SLI and CrossFire Performance
Going into this article, I hadn’t planned on working on SLI or CrossFire testing simply because I didn’t expect much if any scaling in a title that both companies had only just received for the first time. To my surprise, I was able to get both AMD’s CrossFire and NVIDIA’s SLI technologies to scale decently.
For AMD, I used the 7.10 drivers that did NOT have any profiles for UT3 or its demo. CrossFire still worked and showed improvements in game play by simply enabling it in the driver as I normally would.
NVIDIA’s driver wasn’t so nice as it required a little hackery to get running. Enable SLI and then playing the UT3 demo actually showed a slightly negative scaling — something that is very common as some CPU cycles are eaten up trying to figure out how to run SLI before it decides it can’t without a profile. After some experimenting with manually setting SLI modes, I decided to try the executable renaming trick. Knowing that Bioshock used the same Unreal Engine 3, I simply renamed the main “ut3demo.exe” file to “bioshock.exe” and viola! SLI scaling was working!
That being said, I FULLY expect both of these companies to bring improvements in multi-GPU scaling in driver releases in the coming weeks. These numbers here will give us a comparison point to see how much work they’ve actually done since getting the demo.
For these tests I ran at the highest resolution we tested each single card at only in order to see the best case scenario for multi-GPU scaling.
For AMD, I used the 7.10 drivers that did NOT have any profiles for UT3 or its demo. CrossFire still worked and showed improvements in game play by simply enabling it in the driver as I normally would.
NVIDIA’s driver wasn’t so nice as it required a little hackery to get running. Enable SLI and then playing the UT3 demo actually showed a slightly negative scaling — something that is very common as some CPU cycles are eaten up trying to figure out how to run SLI before it decides it can’t without a profile. After some experimenting with manually setting SLI modes, I decided to try the executable renaming trick. Knowing that Bioshock used the same Unreal Engine 3, I simply renamed the main “ut3demo.exe” file to “bioshock.exe” and viola! SLI scaling was working!
That being said, I FULLY expect both of these companies to bring improvements in multi-GPU scaling in driver releases in the coming weeks. These numbers here will give us a comparison point to see how much work they’ve actually done since getting the demo.
For these tests I ran at the highest resolution we tested each single card at only in order to see the best case scenario for multi-GPU scaling.
Once again the 60 FPS lock puts NVIDIA’s top card in an odd spot when we compare it to the multi-GPU scaling with AMD’s HD 2900 XT. The 8800 GTX cards only see a scaling of 14% due to the fact that the frame rate can’t go MUCH higher than what it averaged with a single card. The pair of HD 2900 XTs though see a 30% scaling gain in CrossFire mode bring the dual GPU performance on par with that of NVIDIA’s flagship cards.
Here we have the 8800 GTX and 8800 GTS 640MB cards compared and the results are much improved for NVIDIA. Since the 8800 GTS 640MB card has a lower average FPS to start with, adding in the second card shows better percentage-based improvements. The GTS cards scale about 51% compared to the AMD HD 2900 XTs that scale just over 30%.