Two different things. This guide is about a lab building its own Slurm cluster in its Ursa Major project. That is recharged to a lab funding source (COA) under an MOU.
Separately, Research Computing runs a shared HPC cluster in Google Cloud for Tier 1 exotic hardware the HPCC does not have, such as TPUs or Arm processors. Access starts with a consultation, and it is campus-supported under current terms, within limits set per project. See KB005. Ask us which fits.
For most batch and GPU work, the campus HPCC cluster is the first place to look and usually costs the lab less.
What it is
Using Google's Cluster Toolkit, a lab can run its own Slurm cluster in its Ursa Major project:
- personal, lab or shared clusters;
- software installed to the lab's needs;
- compute nodes that scale up with the queue and down when idle (recharged by usage);
- clusters that can be stopped and restarted without losing data on persistent storage;
- connection to Cloud Storage and managed file systems (storage is charged separately).
Getting started
- Launching an Ursa Major HPC cluster
- Connecting to a cluster and running a first job
- Google Cloud HPC solutions
Talk to Research Computing before you build a cloud cluster. We can help you compare it with the HPCC and estimate costs.