Hosted at UCLouvain (CISM), this cluster consists of more than
5000 AMD Epyc Genoa cores @ 3.7 GHz in the default partition. In that partition, hyperthreading is activated, and Slurm is configured to allocate by default both hardware threads of a core for every core that is requested, resulting in double the amount of CPUs. Also, there is a limit of 6GB of RAM per requested core. Jobs requesting more than 6GB of RAM per core will be allocated as many cores necessary so that this limit is respected.
It also offers five nodes equiped with Accelerated Computing Units (APUs) AMD MI300a that combine, on a single chip, a CPU, a GPU and their shared memory. On that partition, the allocation unit is the socket, not the core ; each socket must be requested as a GPU in Slurm. Software optimized for this partition is available in release modules suffixed with apu.
Compute nodes in the default partition have access to a fast 320 TB BeeGFS /scratch space, while those in the apu partition see a specifc globalscratch of 207TB.
All nodes are interconnected through a 100 Gbps HDR InfiniBand fabric.
Suitable for:
MPI parallel jobs (many communications, dozens of cores),
SMP/OpenMP parallel jobs on the default partition, and GPU workloads (double precision and lower precision) on the apu partition.
Max wall time: 2 days.
Resources
- Home directory (100 GB quota per user)
- One lobal working directory
/scratch($GLOBALSCRATCH) per partition - Node local working directory
$LOCALSCRATCH - Default batch queue*
Access / Support
SSH: lemaitre4.cism.ucl.ac.be (port 22) via the university gateway,
with proper login and an id_rsa.ceci key.
Support: CISM
Server SSH key fingerprints:
What’s this ?
ECDSA:
SHA256:krYWLlE32ygG0u8uYbXUNBRTpbxDoDVyCvg3B1zLvGQ
ED25519:
SHA256:mWlgUkE+tBNbklXLgvrt7pL/3Ohn7uidqFfBUU0fSkQ
RSA:
SHA256:NIhjzqQgxgkG7K1x4kqoFnNSGrbc9b8AUG8+JT68jg4
