Having struggled to get up and running with GPUGRID, I'm now hitting another problem. I understand why this limitation has been put in place but 10 WU's per 24 hours is not enough. It was my intention to configure 3 machines to fold GPUGRID. The first has 6x GPU's (3x GTX295), the second and third have 8x GPU's (4x GTX295) each. I'm looking at around 12 hours to completion for each WU, so on the 8x GPU machines I will crunch 16 in a 24 hour period. Please let me know if this restriction can be eased, otherwise I'll start looking at other projects to donate the 'spare' resource to.

If I remember correctly, the restriction is eased automatically once the server software sees you return the first few workunits fast enough. Give it a day or two to see if this happens. You may need to adjust your settings so that it can ask for more workunits shortly after it returns the first one, though.

However, if you want an alternate BOINC project that can use the GPU, I believe that this one is the only one active at present:

http://www.rechenkraft.net/yoyo/

Still in beta test, if that matters for you.

There are rumors that SETI@home will have such an application available soon, but it's been some time since I checked.

You can probably give preference to GPUGRID by simply assigning any other such projects a rather low fraction of the CPU time they can use; let us know if this works.

My system with 2 GTX 295 cards, giving me 4 GPUs has a limit of 15 per day PER CPU ... with the i7 running things should not be a sweat. ANd if the text is wrong and the scheduler counts the GPUs it is still no sweat as the 295 cards take about 6 hours per task (about 15-18 tasks completed per day for the whole system.

And as robert suggested, there are other GPU projects out there:

SaH
Collatz
MW
The Lattice Project
Aqua
Einstein (early beta)

Actually there are now almost so many that I have lost track ...

MW and Collatz are kind of special as they have applications for ATI cards if you happen to have the right ATI cards ...

Thanks guys. I had a quick look at SaH and Aqua yesterday. Neither appeared to have GPU units available. I'll do a little research on the others.

Thanks guys. I had a quick look at SaH and Aqua yesterday. Neither appeared to have GPU units available. I'll do a little research on the others.

SETI tasks, like MW tasks are no different. To put it another way, you configure your system to run tasks for SETI on the GPU vice the CPU and off you go.

AT THE MOMENT, for most projects you have to fiddle with settings and maybe install anon platform applications to get the GPU applications to run. We are slowly moving to daylight wherein the project will download the right applications and you don't have to do that...

The problem is that most of this is very project specific and BOINC version specific. For GPU work the only versions of BOINC I recommend are 6.5.0 and now 6.10.3 and even those may not work for you ...

SETI tasks, like MW tasks are no different. To put it another way, you configure your system to run tasks for SETI on the GPU vice the CPU and off you go.

Even on Linux. (Did I mention Linux! ;)) Are there SETI cuda tasks available for any platform other than Windows?

The problem is that most of this is very project specific and BOINC version specific. For GPU work the only versions of BOINC I recommend are 6.5.0 and now 6.10.3 and even those may not work for you ...

Yes, I had my very own little introductory learning experience with boinc versions yesterday. Is there any version of the app that does not fix an issue, only to introduce 2 more? GPU scheduling is totally borked in 6.10.3/6.10.4 on Linux. I tried that after the Fedora 6.6.37 package, which has borked GPU scheduling - all processes run on device 0. I settled on the 6_6a code branch (6.6.39), a couple of other patches from svn and a patch of my own concoction. All seems to be working quite well.

Um, don't know ... and it is not an issue with the tasks, but on the availability of applications to run the tasks. So you could run the tasks on linux and match up with me running the same task on windows. So the key is not the tasks, but on the availability of applications for the various platforms.

My take on BOINC development is pretty sour mostly for the same reasons you note. Each new version seems to have significant new issues with out always resloving old issues. There are versions that are at one level or another relatively free of significant issues, but, this is also platform dependent.

6.10.5 is supposed to fix the issues from 6.10.3 and .4 though I have not yet had the courage to try it yet. I am running one system fine on 6.10.3 where I have ATI running MW tasks and I am loathe to fiddle with it. For the rest of my systems I am running a few other versions though I think the whole 6.x.y series has had serious problems for a long time.

Sadly, the developers are not real interested in addressing participant concerns, or at least don't appear to be that interested in addressing them ...

Part of the problem is that UCB has lost sight of what BOINC was intended to do in the first place and it has become a play-toy ...