Titan Xp not matching the installed CUDA-8.0/PGI-17.4?

Hi.
I’m pretty new to OpenACC, and have tried to write a basic test to educate myself. My small package is available on the github at https://github.com/moravveji/OpenACC/tree/master/pi. If you want to build it, run this command: “make pgi”.

I have access to two GPUs:

  • K40c installed at my university cluster
  • Titan Xp card installed at my home desktop
    I always use PGI 17.4 community edition. Cuda 8.0 is also installed on both machines.

I can successfully compile and execute my code with the K40c on our cluster. However, at home, after I successfully compile the same code, I cannot execute it, since the execution leads to the following error:

Current file:     /hdd/ehsan/codes/OpenACC/pi/lib.f90
        function: pi_gregory_leibniz
        line:     55
Current region was compiled for:
  NVIDIA Tesla GPU sm30 sm35 sm30 sm35 sm50
Available accelerators:
  device[1]: Native X86
The accelerator does not match the profile for which this program was compiled

What I understand from the message is that the installed PGI/CUDA libraries do not support the Titan Xp card. But, I strongly doubt it.

Can you please enlighten me what else I must do to execute my executable file?

Regards,
Ehsan.

Hi Ehsan,

Titan XP uses compute capability 6.0, so you’ll need to compile with the flag “-ta=tesla:cc60”. Since CC60 requires CUDA 8.0, the 17.4 compilers don’t target this device by default.

I’d also double check that the PGI runtime can see the device by running the “pgaccelinfo” utility.

Hope this helps,
Mat