Missing GPU Binary

Hi,

I have recently got a GTX 970 and wanted to test out some fortran code on windows. It compiles fine, but I get error messages when trying to run –

$ pgf90 ModInvACC.f90 -acc -Minfo=accel -ta=nvidia,cuda6.5
NOTE: your trial license will expire in 13 days, 6.33 hours.
NOTE: your trial license will expire in 13 days, 6.33 hours.
modinverse:
     13, Accelerator kernel generated
         13, Sum reduction generated for sum
         16, !$acc loop gang ! blockidx%x
     13, Generating Tesla code
test:
     29, Generating acc routine vector
         Generating Tesla code

byo@byo7 ~/Euler/451
$ ./ModInvACC.exe
call to cuModuleLoadData returned error 209: No binary for GPU

I was getting this same error message when trying CUDA samples until I installed the special CUDA sdk for newer cards. https://developer.nvidia.com/cuda-downloads-geforce-gtx9xx

I now have the situation where I can run the CUDA samples compiled with nvcc fine, but the pgf90 situation remains the same. Any help would be greatly appreciated.

Hello,

The explanation is very simple.

PGI compilers do not support Maxwell GPUs, and we have no
schedule of when we will or news of if we will.

The devices with compute capacity of 1.0 to 3.5 are the devices we
support. The CC 5.0 and above are not supported in our compilers.


dave

Hi,

Thanks for the quick response.

Peter