Investigations into Athlon X2 Overclocking

Benchmark Information

System components aren't the only changes that we've made. We eventually discovered an error in our Battlefield 2 benchmark last time and removed the scores. The problem was that frame rates were tracked while in the menu screen, skewing the results. We've corrected the benchmark and will provide complete results this time, and we also switched to version 1.03 of BF2 rather than using 1.00. (We'll be switching again to a new demo and version 1.12 soon. Wouldn't it be nice if game updates didn't break old demos?) In addition, Quake 4 and FEAR are now part of the game list. Quake 4 and Doom 3 are very similar, though we did notice that Quake 4 timedemo benchmarks actually disable some of the extra graphical effects (like some of the shadows and lighting).

Here's the list of all the benchmarks that were run, along with information on how they were run:

Benchmark Information
Winstones 2004 (v1.01)	Default settings except reboots between benchmark runs were disabled.
PCMark04/05	Default settings.
AutoGK 1.96	Encode Ch. 9 of The Sum of All Fears to 75% quality Encode a 30 second commercial to 5MB size with audio DivX version 5.2.1 and Xvid version 1.0.3
Battlefield 2 (v1.03)	High detail setting with lighting set to High as well.
Doom 3 (v1.03)	High detail setting.
Quake 4 (v1.00)	High detail setting.
Far Cry (v1.33)	Very High setting with 8xAF.
Half-Life 2	All settings at High plus Reflect World and 8xAF.
F.E.A.R. (v1.01)	High detail setting with 8xAF and no soft shadows.
3DMark03/05	Default settings.
CPU-Z (v1.30) Latency.exe	CPU cycles using 512 byte stride size with 32M data set.

Please pay attention to the scales used on the graphs. The numbers are also included for reference, and in order to avoid having all of the results overlap, the charts for the most part do not start at the 0 point. This was not done to obfuscate the results, but rather to make the charts less cluttered. A steep line slope will not indicate a significantly faster score in most cases.

Battlefield 2 Benchmark Utility

We received some requests for our BF2 benchmark utility, and since I created it, I'm going to provide it for download here. First, a quick disclaimer: Battlefield 2 benchmarking is a little odd. The built-in benchmark feature runs the demo and pops up a modal dialog at the end with the results - except that you can't see the results because BF2 is still running fullscreen. Pressing space will clear the dialog and allow the game to exit, at which time you can open the results file. The problem is that the results file shows the average frame rate skewed by the menu - the menu will often render at hundreds of frames per second! The timedemo_frametimes.csv file contains the time used for every frame rendered, though, and we know the exact number of frames in our demo file: 6362. By only using the time required to render the last 6362 frames from the CSV, we can calculate the real FPS.

The benchmark takes care of all of this for you, but it's still beta software. Sometimes BF2 will crash and the script will get stuck in a loop; in which case, you'll have to close the command prompt window (or press CTRL+C). Also, some editing of the batch file will generally be required in order to customize the options. Specifically you should set the resolutions that you want to test as well as the drive and directory where BF2 is installed. Don't select resolutions that your monitor can't support - BF2 will simply exit and the script will be stuck in a loop. SLI support also appears to be questionable, at least with the tested versions of NVIDIA's drivers and BF2 1.03.

With the disclaimers done, here's the benchmark tool - including the bf2demo and bf2cam files. Extract it to your C: drive (C:\BF2Bench) and it should work with only a few edits. If you want to extract it elsewhere, you will need to edit the batch file a bit more, but it should still work. Included are freeware versions of a few helper utilities that are required for the script to work. Sleep.exe is used to wait (without using CPU resources) for the benchmark to complete. Gawk.exe is used to calculate the actual FPS for the demo, as well as the amount of time required to load the level. (If you haven't heard of GAWK or AWK before, it is an interpreted programming language of sorts that specializes in the parsing of data files and the generating or reports.)

The repeated calls to sleep.exe may affect BF2 performance slightly (more or less depending on numerous factors), so scores should only be compared with results obtained in the same manner. Suggestions for change and comments are, of course, welcome. You may also edit and/or redistribute the script, provided that my name as well as AnandTech is not removed. If you wish to compare scores with our current and previous results, you must test with BF2 version 1.03. I have also created a new version of the script (and a new demo recording) for BF2 1.12, but results in this article are from the old version. The latest patch also made benchmarking a bit easier, so the new script doesn't have to be as complex. It still has to calculate manually the frames per second in order to avoid the impact of rendering the menu screens, and the new demo file is 8336 frames long. Enjoy!