compute_ru

User Tools

Site Tools

Trace:
compute:step2

Differences

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

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
compute:step2 [2023/04/24 10:13] – bjornscompute:step2 [2024/10/14 14:24] (current) – external edit 127.0.0.1
Line 1: Line 1:
 ====What's next ?==== ====What's next ?====
 +Home folder for students is /student/<user>\\
 Most users are using Python for their jobs\\ Most users are using Python for their jobs\\
 Setup your environment, Python virtual environment/Anaconda and install in you home folder the necessary libraries for your project\\ Setup your environment, Python virtual environment/Anaconda and install in you home folder the necessary libraries for your project\\
 +\\
 +When you are ready to send your job to the slurm queue, adjust the slurm command script\\
 +Slurm will put you on the queue, if all the resources you ask for are available you will go to Run(R) state, otherwise it will be Pending(Pd)\\
 \\ \\
 **The slurm command script** : sbatch script.sh\\ **The slurm command script** : sbatch script.sh\\
Line 10: Line 14:
 |''#SBATCH --mem=4G''|Job wants 4gb on memory| |''#SBATCH --mem=4G''|Job wants 4gb on memory|
 |''#SBATCH --time=0''|Time limit on my job e.g. time=11::00 (11 hours), 0 means nolimit| |''#SBATCH --time=0''|Time limit on my job e.g. time=11::00 (11 hours), 0 means nolimit|
-|''#SBATCH --partition=Lokaverk''|Run job on this queue|+|''#SBATCH --partition=Lokaverk''|Send job to run this queue|
 |''#SBATCH <nowiki>--</nowiki>output=myBatch.log''|Log file for environment & slurm| |''#SBATCH <nowiki>--</nowiki>output=myBatch.log''|Log file for environment & slurm|
 +|python3 file.py | you put your run command after the directives|
  
 When you job need GPU add this line to you slurm cmd file\\ When you job need GPU add this line to you slurm cmd file\\
 #SBATCH %%--%%gpus-per-node=1\\ #SBATCH %%--%%gpus-per-node=1\\
 +**Note** : We recommend all students in this 3-week-course run their **GPU job** for a short time, using "--time=hh:MM" slurm directive, so all jobs get some GPU time\\
  
 ==Other slurm directives== ==Other slurm directives==
 #SBATCH %%--%%mem-per-cpu=2G\\ #SBATCH %%--%%mem-per-cpu=2G\\
 #SBATCH %%--%%cpus-per-task=2 : 2 cores per process/task\\ #SBATCH %%--%%cpus-per-task=2 : 2 cores per process/task\\
-#SBATCH %%--%%ntasks-per-node=4 : 4 procees per node(worker\\ +#SBATCH %%--%%ntasks-per-node=4 : 4 procees per node/worker\\ 
 +[[https://slurm.schedmd.com/sbatch.html|Slurm sbatch]]\\
 ===Tools and libraries installed=== ===Tools and libraries installed===
 CUDA toolkit version 11,7\\ CUDA toolkit version 11,7\\
compute/step2.1682331190.txt.gz · Last modified: 2024/10/14 14:24 (external edit)