CLAIX-2023
The CLAIX-2023 cluster is the current computing component of the HPC system CLAIX. Installed in 2023 by NEC, it has been available to users since January 2024. The cluster is funded through three sources: Tier-2 NHR, Tier-3, and WestAI.
Picture of CLAIX-2023.
The cluster consists of two segments. The traditional HPC Segment is equipped with Intel Xeon 8468 Sapphire Rapids CPUs, while the Machine Learning (ML) Segment includes additional NVIDIA H100 GPUs for accelerated ML workloads (see details in table below). In addition, CLAIX-2023 provides several login-nodes.
The HPC Segment offers compute nodes with varying amounts of main memory. You can select a node that meets your memory requirements by specifying the appropriate partition in your batch job.
All nodes are interconnected via an NVIDIA/Mellanox NDR InfiniBand network, providing up to 25 GB/s of unidirectional communication bandwidth between nodes. A fat-tree network structure ensures high bandwidth between all nodes.
The cluster employs direct liquid cooling for CPUs, GPUs and main memory offering improved and efficient heat dissipation compared to traditional air cooling.
Segments | Number of | Total | Total number of | Available computing |
|---|---|---|---|---|
CLAIX-2023-HPC Per Node:
| 632 | 4077 (CPU) | cores: 63552 | NHR: 346 mio. core-h |
| Tier-3: 185 mio. core-h | ||||
CLAIX-2023-ML Per Node:
| 52 | 335 (CPU) + 7072 (GPU) | cores: 4992 GPUs: 208 | NHR: |
| Tier-3: 167 k GPU-h | ||||
| WestAI: 500 k GPU-h |
