moved to https://docs.hpc.taltech.ee not changed to rocky yet
LUMI
What is LUMI?
2560 GPU-based nodes (LUMI-G), each node with one 64 core AMD Trento CPU and four AMD MI250X GPUs.
1536 dual-socket CPU nodes LUMI-C with 64-core 3rd-generation AMD EPYC™ CPUs, and between 256 GB and 1024 GB of memory.
large memory GPU nodes LUMI-D, with a total of 32 TB of memory in the partition for data analytics and visualisation.
Main storage - LUMI-P has 4 independent Lustre file systems with 20 PB and an aggregate bandwidth of 240 GB/s each. Each Lustre file system is composed of 1 MDS (metadata server) and 32 Object Storage Targets (OSTs).
Flash storage - LUMI-F has Lustre file system with a storage capacity of 8 PB and an aggregate bandwidth of 1 740 GB/s.
Object store - LUMI-O provides 30 PB storage.
More about LUMI system architecture can be found in overview and LUMI’s full system architecture.
LUMI uses Slurm as job scheduler and resource manager. Slurm partitions can be allocated by node or by resources. More about partitions can be found here.
Why LUMI?
if job is run using GPUs
if job needs large memory
if queue on HPC is too long