awslabs / amazon-eks-ami

Packer configuration for building a custom EKS AMI
https://awslabs.github.io/amazon-eks-ami/
MIT No Attribution
2.41k stars 1.13k forks source link

Radeon Drivers #1364

Open aldmbmtl opened 1 year ago

aldmbmtl commented 1 year ago

What would you like to be added: Radeon Drivers

Why is this needed:

We currently are looking to use AMD GPU's for some workloads and we were wondering if that is planned at some point. We currently use containerized workstations and stream them to the desktop. The pricing on the g4ad's are so much better than the nvidia's and we would love to use the AMD's for some workstations.

aldmbmtl commented 1 year ago

I forked this repository and added support on my own, but I wish I could just use your managed template instead of having to create my own AMI and then specifying a launch template.

https://github.com/awslabs/amazon-eks-ami/compare/master...aldmbmtl:amazon-eks-ami:master

Also, these images are fantastic! Thank you so much for your hard work!

cartermckinnon commented 1 year ago

Thanks for sharing your patch! We're planning some work on the GPU AMI in the near future; I'll bring this up with the team. 👍

aldmbmtl commented 1 year ago

Awesome!

Just an FYI, my patch didn't work :disappointed:. I can get into the launched machine and see that the driver is installed when I list installed packages with yum, but the driver isn't loaded when I lsmod it. I also lshw and it kicked back saying the device is "UNCLAIMED" which tells me the driver isn't running at start. I also used dmseg to check if the amdgpu is initialized and it isn't. I followed the docs from AWS but it looks like I would need to reboot the machine to finish the install which should be happening after the machine is shutdown and the AMI is generated no?

I also tried to do the incredibly hacky solution of just mounting the GPU into the container directly but since it isn't mounted in /dev/dri, that also doesn't work.

I have tried everything that I know how to do at this point. We are running on the NVIDIA's but I would absolutely LOVE having this set up to get our savings.

Thanks again for the quick reply!

megagabay commented 1 year ago

This would actually be great for us too. We currently are using Nvidia and we really do not have to. If we could switch to the G4ad's, that would save a ton for us as well.

lamepennies commented 1 year ago

We'd also love to have this. We can better split our costs on different configurations for our user groups.

oneBlueFish commented 1 year ago

Ironically, this popped into my head today as well, glad I stumbled onto this. Adding the AMD drivers would be amazing.

danielpodwysocki commented 1 year ago

+1 - would be really useful for moving workloads that don't need all the fancy features of Nvidia and making them cost-effective on AWS.

ddesmond commented 5 months ago

any update on this one yet?