I am using the following system specs
GPU: 2080 ti
Nvidia Driver: 418.67
Cuda Version: 10.1
OS: Ubuntu 18.04
The problem I am having is when attempting to run a RNN using pytorch I receive the error stated in the title. This error occurs when I attempt to send the encoder to the GPU with .to(device). (No reduction in size prevents this error)
When installing the Nvidia software I installed cuDNN 7.6 but when installing pytorch it installed 7.4.
I cannot seem to find the right configuration to allow my RNN to run efficiently with the use of cudNN. I have worked around it for the time being by disabling cudnn from the pytorch command but my performance diminshes drastically.
Is it possible to use pytorch with cuda 10.1?
Do I need to downgrade to CUDA 10.0 with cuDNN 7.5 to operate pytorch 1.0?