Partitions
On this page you will learn about the available SLURM-partitions for both of our clusters CLAIX-2023 and CLAIX-2025.
Partitions in Slurm are sets of computing nodes with dedicated queues that represent different sets of nodes with specific hardware configurations. In our case we distinguish nodes with different memory configurations and gpus.
CLAIX-2023 Partitions
For jobs on CLAIX-2023 it is not necessary to request a specific partition by default. Without any special requirements, jobs will be scheduled to the c23ms partition. If you request GPU resources, your job will be scheduled to the c23g partition.
In some cases it may be useful to specify a partition explicitly if your job has specific requirements regarding memory. See the table below for a detailed overview of the partitions:
| Partition | Nodes | Cores per node | Memory per core* | Billing | remarks |
| c23ms | 625 | 96 | 2540 MiB** | Regular | Small memory, Default partition |
| c23mm | 166 | 96 | 5210 MiB | Medium memory | |
| c23ml | 2 | 96 | 10560 MiB | Large memory | |
| c23g | 50 | 96 | 5200 MiB | Using one GPU for one hour is billed as 24 core-h | (each node has four H100 GPUs) 1 GPU is limited to 24 cores and 122GB |
| devel | 2 | 96 | 975 MiB | Free | Designated for testing purposes. Maximum runtime of 1 Hour Only few simultaneous jobs Please use without project |
* This is the default value and the recommended maximum for #SBATCH --mem-per-cpu
** 1 MiB = 220 Bytes = 1024 kiB = 1048576 Bytes = 1.048576 MB
Memory Tiers
The nodes in the Claix-2023 cluster offer three memory configurations: small, medium and large. The c23ms partition includes all nodes, the c23mm partition includes nodes with medium and large memory configurations, and the c23ml partition includes only nodes with the large memory configuration.
This setup allows you to select the partition that meets your memory requirements, while also ensuring that all nodes are utilized, even if the c23mm or c23ml queues are empty.
Testing
The devel partition is designed for quick testing of computing jobs, offering short wait times but limited run times. You can find more information here.
Low-Partitions
Slurm Jobs might be automatically placed on a c23*_low Partition if the compute quota exceeds the 3-month quota.
These Jobs run on the same hardware / partition they were submitted to, but are scheduled with less priority.
These lower priority jobs (_low Partition / QoS) will only start and run when the respective partition resources are completely free and no other normal priority jobs are pending.
CLAIX-2025 Partitions
Please note that CLAIX-2025 is accessible only to users with computing time allocations through NHR or WestAI computing time projects.
| Partition | Nodes | Cores per node | Memory per core* | Billing | remarks |
| c25ms | 418 | 192 | 1940 MiB** | Regular | Small memory, Default partition |
| c25ml | 108 | 192 | 3960 MiB | Large memory | |
| c25g | 20 | 128 | 5960 MiB | Using one GPU for one hour is billed as 24 core-h | (each node has four H100 GPUs) 1 GPU is limited to 24 cores and 122GB |
* This is the default value and the recommended maximum for #SBATCH --mem-per-cpu
** 1 MiB = 220 Bytes = 1024 kiB = 1048576 Bytes = 1.048576 MB
Memory Tiers
The nodes in the CLAIX-2025 cluster offer two memory configurations: small and large. The c25ms partition includes all nodes, the c25ml partition includes nodes with small and large memory configurations.
This setup allows you to select the partition that meets your memory requirements, while also ensuring that all nodes are utilized, even if the c25ml queue is empty.
Testing
The CLAIX-2025 cluster has no devel partition.
