Current version we're using (5.8.1) has two issues, the combination of which breaks the TPL build:
it ignores CUDA=no flag we're passing to the build when CUDA_PATH is set in the environment (which is the case if the cuda module is loaded) and still tries to compile some GPU kernels
it forcibly tries to compile code for compute_30 virtual architecture which is no longer supported by CUDA 11 compiler
Needed to compile TPLs with CUDA 11.
Current version we're using (5.8.1) has two issues, the combination of which breaks the TPL build:
CUDA=noflag we're passing to the build whenCUDA_PATHis set in the environment (which is the case if the cuda module is loaded) and still tries to compile some GPU kernelscompute_30virtual architecture which is no longer supported by CUDA 11 compiler