GPU Shootout with Unreal Tournament 2003 Part II: CPU Scaling

One week ago we put a total of 19 different GPUs through the toughest tests they’ve encountered in our labs: non-stop benchmarking using Unreal Tournament 2003.  Using a couple of levels of UT2003 we were able to get an idea for the performance levels these GPUs would enjoy when the forthcoming game is released. 

In order to save time we conducted that comparison entirely on one platform, an Athlon XP 2100+ on an EPoX KT333 motherboard.  When we queried you all about whether you’d like to see CPU scaling tests we received an astounding number of replies asking for just that. 

Today we bring you an update to our 19-way GPU shootout with a focus entirely on how these GPUs scale with CPU clock speed.

What are we testing?

The flyby demos Epic provided for us in UT2003 give a good idea of graphics performance but really don’t take into account things like physics and AI calculations, which are inevitably handled by the host CPU. 

Instead, these demos measure the speed at which the CPU can provide the GPU with vertex data, which is the first step in the rendering process.  The end result is that we’re not taking into account what a faster CPU will be able to do for physics and AI calculations, only how quickly it can feed the GPU the vertex data it needs to run at those high frame rates we desire.

How are we testing?

The first time we introduced our updated CPU scaling methodology was in our Sub-$200 Video Card Roundup, here’s a quick refresher for those of you that don’t remember:

Our goal with these tests was to isolate CPU clock speed as the only variable, not cache sizes, not architectures, not FSB speeds, only clock speed.  In order to achieve this we took an Athlon XP 2100+ and unlocked it so that we could adjust the clock multiplier.  We kept the FSB set at 133MHz and adjusted the multiplier from 6.0x, yielding an 800MHz clock speed, up to the default 13x multiplier.  Although this method produces some unofficial and unavailable processor/FSB/clock speed combinations, it serves our needs perfectly.  So although a 800MHz Athlon XP processor running on a 133MHz FSB won’t be directly comparable to a genuine Athlon running at 800MHz, you’ll at least get an idea of what a CPU around that speed would give you, performance-wise.

We stuck to one resolution (800x600) in order to use a low enough resolution to give the slower cards a chance while not making the test entirely CPU-bound.  We also ran all of our benchmarks under the DM-Asbestos map, which is entirely indoors and is noticeably less GPU bound than some of the other maps.

Finally, we ran all of the tests at the same two detail settings from our initial roundup: High and Medium.  The High Detail settings have every detail option maxed out while the Medium Detail settings disable detailed textures, lower the texture detail level, and lower the model detail level among other image quality/performance tradeoffs.  If you haven’t already, be sure to read the original UT2003 performance roundup to get a better feel for the tests we ran.

The test system remained the same from the original comparison.