Curie

Revision as of 14:36, 7 August 2021

Description

Curie is a cluster featuring hybrid CPU/GPU nodes. It is equipped with the following hardware/software:

• 3 nodes with 2 x CPU AMD Opteron 6238 (12 cores), 64 GB of RAM, 1 x 500 GB HD SATA, 3 x GPU Tesla K20Xm, Infiniband, OS CentOS 6.5.
• 3 nodes with 2 x CPU Intel Xeon Gold 5218 (16 cores), 192 GB of RAM, 2 x 480 GB HD SATA, 2 x GPU Tesla V100 16 GB of RAM, Infiniband, OS CentOS 6.5.

for a total of 6 nodes, 84 cores, 12 GPU's.

Note: the oldest cores with the M20xx Tesla GPU's are down.

Access

• Linux and Mac OS users can login using a terminal. Windows 10 users can use the PowerShell.

The first step si to open a tunnel. Within the DiSC internet this can be simply done as:

ssh -L 2000:192.168.20.253:22 account@192.168.9.18 -p 7000

or from outside the Department using

ssh -L 2000:192.168.20.253:22 account@147.162.63.10 -p 7000

where "account" is the user's account.

Then, in a different shell, login as:

ssh -p 2000 account@localhost

Through port 2000 users can also transfer files directly via the tunnel using the command scp.

Queues

The queue manager is SLURM and the following queues are available:

• curie2: max nodes 2, max walltime 336:00:00 (2 weeks), to use the Tesla K20Xm
• curie3: max nodes 2, max walltime 336:00:00 (2 weeks), to use the Tesla V100

Example SLURM file

A typical SLURM script will be as follow (see the Support page for more help):

#SBATCH --job-name=name
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --partition=curie2
#SBATCH --account=curie2
#SBATCH --time=100:00:00

commands to execute

where the parts in italic should be changed as appropriate.