L

Rockfish.jhu.edu is the main cluster at the Hopkins High Performance Computing Center (HHPCC) starting over 17,000 cores (November 2020) and a combined theoretical performance of over 1.7 PFLOPs. The compute nodes use Intel Cascade Lake 6248R processors with HDR100 connectivity. Rockfish also has a GPFS (General Parallel File System) file system with 13 Petabytes total capacity.

The standard compute nodes areĀ  Intel Cascade Lake 6248R, 48 cores per node, 192Gb memory, a 1 TB NMVe, HDR100 Infiniband connectivity, and base frequency of 3.00GHz.

The large memory nodes are Intel Cascade Lake 6248R with 1.5 TB memory and similar features as above.

The HDR100 Infiniband topology is 1:1.15 with 100 Gbps bandwidth. The GPFS file system provides an aggregate bandwidth of 25 GBps (read) and 20 GBps (write).

NodesType  DescriptionTotal coresTheoretical Performance/TFLOPs
368
Regular compute nodesIntel Cascade Lake 6248R, 48cores, 192GB RAM, 3.0GHz, 1TB NMVe local SSD17,6641,695.7
10
Large memory nodesIntel Cascade lake 6248R, 48 cores, 1.5TB RAM480
46.08
TBDGPU nodes

GPFS13 PB usable high performance file system
10ZFS2 PB raw
TOTALS18,144Total combined: 1,741.8 TFLOPs
20Regular compute nodes1,34455.9
5LM
TBD





New TOTALS with Additional hardware23,280Total combined: 1.4PFLOPs

VisibilityPublishAuthor30 RevisionsTemplate:Parent Page:Order