I've been away for a while. When I left some guy over on the lunatics forum was hacking together the first GPU exe, and the V8 exe had just been released.
Since then it looks like we've gotten an official gpu client which most of my questions are about.
- Are there special GPU WUs or do they run the regular CPU WUs?
- Does each GPU require a dedicated CPU core to feed it?
- I don't seem to be getting any GPU work (assuming they have special WUs). Is this common?
- Do I want to use the killVLAR optimized GPU exe?
- What's a VLAR?
Wow, you have been gone a while. I'll try to answer these to the best of my knowledge.
Are there special GPU WUs or do they run the regular CPU WUs?
Sort of. They are basically the same work, but the client assigns them to one or the other (I think). There is a program that you can use to change them back and forth in case you run out of GPU work or something.
Does each GPU require a dedicated CPU core to feed it?
Depends, I use one core to feed my 9800gx2 which is technically 2 cards. A lot of people just leave one core free to feed however many cards they have. As far as I can tell, this seems to be the best way.
I don't seem to be getting any GPU work (assuming they have special WUs). Is this common?
SETI only does GPU work on CUDA enabled cards (newer Nvidia cards only). Do you have a CUDA card? If so, when you first open BOINC manager and go to the messages tab does it say something like:
NVIDIA GPU 0: GeForce 9800 GX2 (driver version 19745, CUDA version 3000, compute capability 1.1, 512MB, 448 GFLOPS peak)
NVIDIA GPU 1: GeForce 9800 GX2 (driver version 19745, CUDA version 3000, compute capability 1.1, 512MB, 448 GFLOPS peak)
Do I want to use the killVLAR optimized GPU exe?
Yes. The vlars can take forever on a GPU, but run at normal speed on a CPU.
Its Very Low Angle Range work units. There is a thread explaining how they slow down you GPU
here.
