The general experience, *also* under Windows, is that HCC-GPU on some, not all NVIDIA cards, is not able to run without affecting the user when at computer. Default setting for GPU computing applies: Only GPU compute when idle! NVIDIA is just *not* excelling at OpenCL applications such as what HCC-GPU has been coded in.

P.S. I've implemented SNA forcing on my 2670QM, force user apps to run the On-intel CPU graphics, so that BOINC may use the ATI card for crunching.

edit: Think GT320 is in that "not powerful enough" class to compute HCC-GPU while using the computer. Have had frequent driver crashes [W7 and W8 recover well], so set to only idle GPU compute on NVidia GPU.

edit2: Not shouting or screaming, Bolded in red, the important needing clear GPU crunching awareness for HCC-GPU! When electing to run HCC-GPU when system is idle, you'll never notice ;>) Recommend a delay setting of 5-10 minutes [about half time of what a GPU task takes, before resuming, as restarted tasks will only pick up crunch time when beyond the 50% midpoint, when a checkpoint is saved. .

The beauty about setting GPU to crunch when computer is not in use is that when no one at the keyboard I have 3 CPU and 1 GPU task active, that is 4 in total. When I want to play World of Tanks and sit at key board I have 4 CPU task active, that again is a total of 4 tasks so nothing is lost as far as CPU time goes, only points total is lower!

With respect to GPU WU crunching, it has been my experience that any NVIDIA card below 570GTX will have material video issues when you want to use the computer. 580GTX should run fine but video streaming or heavy video usage usually may result in video stuttering issues. Doubling (SLI mode) the cards will help video performance but expect some residual lagging issues from time to time.

Did not bookmark in my webbrowser [corrected now], but did post the link in the Linux Adept Confez thread where I discussed originally: http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=397887

Have beautiful, snappy graphics in Ubu 12.10 Unity.

In turn, part and parcel of the reason for that, I believe, is that Nvidia wants to steer all GPU-computing towards its exclusive, closed, and proprietary CUDA. Paying lip-service to OpenCL if Nvidia can get away with it, which Nvidia can't, or treat OpenCL as a 2nd-class citizen, which Nvidia does.
;
; andzgridPost#746
;