Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision		 Previous revision
habrok:advanced_job_management:running_jobs_on_gpus [2024/01/15 10:50] – [Running jobs on GPUs] camarocicohabrok:advanced_job_management:running_jobs_on_gpus [2026/02/27 10:20] (current) – [Example] fokke
Line 18: Line 18:
 ==== Available GPU types ==== ==== Available GPU types ====
  
-^ Node ^ GPU type ^  GPUs per node ^  Memory per GPU ^  CPUs per node ^  Memory per node ^ Slurm name ^ Notes ^ +^ Node         ^ GPU type            ^  GPUs per node ^  Memory per GPU ^  CPUs per node ^  Memory per node ^ Slurm name ^ 
-| A100_1 | Nvidia A100  |  4 |  40 GB |  64 |  512 GB | a100 | Full A100 cards | +| A100         | Nvidia A100         |  4 |  40 GB |   64 |  512 GB | a100 | 
-| A100_2 | Nvidia A100  |  8 |  20 GB |  64 |  512 GB | a100.20gb | Two virtual GPUs per A100 card | +| V100         | Nvidia V100         |  2 |  32 GB |   36 |  128 GB | v100 | 
-| V100 | Nvidia V100  |  1 |  32 GB |  8|  128 GB | v100 | |+| RTX Pro 6000 | Nvidia RTX Pro 6000 |  8 |  96 GB |  128 |  192 GB | rtx_pro_6000 | 
 +| L40S         | Nvidia L40S         |  2 |  48 GB |   56 |  512 GB | l40s |
  
 ==== Example ==== ==== Example ====
Line 29: Line 30:
 <code> <code>
 #SBATCH --gpus-per-node=a100:2 #SBATCH --gpus-per-node=a100:2
-</code> 
-If you want to request a node with half of an NVIDIA A100, use the following: 
- 
-<code> 
-#SBATCH --gpus-per-node=a100.20gb:1 
 </code> </code>
  
Line 41: Line 37:
 #SBATCH --gpus-per-node=1 #SBATCH --gpus-per-node=1
 </code> </code>
 +
 +Note that this will only sent jobs to the V100 and A100 nodes. This because not all software is compatible with the RTX  Pro 6000 GPUs. Furthermore the more capable RTX Pro 6000 nodes should not be swamped with jobs that don't need its capabilities. See [[rtx_pro_6000_gpu_nodes]] for more details.
  
 ==== Interactive GPU node ==== ==== Interactive GPU node ====
Line 47: Line 45:
  
 <code> <code>
-gpu1.hpc.rug.nl +gpu1.hb.hpc.rug.nl 
-gpu2.hpc.rug.nl+gpu2.hb.hpc.rug.nl
 </code> </code>
  
-These machines have an NVIDIA V100 GPU each, which can be shared by multiple users. The tool ''nvidia-smi'' will show if the GPU is in use. +These machines have an NVIDIA L40S GPU each, which can be shared by multiple users. The tool ''nvidia-smi'' will show if the GPU is in use. 
  
 ** Please keep in mind that this is a shared machine, so allow everyone to make use of these GPUs and do not perform long runs here. Long runs should be submitted as jobs to scheduler. ** ** Please keep in mind that this is a shared machine, so allow everyone to make use of these GPUs and do not perform long runs here. Long runs should be submitted as jobs to scheduler. **
 ==== Running interactive jobs ==== ==== Running interactive jobs ====
  
-You can request an interactive session by using a command like:+You can usually request an interactive session by using a command like:
 <code> <code>
 srun --gpus-per-node=1 --time=01:00:00 --pty /bin/bash srun --gpus-per-node=1 --time=01:00:00 --pty /bin/bash
 </code> </code>
 +
 +There is currently an issue with using ''srun %%--%%gpus-per-node'', but there is a workaround by using '' %%--%%gres'' instead:
 +<code>
 +srun --gres=gpu:1 --time=01:00:00 --pty /bin/bash
 +</code>
 +
 +or:
 +<code>
 +srun --gres=gpu:v100:1 --time=01:00:00 --pty /bin/bash
 +</code>
 +
  
 When the job starts running, you will be automatically logged in to the allocated node, allowing you to run your commands interactively. When you are done, just type ''%%exit%%'' to close your interactive job and to release the allocated resources. When the job starts running, you will be automatically logged in to the allocated node, allowing you to run your commands interactively. When you are done, just type ''%%exit%%'' to close your interactive job and to release the allocated resources.
  
 **N.B.: interactive jobs currently don't (always) use the software stack built for the allocated nodes, you can work around this by first running ''unset SW_STACK_ARCH && module restore'' after the job has started.** **N.B.: interactive jobs currently don't (always) use the software stack built for the allocated nodes, you can work around this by first running ''unset SW_STACK_ARCH && module restore'' after the job has started.**