Oscar has two Grace Hopper GH200 GPU nodes. Each node combines Nvidia Grace Arm CPU and Hopper GPU architecture.
Each GH200 node has 72 Arm cores with 550G memory. Multiple-Install GPU (MIG) is enabled on only one GH200 node that has 4 MIGs. The other GH200 node doesn't have MIGs and only one GPU. Both CPU and GPU threads on GH200 nodes can now concurrently and transparently access both CPU and GPU memory.
The two GH200 nodes are in the gracehopper partition.
A gk-condo user can submit jobs to the GH200 nodes with their gk-gh200-gcondo account, i.e.,
#SBATCH --account=gk-gh200-gcondo
#SBATCH --partition=gracehopper
For users who are not a gk-condo user, a High End GPU priority account is required for accessing the gracehopper partition and GH200 nodes. All users with access to the GH200 nodes need to submit jobs to the nodes with the ccv-gh200-gcondo account, i.e.
#SBATCH --account=ccv-gh200-gcondo
#SBATCH --partition=gracehopper
To request a MIG, the feature mig needs be specified, i.e.
#SBATCH --constraint=mig
NGC containers provide the best performance from the GH200 nodes. Running tensorflow containers is an example for running NGC containers.
{% hint style="info" %} A NGC container must be built on a GH200 node for the container to run on GH200 nodes {% endhint %}
The two nodes have Arm CPUs. So Oscar modules do not run on the two GH200 nodes. Please contact [email protected] about installing and running modules on GH200 nodes.