As stated in the overview chapter, in Cluster DEI there are currently 9 servers with GPUs:
Limitations regarding GPUs
- Each user has a limit of 8 GPUs: it is strongly advisable to request the minimum number of GPUs needed for the job to start in a reasonable time and not to waste resources;
- jobs requiring A40 GPUs are limited to 8 cores per GPU;
- jobs requiring RTX 3090 GPUs are limited to 6 cores per GPU.
In order to request one or more GPUs a user needs to use the SBATCH instruction followed by the --gres option with the indication of the group of GPUs and also the number, if more than one graphic card is needed. It is mandatory to choose a group, it can't be requested a generic card.
The groups are the following:
The number is only needed if more than one GPU is requested and has to be separated from the GPU group by a colon sign. So if I need 3 Nvidia A40s, I will use the following option of the sbatch command:
Here are some possible uses of the SBATCH command:
#SBATCH --gres=gpu:rtx Use one Nvidia RTX 3090 GPU
#SBATCH --gres=gpu:rtx:3 Use three Nvidia RTX GPUs
#SBATCH --gres=gpu:p2000 Use one Nvidia Quadro P2000 GPU
#SBATCH --gres=gpu:a40 Use one Nvidia A40 GPU
Warning
Do not request GPUs that you won't use!
Notice about resources
Do not underestimate the importance of correctly calibrating the request of resources for a job.
Please double check if your CPUs, RAM e GPUs requests are adequate for the code you are going to execute. The efficiency must be high, otherwise your execution can result in other users having problems to launch their jobs and, in the worst case, it can lead to a forced termination of a job launched with an unwise resource allocation.Consult the chapter How my jobs are performing for more information.