The original calling command is: pgfortran -Mcuda=cuda10.0,cc70,ptxinfo,maxregcount:32 name.cuf. After changing it to a new card (GPU_Nvidia_Tesla-T4_16G), what should the command be changed to? Does the code need further adjustments?
A T4 uses the Turing architecture which uses compute capability 7.5, so you’d need to change “cc70” to “cc75”.
For the CUDA version, you’ll want to update this to be the same as the CUDA driver version installed on your system.
thanks，Does the code need further adjustments?
I wouldn’t think so but I don’t know your code so can’t offer any advice.
If CUDA10.0 and PGI18 are installed, can I call cc75?
Support for cc75 was added in the PGI 19.1 compilers. You may consider updating your compilers to using the recently released NVHPC SDK 21.3 (PGI was rebranded to be the NVHPC Compiler). However, CUDA 10.0 is no longer supported in 21.3, so you would need to update your CUDA driver to at least 10.2, but preferably 11.2.