GPU project on BOINC

[Dave gpu]

Just saw this on primegrid message board.




Michael GoetzProject
Volunteer moderator
Project scientist


If we build it, will they come?

We have an option to start running a GPU project on BOINC which will have very short GPU tasks -- on my GTX 580, the tasks run in a bit under a minute.

This project can find primes, but the primes are too small to be eligible for the Top 5000 list. To be clear: primes found on this project will never be reported to the Top 5000 primes list. The primes will, however, show up in PrimeGrid's prime database.

This project (GFN32768) is currently running on PRPnet, but without the GPU option. These tasks are a lot slower on a CPU.

I want to know if there's sufficient interest in searching for these relatively small primes. They're about 200K digits.

Who would want to run GFN tasks that are very short but will be finding primes too small to make the Top 5000 list?
____________

[denim]

If it's PG, I'm in.

[Mumps]

If this becomes another option for older/slower GPU's, I've got the hardware for that

[Al]

If this becomes another option for older/slower GPU's, I've got the hardware for that

You sure do! So do I...it makes us try harder.

I'd love another gpu project for my older hardware.

[denim]

If it is, I have a NIB GT430 that is free to a good home.

[dcushing]

If this becomes another option for older/slower GPU's, I've got the hardware for that

Yep, I got 5 of them gtx 460's, 3 of them running on pg right now, so wouldnt matter if Id find a prime or not, just give me the damn credits!

p,s, someone sign us up for that boincstats primegrid challenge that starts 14Sep0000 UTC please?

[Al]

Done!

[dcushing]

Done!

Thank you, bunker started.
Should I start a thread for that?

[Al]

Sure, feel free!

[Dave gpu]

Update


Michael GoetzProject donor
Volunteer moderator
Project scientist


My original plan was to have a combined OCL/OCL2/OCL3 app that would automatically switch between the three transforms, as this could run on any range. I still plan on doing that, but combining the apps takes time, so we're going to start with just the OCL3 app on 15, 16, 17, 17-Low, 18, 19, and 20. It will be a long time before most of those exceed OCL3's limit, and we should have the combined app before then.

For 17-Mega, we'll run the OCL2 app.

(I'm not saying we're starting all those ranges immediately, but I am setting up the system to run all of them.)

21 will start up tomorrow, because that's when we'll exhaust n=20. I'll also turn on CPU tasks on World Record tasks. Beyond that, I'm not sure which of the others will start first, or exactly when. Normally we'd be waiting longer to do more testing, but a lot of these apps need to be online by tomorrow anyway, so they'll get a lot of testing in production. It's not ideal, but it does mean we'll have more confidence about opening up the other ranges quickly.

If there's no problems with the new apps tomorrow (which include the trickle-up deadline extension mechanism), we still need to get Iain's sign-off on the OCL2 and/or OCL3 apps, as well as Jim's input on which ranges are sufficiently sieved. Then we can decide what goes live, and when.