High-End Computing Capability
Computing power to answer NASA's complex science and engineering questions
Cabeus Supercomputer
Cabeus is NASA's latest GPU system, currently comprised of 128 NVIDIA GPU A100 nodes and 538 NVIDIA GH200 superchip nodes. The statistical information on this page is split into two groups based on their operating system: the A100 augmented nodes using x86_64 and the GH200 nodes using aarch64.
Named after a lunar crater in the south polar region of the Moon, Cabeus is a cluster system built to support high performance computing and machine learning/artificial intelligence applications that can benefit from GPU technology.
System Architecture
A100 Nodes
- Manufacturer: HPE
- 12 racks (128 nodes)
- 5.31 Pflop/s peak cluster
- Total CPU cores: 8,192
- Total GPU cores: 1,769,472
- Total memory: 106 TB
- CPU host memory: 65 TB
- GPU memory: 40 TB
GH200 Nodes
- Manufacturer: Supermicro
- 23 racks (538 nodes)
- 20.10 Pflop/s peak cluster
- Total CPU cores: 38,736
- Total GPU cores: 4,545,024
- Total memory: 309.9 TB
- CPU host memory: 258.2 TB
- GPU memory: 52.6 TB
| Grace Hopper-GH200 Nodes | Milan-A100 Nodes | |
|---|---|---|
| Number of Nodes | 538 | 128 |
| Processors per Node | One superchip containing a 72-core CPU and GPU pair | One 64-core host processor and four GPU coprocessors |
| Node Types | NVIDIA Grace Neoverse V2 (CPU); NVIDIA Hopper H100 (GPU) | AMD 7763 EPYC (host); NVIDIA Ampere A100 (GPU) |
| Processor Speed | 3.47 GHz (host); 1,980 MHz (GPU) | 2.45 GHz (host); 1,275 MHz (GPU) |
| L3 Cache | 114 MB shared by 72 cores per CPU processor | 32 MB shared by 8 cores in a Core Complex (CCX), total of 256 MB per processor |
| Memory Type | LPDDR5X (host); HBM3 (GPU) | DDR4 FB-DIMMs (host); HBM2e (GPU) |
| Memory Size | 480 GB (host); 96 GB (GPU) | 512 GB per node (host); 80 GB per GPU card |
| Interconnects | InfiniBand NDR host channel adapter and switches (host) | InfiniBand HDR host channel adapter and switches (host) |
Operating Environment
- Operating System: Tri-Lab Operating System Stack (TOSS)
- Job scheduler: Altair PBS Professional
- Compilers: Intel, NVIDIA HPC, and GNU C, C++ and Fortran
- MPI: HPE MPT, NVIDIA HPC OpenMPI
Quick Links
- News
- User Documentation
- myNAS User Portal
- System Status
- FAQ
- Get Accounts
- New User Orientation
- Training Webinars
Can't find what you're looking for? NAS User Support is available 24x7x365:
(800) 331-8737
(650) 604-4444
support@nas.nasa.gov
Now at HECC
Live status information for Athena, Aitken, Electra, Cabeus and Endeavour can only be viewed with Javascript enabled in your browser.
