Extended beta testing starts on LUMI-G

An extended beta testing period starts on the GPU partition (LUMI-G) today, 19 December 2022. This means that all computing projects which have been allocated GPU resources, in addition to the pilot projects, will be able to submit jobs on the whole LUMI-G.

The extended beta testing period will continue until mid–late January 2023, until all the steps for formally accepting the installation have been completed.

There are now three queues for the GPU partition:

• standard-g: max 1024 GPU nodes, max runtime 48 hours. Exclusive mode, i.e., full nodes reserved.
• small-g: max 4 GPU nodes, max runtime 72 hours. Possibility of allocating individual GPUs.
• dev-g: max 16 GPU nodes, max runtime 6 hours.

At the same time, the EAP (Early Access Platform) and pilot queues will be closed.

Please note that currently the GPU nodes have 63 cores available in the host CPU (as the 1st core is reserved for operating system purposes). This causes jobs submitted to the queue asking, for example, 8 MPI x 8 OpenMP, to get stuck and/or denied due to the resources being unavailable. Refer to docs.lumi-supercomputer.eu for batch job script examples.

Note that the system is not in its final shape. Performance variations, shortcomings in the programming environment etc. may be observable. Please do not hesitate to report all observed shortcomings and issues to the LUMI User Support Team helpdesk in order for them to get fixed.