As the world of artificial intelligence and machine learning (AI/ML) evolves into tools that can be used for scientific discovery, the NCCS has built a 87-node GPU cluster, with 88 NVIDIA v100 GPUs, 8 Nvidia A100 GPUs and 62 H100 GPUs to facilitate this new research paradigm.
Prism has access to over 70 petabytes of curated NASA data products as well as data produced on both the Discover supercomputer and the ADAPT Science Cloud.
For those who enjoy having system names be acronyms, PRISM stands for Promoting Research In Science using Machine learning.
System Specifications
22 Supermicro Compute Nodes (X86)
One NVIDIA DGX Node
- 4x NVIDIA V100 GPUs with 32 GB of VRAM and NVLink
- Dual Intel Xeon Cascade Lake Gold 6248 CPUs;
20 cores each at 2.50GHz - 768 GB RAM
- Dual 25Gb Ethernet network interfaces
- Dual 100Gb HDR100 Infiniband high speed network interfaces
- 3.8 TB RAID protected NVMe drives, mounted as /lscratch
- 8x NVIDIA A100 GPUs with 40 GB of VRAM and NVLink
- Dual AMD EPYC Rome 7742 CPUs;
64 cores each at 2.25GHz - 1 TB RAM
- Dual 25Gb Ethernet network interfaces
- Dual 100Gb HDR100 Infiniband high speed network interfaces
- 14 TB RAID protected NVMe drives, mounted as /lscratch
62 Supermicro Grace Hopper Compute Nodes (ARM)
2 Supermicro Grace Grace Compute Nodes (ARM)
- 1x NVIDIA H100 GPUs with 96 GB of VRAM
- Coherent GPU RAM up to 480 GB
- 1x Grace CPU;
72 cores each at 2.0GHz - 480 GB RAM
- Dual 100Gb Ethernet network interfaces
- Dual 100Gb NDR100 Infiniband high speed network interfaces
- 1 TB NVMe drives, mounted as /lscratch
- CPU only no GPU
- 2x Grace CPU;
72 cores each at 2.0GHz - 480 GB RAM
- Dual 100Gb Ethernet network interfaces
- Dual 100Gb NDR100 Infiniband high speed network interfaces
- 1 TB NVMe drives, mounted as /lscratch
Using PRISM
Everything you need to know about how to work with ADAPT:
- Environment
- Access
- Tools
- Data Collections
- Storage