I'd like to see user defined GPU usage in multiple GPU systems. For example, I could be able to allow a secondary card to crunch while the primary was free.

In this manner, more flops get devoted because there is never a need to disable BOINC for using the system on the primary graphics card. Furthermore, it allows for different cards to be used in each slot whilst disabling SLI, expanding options for users (i.e. can throw in a 9800GT into the secondary slot and leave my GTX 280 in the primary for gaming).

I think with this feature, we'd see overall gain in time spent towards the project.

I'd also love to see this happen, if it's possible to do with BOINC/the clients/the CUDA libraries?

In addition to this feature, it would be nice to be able to schedule down time as well for the CUDA WUs. I have had some issues with overheating on my video card and have decided to turn it off from crunching CUDA WUs on one of my systems. I am still testing this to see if I can find a balance but if BOINC had some sort of periodic scheduler, I would be able to use the GPU more often than not. For example, allow to run for 30 minutes, 10 minute cooldown, run again 30 minutes, 10 minute cooldown, etc...

Of course I would imagine people would want the ability to customize all this; for me something that would allow me to define a periodic cooldown would be sufficient.

Thanks for the consideration...

---

I would guess that repeated cycling of the GPU temperature would be worse than letting it run at a higher temperature constantly. Better to take manual control of the GPU fans to keep the temperature down, I think.

---

--
Classic 82353 WU / 400979 h