I got a docker container up and running with Pytorch that's using my AMD GPU to accelerate training. You need to install Rocm on the host machine and I'll walk you through the steps in this video. I used Ubuntu 22.04 so the steps here are for Ubuntu 22.04
Here's a link to Rocm's 5.4 installation page: docs.amd.com/b...
link to Rocm framework installation docs.amd.com/b...
I'm reading the steps from a medium article I wrote about how to get it up and running: / how-to-get-rocm-runnin... is the link to that article.
I assume you have Docker installed on your machine for these steps to work.
steps to do
1. sudo apt-get update
2. wget repo.radeon.co...
3. sudo apt-get install ./amdgpu-install_5.4.50400-1_all.deb
4. sudo amdgpu-install --usecase=rocm
5. docker pull rocm/pytorch:latest
6. docker run --net=host -it --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --device=/dev/kfd --device=/dev/dri --group-add video --ipc=host --shm-size 8G rocm/pytorch:latest
that will get you into the container, to test that it worked you can run
1. python
2. import torch
3. torch.cuda.is_available()
there is also a test script located in lib if I remember correctly. if you haven't changed directories from running the container, you can cd apex/tests/L0 to get there and run ./run_rocm.sh to test it.
In the amd framework link above there are instructions to get a tensorflow container running as well. I tested that and got it to work. I saw they have Magma stuff to but I'm not sure how to use that so I didn't test that container. Let me know if you guys have any questions.
Affiliate links
-------------------------------------------------------------------
the 6900xt 16GB that is a pretty good deal: amzn.to/3mtQvNP
Each Amazon link is linked to my Amazon Associate's account and a small percentage of your purchase will support the channel at no extra cost to you.
22 окт 2024
