RTFA: http://www.extremetech.com/article2/0,1697,2284065…

Dr. Pande: The CPU clients do a somewhat different calculation (”explicit solvent”) vs. the GPU/PS3 (which do “implicit solvent”). All of this depends on how we deal with water. Do we deal with water as individual molecules (”explicit”) or as a mathematical continuum (”implicit”)? Both have various pros and cons. Implicit solvent maps better to the PS3 & GPUs (at least with today’s hardware).

ET: Obviously everyone is interested in performance. Understanding that the client is still in beta, how is performance shaking out so far, and how does it compare to other popular clients like the SMP client or PS3 (in terms of gigaflops per client, that is)?

Dr. Pande: It’s tricky to compare clients in overall performance (since the GPU and SMP clients are running different calculations), but we are very happy with the GPU2 performance so far. We haven’t done extensive benchmarking vs. the PS3 (which is closer), but qualitatively, the GPU performance is greater than the PS3 on most new ATI GPUs.

ET: That leaves a lot of wiggle room. Is it 5% faster? 2x faster? 10x faster? I know you can’t be exact, but can you give me a ballpark?

Dr. Pande: The problem is that there’s a large range of GPU’s out there (ATI 2400 to 3870) and even their relative speed to each other depends on the Work Unit, so it’s hard to give a single number. However, this isn’t a minor speed up at the fastest GPUs.

Neat - using a 3d card to perform protein folding. Not only can you use an implicit solution, but modern desktop computers will be able to perform this bio-work faster by utilizing different types of processors at the same time. Isn’t it crazy that 50 years ago, people were still working out the clothes washing/folding@home problem?