|
| 1 | +# New Tier-2 cluster: `litleo` |
| 2 | + |
| 3 | +!!! warning |
| 4 | + The `litleo` cluster is currently only available to members of the `gpilot` user group. |
| 5 | + |
| 6 | + |
| 7 | +In December 2024, a new cluster was added to the HPC-UGent Tier-2 infrastructure: `litleo`. |
| 8 | + |
| 9 | +If you have any questions on using `litleo`, you can [contact the {{ hpcteam }}]({{ hpc_support_url }}). |
| 10 | + |
| 11 | +For software installation requests, please use the [request form](https://www.ugent.be/hpc/en/support/software-installation-request). |
| 12 | + |
| 13 | +--- |
| 14 | + |
| 15 | +## `litleo`: GPU cluster |
| 16 | + |
| 17 | +`litleo` is a new GPU cluster. |
| 18 | + |
| 19 | +It adds GPU capacity to existing clusters `joltik` and `accelgor`. |
| 20 | + |
| 21 | +It is only intended for ***single GPU*** compute use. |
| 22 | + |
| 23 | +This cluster consists of 8 workernodes, each with: |
| 24 | + |
| 25 | +* 1x 48 core AMD EPYC 9454P (Genoa @ 2.75 GHz) processor; |
| 26 | +* ~315 GiB of RAM memory; |
| 27 | +* 1.4TB local NVMe disk; |
| 28 | +* NDR-200 InfiniBand interconnect; |
| 29 | +* RHEL9 as operating system; |
| 30 | +* 2x NVIDIA H100 NVL (96GB VRAM) |
| 31 | + |
| 32 | +To start using this cluster from a terminal session, first run: |
| 33 | +``` |
| 34 | +module swap cluster/.litleo |
| 35 | +``` |
| 36 | + |
| 37 | +You can also start (interactive) sessions on `litleo` using the [HPC-UGent web portal](../../../web_portal.md). |
| 38 | + |
| 39 | +### Difference compared to `joltik` and `accelgor`. |
| 40 | + |
| 41 | +The main difference between `litleo` and the other GPU clusters is that *jobs are limited to using only one GPU per job*. |
| 42 | +The motivation comes from historical analysis of GPU usage on the `joltik` and `accelgor` clusters that show a very |
| 43 | +high utilisation of a single GPU per job; combined with cost of the GPUs in 2024. It allowed us get the most out of the |
| 44 | +budget. |
| 45 | + |
| 46 | +--- |
| 47 | + |
| 48 | +## Litleo pilot phase (24/01/2025-01/03/2025) |
| 49 | + |
| 50 | +As usual with any pilot phase, you need to be member of the `gpilot` group, and to start using this cluster run: |
| 51 | + |
| 52 | +``` |
| 53 | +module swap cluster/.litleo |
| 54 | +``` |
| 55 | + |
| 56 | +Shortly after the Tier-2 maintenance in February 2025, the cluster will enter production, without any significant changes. |
0 commit comments