The Initial Release Period for the Cabeus Grace Hopper nodes (gra_h100) ended on March 17, 2025, at 12:00 p.m. (Pacific Time). Any jobs that were running on the nodes at that time are not being charged. Any jobs that were queued in the alpha_test queue have been moved to another appropriate queue and are being charged the Standard Billing Unit (SBU) rate of 18.34.

To run on the GH200 nodes, submit your PBS interactive or batch job from one of the two Grace Hopper front ends, cghfe01 or cghfe02. For more information, please see Preparing to Run on Cabeus Grace Hopper Nodes.

Original Notification:

Cabeus has been augmented with 350 Grace Hopper GPU nodes, which are now available for production use for those users with GPU allocations.

Each Grace Hopper node contains one GPU, for a total of 350 GPUs. For the initial two weeks, user jobs will not be charged. To access the new nodes, specify the model type (gra_h100), queue name (alpha_test), and submit your job from one of two new front ends (cghfe01 or cghfe02). For example:

#PBS -lselect=1:model=gra_h100:ncpus=16:ngpus=1:mem=60GB
#PBS -q alpha_test

For more information about using Grace Hopper, see the following articles in the HECC Knowledge Base:

• Preparing to Run on Grace Hopper GPU Nodes
• Cabeus GPU Nodes
• Requesting Cabeus GPU Resources
• Programming for NVIDIA GPUs

We encourage your feedback on providing performance results and any questions or issues you have when using these nodes. Please send an email to support@nas.nasa.gov with any feedback or questions.