Depends. Of course GPUs have massive parallel FPU performance but the direction Stanford is heading on SMP right now is also quite interesting. I was getting 109k PPD on an SMP unit last week. Something about accelerated algorithms speeding up progress a lot.
Big factor is how Stanford is going to handle data decomposition. If WUs continue to be large batches of simple short instructions, GPUs will outpace CPUs by several 100 times very soon (especially AMD GPUs, those have the most short pipelines and simple ALUs of all common processors). If these simulations will get more advanced and various in use, CPUs will keep their edge. Currently CPUs are doing very inefficient work but are easier to program (more CPU WUs than GPU WU is the result). GPUs need to have specifically written code for a specific architecture and are changing rapidly. GROMACS_15 will probably not run on archs after Maxwell just as GROMACS_13 does not run on Fermi based archs. Radeons are a complete disaster. From 2009 - 2012: VLIW-5, VLIW-4, GCN.
I think eventually CPUs are going to do different simulations or it's going to be dropped if the time comes where GPGPU gets mature.
OK Guys, I've lost the 8057's. And to boot my history shows "0" points on all 2.25 core WU's. But my PPD's are there because my monthly total is over 374,000 in less than five days. Most of that has been in the last two days. I don't know that sending my Work Unit History will do any good. Please advise. Also, I'm still showing 90K PPD running 7015's, but HFM shows "0" PPD. ????????
EDIT: I was wrong! I had the CPU highlighted in Client instead of GPU. I AM still folding 8057's.
EDIT3: Any input on this would be appreciated. According to this the WU's don't even exist, but yet are completed. ??????
Edited by 4thKor - 11/5/12 at 4:18pm