Scientific Data & Computing Center

["Caramarcu Costin" <caramarc AT bnl.gov>]

Hi all,

 

As it was announced in our last liaison meeting the Nvidia P100 nodes are now ready for production. The new nodes have the same specs as the current one (same: cpu, memory, ib) and instead of 2x Tesla K80 we have 2x Pascal P100.

 

In order to run on the P100 gpu’s you need to specify the constraint (--constraint=pascal or -C pascal) not that P100 have only 1 GPU per card (--gres=gpu:2). If you need to make sure you run on K80 same rule applies (--constraint=tesla or -C tesla).

 

Something to take note of:

-          constrains are not mandatory,

-          specifying just --gres=gpu:2 or --gres=gpu:1 can land your job in pascal, tesla or mixed

 

For your convenience here are some sample scripts:

P100:

#SBATCH -p long

#SBATCH -t=01:00:00

#SBATCH --account accountname

#SBATCH -N 2

#SBATCH -n 64

#SBATCH -C pascal

#SBATCH --qos normal

#SBATCH --gres=gpu:2

module load mpi/openmpi-1.10.2

module load gcc/5.3.0

srun ./my_executable

 

K80:

#SBATCH -p long

#SBATCH -t=01:00:00

#SBATCH --account accountname

#SBATCH -N 2

#SBATCH -n 64

#SBATCH -C tesla

#SBATCH --qos normal

#SBATCH --gres=gpu:4

module load mpi/openmpi-1.10.2

module load gcc/5.3.0

srun ./my_executable

 

CPU ONLY:

#SBATCH -p long

#SBATCH -t=01:00:00

#SBATCH --account accountname

#SBATCH -N 2

#SBATCH -n 64

#SBATCH --qos normal

module load mpi/openmpi-1.10.2

module load gcc/5.3.0

srun ./my_executable

 

 

For any problems please contact us: https://www.sdcc.bnl.gov/#support

 

Regards,

Costin & Zihua