[2018-09-05 08:03:24.591] [scope] [critical] ERROR[CUDA driver version is insufficient for CUDA runtime version] on /home/pearson/repos/scope/comm_scope/src/memcpy/args.hpp::31 In ArgsCountGpuGpuPeerNoSelf:(cudaGetDeviceCount(&n)) FAILED
I think this is because the benchmarks are being registered first, which causes the argument functions to be applied, which calls cudaGetDeviceCount, which fails when there is no GPU.
Running
causes
I think this is because the benchmarks are being registered first, which causes the argument functions to be applied, which calls cudaGetDeviceCount, which fails when there is no GPU.