Wiki source code of GPU Server
Version 12.1 by Thomas Coelho (local) on 2024/06/07 10:12
Hide last authors
| author | version | line-number | content |
|---|---|---|---|
 |
1.1 | 1 | {{box cssClass="floatinginfobox" title="**Contents**"}} |
| 2 | {{toc/}} | ||
| 3 | {{/box}} | ||
| 4 | |||
| 5 | = The GPU Server = | ||
| 6 | |||
| |
8.1 | 7 | The GPU machine is a two socket server with AMD EPYC 7313 processors. One processor a 16 Cores, actually with SMT enabled (32 Threads). It comes with 512 GB of memory and 2 x 4 TB U.3 (NVMe) SSDs as fast storage. There are** 8 AMD Instinct Mi 50** GPU cards for computing. |
| |
1.1 | 8 | |
| 9 | Access is given by SLURM and the separate partition "gpu". | ||
| 10 | |||
| |
10.1 | 11 | As software stack AMD ROCm is installed. This supports the ROCm and openCL interface. Current ROCm Stack is version 6.1. This is also packaged in Ubuntu 6.1. |
| |
1.1 | 12 | |
| 13 | (% class="box infomessage" %) | ||
| 14 | ((( | ||
| 15 | Because GPU computing is a new discipline, we can only provide limited information here. If you have something to share, please fell free to edit this page. | ||
| 16 | ))) | ||
| 17 | |||
| |
12.1 | 18 | {{warning}} |
| 19 | To have built in ROCm support in slurm, this machine has already been updated to Ubuntu 24.04. There are some parts of the ROCm stack included in the distribution which is a mixture of 5.7 and 6.0. Official Support from AMD for Ubuntu 24.04 is not yet available. Pytorch has succesfully tested with this setup. | ||
| 20 | {{/warning}} | ||
| 21 | |||
| |
1.1 | 22 | == Submitting == |
| 23 | |||
| 24 | GPUs are handled as generic resources in Slurm (gres). | ||
| 25 | |||
| 26 | Each GPU is handled as allocatable item. You can allocate up to 8 GPUs. You can do this by adding "~-~-gres=gpu:N", where N is the number of CPUs. | ||
| 27 | |||
| 28 | CPUs are handled as usual. | ||
| 29 | |||
| 30 | Example: Interative Seesion with 2 GPUs: | ||
| 31 | |||
| 32 | {{code language="bash"}} | ||
| 33 | srun -p gpu --gres=gpu:2 --pty bash | ||
| 34 | {{/code}} | ||
| 35 | |||
| 36 | == PyTorch == | ||
| 37 | |||
| 38 | A popular framework for machine learning is PyTorch. An up-to-date version with ROCm support must be installed with pip3 in a venv. | ||
| 39 | |||
| 40 | {{code language="bash"}} | ||
| 41 | python3 -m venv venc | ||
| 42 | . venv/bin/activate | ||
| 43 | {{/code}} | ||
| 44 | |||
| 45 | Install Pytorch: | ||
| 46 | |||
| |
4.1 | 47 | {{code language="bash"}} |
| |
9.1 | 48 | pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0 |
| 49 | |||
| 50 | |||
| |
4.1 | 51 | {{/code}} |
| |
1.1 | 52 | |
| |
11.1 | 53 | At time of writing it's not available for 6.1. Please check the pytorch Website for updates. |
| |
10.1 | 54 | |
| |
4.1 | 55 | You can test the installation with |
| 56 | |||
| 57 | {{code language="python"}} | ||
| 58 | import torch | ||
| 59 | |||
| 60 | print(torch.cuda.is_available()) | ||
| 61 | |||
| 62 | {{/code}} | ||
| 63 | |||
| |
1.1 | 64 | == Links == |
| 65 | |||
| |
2.2 | 66 | GPU Cards: [[https:~~/~~/www.amd.com/en/products/professional-graphics/instinct-mi50>>https://www.amd.com/en/products/professional-graphics/instinct-mi50]] |
| 67 | |||
| |
1.1 | 68 | ROCm documentation: [[https:~~/~~/rocm.docs.amd.com/en/latest/rocm.html>>https://rocm.docs.amd.com/en/latest/rocm.html]] |
| 69 | |||
| |
5.1 | 70 | Pytorch: [[https:~~/~~/pytorch.org/>>https://pytorch.org/]] |
| |
1.1 | 71 | |
| 72 | |||
| |
10.1 | 73 |