammer, Newcastle, Winchester. Every single game and benchmark until 2004 will run happy with this amount. However upon testing 2005 games such as Quake 4, Fear and others you will encounter problems - out of memory errors because the system will system will SIMPLY need more than 2gb of RAM even with careful considerations such as disabling all background tasks (from my observations MSI afterburner is the biggest offender taking around 400mb of RAM!!).
One option is: enable the page file. This is a bad option as even the fastest SSD connected to the SATA2 port on your motherboard is by several order of magnitudes slower than your RAM. Today we got the luxury of putting enough RAM in our systems that we no longer have to wait for the hard drive/SSD to act asa virtual memory. Also using the page file will screw up your results in benchmarks.

Another option is to put 4 modules so you get more RAM and run them at DDR-333 with all penalties that will incur such a setting.

I also stumbled on this problem when trying to benchmark Athlon 64 Winchester in 2005 games so I started looking for a solution. As all info on the net shows there is none I fall back to the old axiom: RTFM (read the f*cking manual). As I was having problems with the Abit KN8 SLI I took its cousin: AN8 SLI which should be the high end option of nforce4 from Abit with micro guru chip and all the bells and whistles. Since I got several of them with the full package I took the manual and started reading. This board was released before the cheaper KN8 SLI and was available since the beginning of 2005 before any Venice/San Diego chips. The first BIOS is from February 1, 2005 meaning that the board was finalized and in production since the end of 2004. That means the board was tested with what was available in 2004: Athlon 64 Clawhammer, Newcastle, Winchester. All these have problems running 4 memory modules at DDR-400.

However on page 16 of the manual we got this table:
You can see circled in red a situation with 4 memory modules that can be run at DDR-400. You must have at least 2 single rank modules to be able to do that.

In the case of DDR we have regular (non-ECC) modules with 1gb of RAM as maximum capacity. Those are dual sided modules. In the case of DDR all 1gb modules that I have seen are dual rank also. However in the case of 512mb modules you can have either dual sides or single sides (all chips on one side of the dimm). Those dual sided ones are dual rank but those single sided ones are single rank. Since Abit published the manual with the release of AN8 SLI that got me thinking: how could engineers achieve DDR-400 with 4 modules taking into account that the newest revision of Athlon 64 available at the time this board was released was Winchester?
So I thought: since I can not tackle the problem head on I must find a way around it and here in this table engineers from Abit already found it! I started searching for some 512mb modules and found quite a lot. Since CPU-Z does not show if said modules are single rank or dual rank I used Everest 5.50 that always got me out of trouble. I selected 2 that are single side/single rank and tested them together with 2 1gb modules that are dual sided/dual rank.

Using AUTO settings in the BIOS for all memory related settings I got the following:

So I used 2 1gb dimms and 2 512mb dimms for a total of 3gb of system RAM. Many of you will think: you lost 1 GB! This is not truly 4gb because Windows XP SP3 "sees" only 3.25gb of ram if I install 4 1gb modules. I do not know of any socket 939 board that can recognize more than that with XP 32 bit installed but I might be wrong on this one.

I tested some games and never exceeded 2.5gb of RAM even with msi afterburner, fraps and some other software running in the background. Let's say that under the heaviest conditions you might reach 3gb but I highly doubt that you can exceed 3gb of RAM unless you are using virtual machines or some other specialized software.

Clawhammer (the very first Athlon 64 revision) installed and, as you can see, with 3gb of RAM in a 2*1024 + 2*512mb configuration, we get DDR-400 and the CPU is in sync with the memory:

The 3 last Athlon 64 revisions - Venice, Manchester and San Diego - work "out of the box" with 4 double sided/2 ranks memory modules but I used the same memory configuration in order to have only the CPU different and the rest of the system identical:

I know the picture is bad because I got the monitor facing the sun but i have to show this: Quake 4, second level - 1280*1024, ultra quality, advanced settings all on, AA=8X. With 16X AA even the GTS 250 is overwhelmed and the frame-rate plummets to the low 30's because the game needs more than 1gb of video memory!!!

And here is a screenshot from FEAR - 1280×1024 max details, AA 4x, soft shadows on. I don't know if the limitations is from the CPU (Athlon 64 3500 Manchester) or the GPU (GTS 250) but clearly you can not maintain a 60FPS experience in this game with this combination.
PS: in order to obtain such low RAM consumption I disabled many things including some windows services which you would not normally do!

Now that I got a pretty good idea about what games can a single core Athlon 64 run at max details it is time to start testing. The useful time frame for the last and most powerful single cores from AMD is 2001-2005. Most games released after 2005 require a dual core in order to run at maximum details. While there are still some games that could be run with a single core, manually picking individual titles would require extensive testing and, more importantly, time which I do not have. So I split the periods by years and 2005 is the last year where all titles that I tested run well at max details with an Athlon 64 3500/3800 - the highest clocked Athlon that I have. The only game from 2005 that has major difficulties is Age of empires 3 - the games from AoE series are known to require extremely powerful CPU's.

Hardware environment:
CPU: Athlon 64 3500 Clawhammer, Newcastle, Winchester, Venice, Manchester, San Diego.
RAM: 3gb in a 2*1024 + 2*512mb configuration running at 3-3-3-8 CR 2T and DDR-400 with all CPU's.
Board: Abit AN8 SLI standard (the fatality version has 2 40mm fans that have an incredible high pitch sound that is very annoying)
GPU: GTS 250 1024mb
Storage: 240gb SSD
Cooler: standard AMD cooler
ODD: some SATA from storage
PSU: Corsair TX850M
Case: cooler master haf XB evo

Software environment:
- Windows XP SP3
- nvidia platform driver version 15.23 (latest)
- nvidia geforce driver version 197.77 (latest that is compatible with every single game tested such as Farcry for example)
- realtek AC97 driver version 4.06 (latest, and yes nforce4 has only AC97 codec)
- msi afterburner 4.30
- fraps 3.4.7
- irfanview for all screenshots
- everest 5.50

nd22 wrote on 2026-07-10, 04:03:

And here is a screenshot from FEAR - 1280×1024 max details, AA 4x, soft shadows on. I don't know if the limitations is from the CPU (Athlon 64 3500 Manchester) or the GPU (GTS 250) but clearly you can not maintain a 60FPS experience in this game with this combination.
PS: in order to obtain such low RAM consumption I disabled many things including some windows services which you would not normally do!

GPU does not look to be stressed too much by the metrics... even with soft shadows on, and the CPU is hitting 100% if I'm not wrong. It would probably do better with Athlon 4000+ with double the cache and a few more MHz, or an FX single-core. FEAR is definitely a ceiling for single-core CPUs.

https://www.tomshardware.com/reviews/at ... 4-12.html
This benchmark shows how 4850 (which should be slower than GTS 250) performs on Athlon 4000+ and X2 Athlons as well. FEAR is definitely single-core threaded... and FarCry seems to be more CPU limited than FEAR... quite interesting.