In this web content, the author shares their experience setting up and using Intel’s PyTorch extension for Arc GPUs on Ubuntu and Windows Subsystem for Linux (WSL). They mention their initial difficulties with the installation process and provide instructions for setting up the extension. The author then discusses their testing of training performance on native Ubuntu and WSL, comparing the results to their Titan RTX. They highlight that the training time on the Arc GPU is within 10% of the Titan RTX after applying a fix. The author also notes that performance on WSL is slower and recommends a bare-metal installation for optimal performance. Overall, they believe that Arc GPUs are now viable options for deep learning.
https://christianjmills.com/posts/arc-a770-testing/part-2/