EEMCS-HPC Specific resources (gpu), features and partitions

The generic consumable resources on the HPC cluster are :

• gpu[:family]:amount (omit the family directive or use one of the family names listed below).
• others are not supported yet.

To request a gpu add the following line to your sbatch script :

#SBATCH --gres=gpu:1 

For selecting a specific gpu family, add the family name to the gpu request.

• pascal
• turing
• ampere
• lovelace
• blackwell
• hopper

To request a specific gpu family, update the following line in your sbatch script (As an example the lovelace is selected) :

#SBATCH --gres=gpu:lovelace:1 

Keep in mind for gpu's you need to load the module of the required cuda version !

Once you request a gpu(s) resource, the scheduler will set for you the environment variable : CUDA_VISIBLE_DEVICES

This will point to the assigned gpu(s) for your job, you shall only use those and not others !!!

Some gpu boards have the nvlink modules fitted, this will allow you to double the gpu memory and cumputing power. If you request two gpu's with nvlink, you need to force socket binding using the following option :

#SBATCH --sockets-per-node=1 

The available features are :

• amd/intel, refers to the silicon builder of the Cpus
• avx/avx2/avx512, refers to the avx, avx2 and avx512 instruction set, available in the newer nodes (for example: Keras, Tensorflow v1.6 and above)
• tesla, refers to the Tesla family cards (ctit080)
• geforce, refers to the GeForce family cards (ctit[081-083])
• quadro, refers to the Quadro family cards (ctit[084-085])
• p100, refers to the specific Tesla P100 model (ctit080)
• titan-x refers to the specific Titan-X model (ctit[081-082])
• gtx-1080ti refers to the specific GeForce gtx-1080ti model (ctit083).
• rtx-6000 refers to the specific Quadro rtx-6000 model (ctit[084-085]).
• rtx-2080ti refers to the specific GeForce rtx-2080ti model (ctit088).
• titan-xp refers to the specific Titan-X model (ctit088)
• a40 refers to the specific Tesla A40 model (ctit086, 90..94)
• a100 refers to the specific Tesla A100 model (ctit089)
• l40 refers to the specific Tesla L40 model (hpc-node01..08)
• l40s refers to the specific Tesla L40s model (hpc-node09..12,16..17)
• rtx6000pro refers to the specific Rtx 6000 Pro model (hpc-node15)
• h200nvl refers to the specific H200NVL model (hpc-node19)

For example to force only a40 gpu's

#SBATCH --constraint=a40 

Hint : Use ampersand (AND) and pipe (OR) symbol to combine features.

The HPC/SLURM cluster contains multiple common partitions :

Compute nodes with gpus are allowed to run cpu jobs, but keep cpu resources available for gpu jobs !!!!!

As well as faculty/group specific additional partitions :

The following partition is only available for admins:

The main partition is the default partition that will be used to submit a job to any of the nodes. The debug partition is for testing purposes only.

Access to the additional are limited to the funders during the first year of investment, these can be reached using the funders partitions.

Including multiple partitions is also possible. For example :

• EEMCS-DMB members

#SBATCH --partition=main,dmb

• EEMCS-MIA members

#SBATCH --partition=main,am,mia

• EEMCS-Students

#SBATCH --partition=main,students

• etc.

See the EEMCS-HPC Hardware page for all the partition definitions.