Running 6.5.0 and 180.84 on a 9800gtx+ and have the following observations

gpugrid on hold and beta seti active - I had to dedicate a CPU to beta seti else a 20 minute task took 12 hours. I ran thru about 25 small beta seti gpu wu's in a few hours. This was after I set the number of processors to less than 100%. Once the 25 or so finished, I started seeing the messages "cant finish in time". I then switched back to GPUGRID.

I have only a single 9800gtx+ and switch manually from gpugrid to beta seti.

beta seti on hold and gpugrid active: I was able to switch back to using all cpu's (100%) which gave me the 4+1 cpu advantage. Wall clock ET seems to be the same with 4 or 4+1 cpu so I am sticking with 4+1.

Maybe the gpu CUDA task signals it needs more data while the beta seti expects a cpu poll to feed it? I am just guessing.

Possibly I could have reset the seti beta gpu project to get rid of that "cant finish in time" message. The seti people do not provide time per step and elapsed time like this project does in stderr_txt

my 2c.

I tried GPU-Seti today on a 8600GT and it's remarkeable that there was absolutely no lag. If I assume they also use gpu polling that probably means they have much smaller "steps" (not time steps in the case of seti, but some kind of small work packages) .. which means they'd need to poll the GPU much more often and would thus suffer a much greater performance penalty for not having a dedicated CPU.

These are assumptions, though. I did not yet have to time to take a closer look at Seti@GPU.

MrS

---

Scanning for our furry friends since Jan 2002