CUDA Toolkit and CC versions?

I see under the “Target Accelerators” when I select the NVIDIA option the ability to select both the toolkit and CC versions. Appears 12.5 doesn’t yet include the 4.2 toolkit version. I’ve left both of these on auto/default, assume that is fine. Also I had the CUDA toolkits installed separately and there are Windows environmental variables from those installations. I assume PVF doesn’t get confused by these separate installations.

Final question: I have two GPUs installed, a Fermi (GTX 580) and a Kepler (GTX 680). I know PGI doesn’t yet support the Kepler GPUs but will the compiler pick up the Fermi when it builds and then when I execute?

Appears 12.5 doesn’t yet include the 4.2 toolkit version.

Correct, CUDA 4.2 will be added to the 12.6 release.

Final question: I have two GPUs installed, a Fermi (GTX 580) and a Kepler (GTX 680). I know PGI doesn’t yet support the Kepler GPUs but will the compiler pick up the Fermi when it builds and then when I execute?

There were some problems with earlier release but 12.5 should be fine. Granted I personally only test on Linux so there may be Windows issues I’m not aware of.

  • Mat

Thanks Mat, seems to be starting execution with 12.5 while failing using 12.4. But there are problems with the execution that I’ll have to identify and will post separately.

-ralph