That is allot of crunch just for 1 wu and I have
dual 580 GTX video cards and it's taken some time
just for running 1 on both GPUs.
Does anyone know about this?

---

Runs much more efficiently on AMD cards.
If you just want to rack up numbers may I suggest Primegrid PPS?

Alternatively GPUgrid for bio-med.

---

Team Renegades
Forum

That is allot of crunch just for 1 wu and I have
dual 580 GTX video cards and it's taken some time
just for running 1 on both GPUs.

That's how it's supposed to work. Having dual cards means a task should get crunched in half the time it would take if you'd only have one card.

My triple GTX 285 host is taking about 20-25 mins to crunch through one normal size task (you are crunching huge sized tasks). Your host might benefit from setting core used to 3 in your Project Preferences (at http://moowrap.net/prefs.php?subset=project) if you want to optimize throughput. However, that might affect negatively normal usage of the host.

-w

Your host might benefit from setting core used to 3 in your Project

Oops, no, for nVidia cards cores 9, 10 or 11 would be optimum. You should try to experiment with each of values and see what gives you fastest crunch time. For me it was 11 for that triple-card host (it's a dedicated cruncher so normal daily usage doesn't matter on it).

-w

I have all WUs that run on GPUs that are out there but Moo! Wrapper is the only
one I see that uses 1 WU that haves to be veary big to use both video cars like in SLI mode to crunch.

---

I have all WUs that run on GPUs that are out there but Moo! Wrapper is the only
one I see that uses 1 WU that haves to be veary big to use both video cars like in SLI mode to crunch.

It doesnt *have* to use both. Its a multi-threaded application that works on GPUs, its set to use whatever GPUs are available, the more GPUs uses the faster the WU completes.

As a generalisation, two of the same GPUs will, potentially, run the WU twice as fast as one GPU - end result is same throughput (as such - there are other issues to do with scaling up of cards on SLI/Crossfire, but thats another issue separate from the principle of multi threading). DNETC also do multi threaded GPU applications, and its likely over time that others will follow - a bit like when GPUs first started in BOINC, slow to get going, then as time went on more and more GPU applications appeared over a couple of years timespan.

Its the same with CPUs, multi-thread applications are on the increase that use all available cores on a CPU. Aqua is notable for that - and there it is a case of massive WUs running, try running one of their WUs on a single core, and it will take days, likely weeks. So far, however with the GPU multi-thread apps out there, it just uses whats available, although I suspect GPU WUs will increase over time because of it.

Regards
Zy

Hi,

Scary thought

BBC climate models used to take 6 months on a faster machine and up to a year on slower machines

ah the good old days lol


Ian

---

----> Please Join team Scotland HERE