Partitions & SLURM
To view information about the available nodes and partitions, use the following commands:
sinfo
For more detailed information about a specific partition:
scontrol show partition <partition-name>
The login and control nodes of Elja host two compute clusters: HPC-Elja and HTC-Mimir. Partitions and groups are used to separate the two.
HPC-Elja: Available Partitions / Compute Nodes
| Count | Name | Cores/Node | Memory/Node (GB) | Features |
|---|---|---|---|---|
| 28 | 48cpu_192mem | 48 (2x24) | 192 (188) | Intel Gold 6248R |
| 55 | 64cpu_256mem | 64 (2x32) | 256 (252) | Intel Platinum 8358 |
| 4 | 128cpu_256mem | 128 (2x64) | 256 (252) | AMD EPYC 7713 |
| 3 | gpu-1xA100 | 64 (2x32) | 192 (188) | Nvidia A100 Tesla GPU |
| 5 | gpu-2xA100 | 64 (2x32) | 192 (188) | Dual Nvidia A100 Tesla GPU |
| 1 | gpu-8xA100 | 128 (2x64) | 1000 (996) | 8 Nvidia A100 Tesla GPUs |
Core to true (memory) ratio equates to 3.9GB per core.
HPC-Elja: Job Limits
Each partition has a maximum seven (7) day time limit. Additionally, the queues any_cpu and long are provided:
- any_cpu: all CPU nodes, one (1) day time limit
- long: ten 48cpu and ten 64cpu nodes, fourteen (14) day time limit
HTC-Mimir: Available Partitions / Compute Nodes
| Count | Name | Cores/Node | Memory/Node (GB) | Features |
|---|---|---|---|---|
| 9 | mimir | 64 (2x32) | 256 (252) | |
| 1 | mimir-himem | 64 (2x32) | 2048 (2044) |
Core to true (memory) ratio equates to 3.9GB per core on the mimir partition, and 31GB on the mimir-himem partition.
HTC-Mimir: Job Limits
Either partition has a fourteen (14) day time limit.
HPC-Stefnir: Available Partitions / Compute Nodes
| Count | Name | Cores/Node | Memory/Node (GB) | Features |
|---|---|---|---|---|
| 44 | scompute | 64 | 256 (252) | Intel(R) Xeon(R) Platinum 8358 |
Core to true (memory) ratio equates to 3.9GB per core.
HPC-Stefnir: Job Limits
Each partition has a maximum two (2) day time limit.
SLURM Configuration
SLURM is configured such that 3.94GB of memory is allocated per core.
Available Memory
On each node, 2-4 GiB RAM are reserved for the operating system images (hence the true value is in parentheses).