Roles and computing resources¶
Mila assigns one of two affiliation statuses to its professors and principal investigators:
- Core researchers
- Affiliated researchers
Students, postdocs, and other trainees inherit their access rights from their supervisor's status. If your supervisor (or co-supervisor) is a core researcher, access to the Mila computing cluster is granted. If both your supervisor and co-supervisor are affiliated researchers, access is not granted.
To determine your own status, check the Mila affiliation of your supervisor and, if applicable, your co-supervisor.
Overview of available computing resources at Mila¶
The Mila cluster is a heterogeneous cluster with a variety of node types (see Node profile description) and is to be used for regular development. It uses Slurm to schedule jobs.
Mila cluster versus DRAC / PAICE clusters¶
There are a lot of commonalities between the Mila cluster and the clusters from Digital Research Alliance of Canada (DRAC / the Alliance) and Pan-Canadian AI Compute Environment (PAICE) (job scheduling with Slurm, filesystem, etc). At the time being, core researchers also have access to a large allocation of resources on the Alliance clusters through Mila's global allocation.
The main distinguishing factor is that we have more control over our own cluster than we have over the external ones. Notably, the compute nodes in the Mila cluster all have unrestricted access to the Internet, which is not the case in general (although some clusters allow it).
Mila students are advised to use a healthy diet of a mix of Mila and DRAC / PAICE clusters. This is especially true when a preferred cluster is oversubscribed and switching to a different cluster is possible.
See the clusters page for a list of all available clusters.
Guarantees about one GPU as absolute minimum¶
There are certain guarantees that the Mila cluster tries to honor when it comes to giving at minimum one GPU per student, all the time, to be used in interactive mode. This is strictly better than "one GPU per student on average" because it's a floor meaning that, at any time, a GPU can be requested and obtained immediately (although processing the request through Slurm might take a minute).
Interactive sessions are also possible on the Alliance clusters, and there are generally special rules that provide quicker access to resources when requested for a very short duration (for testing code before queueing long jobs). The same guarantee does not apply on the Mila clusters, however.