Before connecting to LUMI, you need to generate an SSH key pair and upload your public key to MyAccessID.
You can do additional setup, like adding your key to an agent or setting up a shortcut for LUMI in your SSH configuration.
Learn more about the hardware¶
In the first phase of the LUMI installation, LUMI-C, the CPU partition and LUMI-D, the data analytics partition are installed. LUMI-C consists of 1536 compute nodes fitted with two last generation AMD EPYC "Milan" 64-core CPUs and 256 GiB of memory.
LUMI-D consists of 12 nodes either with a large memory or NVDIA RTX GPUs as well as on node storage.
Setup your Environment¶
Software on LUMI can be accessed through modules. With the help of the
command, you will be able to load and unload the desired compilers, tools and
Running your Jobs¶
To get started with running your application on LUMI, you need to write a batch jobs script and submit it to the scheduler and resource manager. LUMI uses Slurm as the batch job system.
- Get familiar with Slurm with the quick start guide
- Learn more about the available Slurm queues
- Check out some example batch scripts
- See how to run a job interactively
- See how to submit a large number of independent jobs
Please note that only the data analytics nodes have local storage, when running on the compute nodes in LUMI-C, the input and output data of your application must be stored in the scratch spaces of the parallel file systems.