Is there a GPU version of the app in the works?

In the works: Yes, no, well sort of
Estimated time of arrival: Later
Is it working: Yes, no, well maybe

Me too ...

:)

In the mean time if you have a 9800GT or better, GPU Grid has seemingly worked out their issues with loading down the CPU and I am seeing sub 1% loads on both systems. Production took a small hit with estimates and actual looking at tasks, ranging from 3% or less to 17% ... my own survey of 14 tasks showed a 8-9% decrease in throughput (longer run times) ... but with the CPU load reduced I am seeing a little over a 20% increase in CPU capacity ...

If you are a credit earning type where the rate is never high enough this is horrible, if you are a produce the most science for the most projects we made a significant step forward ...

On my Q9300 it was 21-22% per GPU core (usually 1) and on the i7 it was 7% per GPU core (usually 3) ... as best as I can guess they changed from an "idle" polling loop to an interrupt driven system ... that is only a guess. But I cannot account for it other wise.

The only other alternative is that they managed to find a way to have the poll loop sleep and wake faster than the cycle time of the GPU running out of work ... wait that does not make sense ... They sleep and wake just after the GPU completes a cycle ... there, that makes sense ...

The project did not say though I did ask ...

It might be easier to find out over at SaH how they did it as it is likely the same mechanism though as far as I have seen GPU Grid does not have the "spin-up" load that SaH tasks have ...

That's very difficult to implement. I think you mean something like the throttle function? That can't be added, as the GPU can only do full work, or none.

To tell BOINC not to use CUDA at all can be set through the project preferences (at Seti at least). To tell BOINC to only use CUDA (and not the CPUs) is an option in the 6.6.x range, by using the 0 option in cc_config.xml

The advice when using CUDA is to use it only at times that you aren't using the computer for anything else GPU intensive. Simple.

But you did get an answer from Eric Korpela.

There has been much discussion on the Dev mailing list of many of the situations that are going to arise, though as Richard has indicated the response has been mainly silence.

One of the issues raised was the need to restrict the use of GPUs in even a more generalized case than you describe ... such as I have two GPUs in the system and I only want one of them used ... and in that the GPUs may not be of equal capacity the selection cannot be simply use 1 because, of course, BOINC will not select the one that I want it to use ...

The problem is, I don't think, really appreciated by the developers that this new capability is a significant departure from the past and yet my impression of the comments that have been made is that the approach is that a GPU is really a CPU by another name. Yet, I can put a GTX 295, GTX 280, and a 9800 GT in to a three slot system and the GPUs have vastly different capabilities for processing work. More interesting to me is the continuing research by Richard that (on a perliminary basis at least) seems to indicate that in the case of SaH the additional capabilities of the higher end cards does not produce improved processing performance.

GPU Grid shows other wise ... where the difference between the 260/280/295 cards are up to 4 times faster than a 9800 GT ... or even better than slower cards.

Anyway, it is still early days and though it seems like forever, it really is too soon to get too worked up ... if you do have a GPU in play the BOINC Manger version 6.5.0 works reasonably well ... but you do have to have the right video drivers ...

:)