As of Monday, August 12, the new queues on Sophia are:
- by-gpu queue: This is the default queue and the queue for GPUs.
- by-node: This queue is for a complete DGX node (8 GPUs), and you may request up to eight nodes. We may adjust the maximum node count based on experience.
- bigmem: This queue is to use one of the nodes with 640 GB of aggregate GPU memory. There are no changes to this queue.
NOTE: The single-gpu and single-node queues have become routing queues pointing to by-gpu and by-node, respectively, so existing scripts will not need to be updated, but new scripts will need to be updated to use by-gpu and by-node. For more information, see https://docs.alcf.anl.gov/sophia/queueing-and-running-jobs/running-jobs/