If all you’re doing is building the binary, then there’s no need to have any GPU installed. PGI ships all necessary components required to create the binary to target the P100.
When you compile, be sure to add the flag “-ta=tesla:cc60” so the compiler knows that you’re wanting to target a P100 or other Pascal based NVIDIA device.
Assuming you are using an x86 based CPU, you may also need to set the target architecture flag (-ta) to the architecture of the system where you will be running. For a full list of targets, please see the output from “-help -ta”. You can target any generic x86 CPU via the “-tp=px” option but may loose some performance since the compiler wont be taking advantage of architecture specific optimizations.
Hope this helps,