we looked at earlier to run the Folding@Home software, you’ll get 20 to 40 times the output compared to a CPU. That’s a mighty bold claim, one that Techreport discovered wasn’t quite as it seemed.
Although Stanford claims the beta GPU client runs 20 and 40 times faster on new Radeons, and although Radeons have many gigaflops of processing power, the current client doesn’t quite report this tremendous workload. In the test, the GPU client had 2640 work points completed whereas the CPU client only had 899. That’s far removed from the 20~40x reported. What gives?
Well actually, the discrepancy comes from the fact taht points and work units are on a different scale depending on whether you’re using CPUs or GPUs. The GPU client is actually doing 20 to 40 times more work, but because the benchmark is set so much higher, the result is that the points are, give or take, not all that much different. So since the scoring scheme doesn’t reflect the bigger workload done, don’t expect to show off your Folding@Home scores to your nerd buddies and brag about how much more cancer research you’re doing. Nevertheless, the GPU is doing an order of magnitude more processing than the CPU.
A closer look at Folding@home on the GPU [TechReport]