1

I am trying to run ./TTS/bin/train_tacotron.py with GPU in Powershell.

I followed these instructions, which got me pretty far: the config is read, the model restored, but as training is about to start, I get the message:

UserWarning: NVIDIA GeForce RTX 3060 Ti with CUDA capability sm_86 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37. If you want to use the NVIDIA GeForce RTX 3060 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

The instructions specified don't really help. I tried installing the most recent stable version of PyTorch, as well as trying 1.7.1 (as opposed to 1.8.0 as recommended in the instructions I linked), but I got the same message.

How can I get this to run on my GPU?

Side note: I was successfully able to run training on my GPU in WSL, but it froze after a few hundred epochs, so I wanted to try Powershell to see if it made a difference.

quavaro
  • 13
  • 3

1 Answers1

0

In order to work properly with your current CUDA version, you need to specify the version 11.3 to cudatoolkit. Execute the following commands:

conda uninstall cudatoolkit
conda install cudatoolkit=11.3 -c pytorch
mike
  • 16
  • 2
  • 4