PyTorch GPU issues for CUDA runtime v8 and CUDA driver API v10


I am having issues running any code on the gpu, namely first call to cuda() takes on the order of minutes and I get CUDNN_STATUS_EXECUTION_FAILED error when running a forward pass. I am using cuda runtime v8 (I need this to be v8) but cuda driver api v10, and I installed pytorch from with pip, choosing cuda 8. If I switch to using both cuda driver api 10 and cuda runtime v10, everything runs fine. The GPU I’m using is a RTX 2080. Is this a pytorch compatibility issue given the different runtime vs driver api versions for cuda?

Thank you!