This is the error you get when building a binary targeting a device that’s older than the installed device. For example building for a Kepler (cc35) but running on a Maxwell (cc50).
When building with PGI 15.7 on a Maxwell system similar to yours, I can only replicate the error if I compile targeting cc35. Compiling to cc50 works fine. Can you please double check that you do compile with cc50?
% pgcc -V
pgcc 15.7-0 64-bit target on x86-64 Linux -tp k8
The Portland Group - PGI Compilers and Tools
Copyright (c) 2015, NVIDIA CORPORATION. All rights reserved.
% pgcc -acc -fast -ta=tesla:cc35 add.c -o add ; add
call to cuModuleLoadData returned error 209: No binary for GPU
% pgcc -acc -fast -ta=tesla:cc50 add.c -o add ; add
11
13
15
17
19
% pgcc -acc -fast -ta=tesla add.c -o add ; add
11
13
15
17
19