Since seeing this, I have to been itching to ask… When launching an environment there’s a text that says
Number of GPUs: 0, without a way to actually change that number. I assume there is a good reason not to enable it across the board.
Now a Keras based experiment my team is working on actually needs acceleration. Is there some way of getting (paid) access at least temporarily to a high performance instance? Or include that in a pipeline? Or export the job to some external high-performance cluster? I can’t see anything in the documentation. And it doesn’t seem like anyone has asked about it here, either. Maybe I just missed the memo
Hi @loleg that’s a very good question! We do have GPUs in some of our deployments, but those are not public. Right now, resource access is uniform, meaning there are no special groups of users - we are working on functionality that will allow us to specify specialized hardware that certain user pools would be allowed to run on. I think us leaving that option there is a sign of our optimism for how soon this will be available
Raising my overheating local GPU’s and virtual to you in hopeful attendance.
in attendance, we could try to have your own deployment of Renku linked to renkulab.io’s gitlab and authentication. All you need is a Google Cloud Platform account and a domain. I could help you setup such a small cluster if you’re interested.
That would be great, if still an option - or if there is a better one - could you contact me please by e-mail to discuss.