rsnt_translations
56,420
edits
Using GPUs with Slurm: Difference between revisions
Jump to navigation
Jump to search
no edit summary
No edit summary |
No edit summary |
||
| Line 199: | Line 199: | ||
<!--T:13--> | <!--T:13--> | ||
If you need to run four single-GPU programs or two 2-GPU programs for longer than 24 hours, [[GNU Parallel]] is recommended. A simple example is | If you need to run four single-GPU programs or two 2-GPU programs for longer than 24 hours, [[GNU Parallel]] is recommended. A simple example is: | ||
<pre> | <pre> | ||
cat params.input | parallel -j4 'CUDA_VISIBLE_DEVICES=$(({%} - 1)) python {} &> {#}.out' | cat params.input | parallel -j4 'CUDA_VISIBLE_DEVICES=$(({%} - 1)) python {} &> {#}.out' | ||
</pre> | </pre> | ||
In this example the GPU ID is calculated by subtracting 1 from the slot ID {%} | In this example, the GPU ID is calculated by subtracting 1 from the slot ID {%} and {#} is the job ID, starting from 1. | ||
<!--T:14--> | <!--T:14--> | ||
| Line 214: | Line 214: | ||
... | ... | ||
</pre> | </pre> | ||
With this method, you can run multiple tasks in one submission. The <code>-j4</code> parameter means that GNU Parallel can run a maximum of four concurrent tasks, launching another as soon as | With this method, you can run multiple tasks in one submission. The <code>-j4</code> parameter means that GNU Parallel can run a maximum of four concurrent tasks, launching another as soon as one ends. CUDA_VISIBLE_DEVICES is used to ensure that two tasks do not try to use the same GPU at the same time. | ||
== Profiling GPU tasks == <!--T:65--> | == Profiling GPU tasks == <!--T:65--> | ||