So I've been too busy getting systems actually working to participate in any kind of benchmarking (or do any gaming - i seem to spend 4x time fixing and 1x time playing). Recently put a CrossfireX system in XP after Windows 7 performance was disappointing to say the least. Running 3DMark03 has helped me to a) confirm my setup is working and b) know where it makes a difference and why you don't get 2x performance boost (which I'd read about beforehand). I also played with overclocking the somewhat modest CPU (but the only AM3 one I have). I don't know how much (if any) the CPU is holding back the dual GPU setup.
CPU: Athlon II X2 250 (3Ghz stock speed)
RAM: 4GB something-or-other
Mobo: Gigabyte GA-MA790XT-UD4P (rev. 1.0)
OS: XP Pro x64 Edition
GPU: 2x ATi Radeon HD 4850 512MB (Sapphire)
The CPU OCed to 3.8GHz (FSB increase from 200MHz to 255MHz) without breaking a sweat before becoming unstable. I didn't bother tweaking voltages as I guessed it was the boring RAM that was holding it back. But I was fairly impressed with the 25%+ clock improvement. I've got screenshots but I thought graphs would be more useful:
The attachment Overall Score.PNG is no longer available
So overclocking by 25% with 1x GPU yields a 4% improvement. With 2x GPU we get a 12% improvement, which is much better.
The attachment CPU Scores.PNG is no longer available
As I can only upload 5 attachments, I have not included the overall CPU scores chart but no need as it's simply 1526 at stock speed and 1852 OC. That's a 21% increase, which is almost in line. This is reflected in the individual test results.
The attachment Game Scores.PNG is no longer available
Looking at the game results we can see which ones are more affected by the CPU and which ones are GPU dependent. There is almost no improvement in Wings of Fury (GT1) from having the 2nd GPU. It looks like this is down to the CPU limiting the bandwidth, as overclocking does improve the score by 21% so there is definitely room for improvement. Battle of Proxycon (GT2) and Trolls Lair (GT3) are significantly improved by CrossfireX, 68% and and 61% respectively at stock speeds, 78% in both cases with overclock. There is only a 26% improvement with the 2nd GPU in Mother Nature (GT4) at stock speeds but a 46% improvement with OC.
The attachment Fill Rates.PNG is no longer available
Fill rates is where it's clear what effect CrossfireX has - an obvious doubling of the statistics by adding a second card - but nothing takes advantage 100% of this feature so that's why you don't see double improvement across the board.
The attachment Pixel Shader.PNG is no longer available
Finally, the Pixel Shader 2.0 scores are clearly GPU-dependent, improving by 90%.
So I would suggest that 3DMark03 is a great tool for measuring whether something is working - it is known that it's not a completely rounded measure of performance but it's pretty good imho as it covers many bases. 3DMark2001 confused the hell out of me because overclocking took the results from about 22k to 40k, while adding the second GPU made almost no difference at all.
Now all I need is a Phenom II X6! Now where's that £120 I left lying around...
Edit: crap. Forgot to keep the orb link. And I'll update what the actual RAM is later.
Check out my blog and YouTube channel for thoughts, articles, system profiles, and tips.