I am using a nvidia 550-TI (Fermi) and a nvidia 9500 GT (Non-Fermi) cards for Cuda GPU processing.

I am processing MB WU's using Lunatics 0.38 64-bit code on a win-7 system.

Is the some app_info.xml magic to make the Fermi card run 2 WU's at a time while the non-Fermi card runs only 1 WU??

If so ....what is it? (no, I'm not ready to run 2 copies of BOINC on the same computer ... yet).

Thanks

Ed F

IIRC there are several ways to get both GPU's running. There are several threads that explain this very well. I think the simplest is to run 2 instances of the BOINC and have an app_info set for each.

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Yes, I can run 2 WU's on BOTH cards but the slower get a lot of -177 errors ... not quite a good solution (unless "they" fixed this in the last outage).

Ed F

Unless something new has come along in the last 5 months the answer is no, there is no way to do this.

I used to run a 9800GT with a 460 GTX and had to run them both with just 1 at a time until I upgraded the 9800 GT to another fermi card.

There is one way as described at Vyper's Blog.

If its worth to change the configuration, it is on your decision; but it is possible.

---

- Performance is not a simple linear function of the number of CPUs you throw at the problem. -

Yes, I can run 2 WU's on BOTH cards but the slower get a lot of -177 errors ... not quite a good solution (unless "they" fixed this in the last outage).

Ed F

Using BOINC Rescheduler to avoid the -177 errors would make that configuration practical, and the longer runtimes on the 9500 might make the combined APR come down enough so you wouldn't need to use Rescheduler after awhile.

With other things affecting work availability this is probably not a good time for testing, though.

                                                                   Joe

Yes, I can run 2 WU's on BOTH cards but the slower get a lot of -177 errors ... not quite a good solution (unless "they" fixed this in the last outage).

Ed F

Using BOINC Rescheduler to avoid the -177 errors would make that configuration practical, and the longer runtimes on the 9500 might make the combined APR come down enough so you wouldn't need to use Rescheduler after awhile.

With other things affecting work availability this is probably not a good time for testing, though.

                                                                   Joe

Then when will it ever be a good time? Two types of nvidia gpus and mixing them has results of either -177 errors being made from running too many wu's at one time on the 200 series or older gpus or not efficient use of the 400 or newer series gpus...

One rather obvious solution is to NOT mix different generation GPUs in the same computer (8{).

But I know some folks can't stand to see good computing hardware sitting idle, and can't afford a 2nd (or 3rd) cruncher.

---

Donald
Infernal Optimist / Submariner, retired

But I know some folks can't stand to see good computing hardware sitting idle, and can't afford a 2nd (or 3rd) cruncher.

Or don't have the proper combination of slots available (like me).

Ed F