The U.S. Department of Defense is making a big investment in data analytics and AI computing with the procurement of two HPC systems that will provide the High Performance Computing Modernization Program (HPCMP) with a combined 15 petaflops of computing power (peak double-precision).
Liqid Inc., the Broomfield, Colorado-based provider of “composable infrastructure,” was awarded the nearly $32 million contract for hardware acquisition and software maintenance services after a competitive bidding process, according to a public notice.
The systems will be deployed at the U. S. Army Research Laboratory (ARL) DoD Supercomputing Resource Center (DSRC) — at Aberdeen Proving Ground, Maryland — where they will support both traditional HPC and data analytics workloads across the DOD complex. The HPCMP indicated the systems will enter production in mid fiscal 2021 (roughly March or April 2021).
The two systems – “Jean” and “Kay” — both feature 48-core Intel Xeon Cascade Lake-AP processors, Nvidia A100 Ampere GPUs, 200 Gbps InfiniBand interconnection, and large solid-state file systems. Jean incorporates 1,202 Cascade Lake-AP CPUs, 280 A100 GPUs, 323 terabytes of memory and 12.5 petabytes of usable NVMe-based storage. While Kay will be outfitted with 1,010 Cascade Lake-AP CPUs, 76 A100 GPUs, 240 terabytes of memory, and 10 petabytes of usable NVMe storage.
“These systems significantly enhance the Program’s capability to support the Department of Defense’s most demanding data-intensive computational challenges, and include significant capabilities for artificial intelligence, data analytics, and machine learning. The systems include embedded capabilities to support persistent services in additional to traditional batch-oriented processing,” the HPCMP stated in an announcement made earlier this month.
The systems are being supplied by composable infrastructure specialist Liqid. “Liqid’s composable solutions reduce the cost of deployment by optimizing the ratio of GPUs to CPUs and dynamically changing such ratios as needed, significantly improving the total cost of ownership for high-density computing environments,” the company said of its composable system solution. “The composable model enables GPUs to be incorporated into compute nodes on the fly to take maximum advantage of these powerful compute accelerators through software-defined technology.”
The High Performance Computing Modernization Program was established in 1992 with a mandate to modernize the DoD’s high-performance computing program. The HPCMP funds and oversees the operation of five DoD Supercomputing Resource Centers (DSRCs) – AFRL, ARL, ERDC, Navy, and MHPCC. The U.S. Army Corps of Engineers, Engineering and Support Center, in Huntsville, Alabama, awarded the $31,850,000 supercomputing contract to Liqid.
Two other new HPCMP systems – Narwhal and Warhawk – are slated to be ready for use in early 2021 (calendar year). Both are HPE Cray systems that leverage AMD Epyc Rome processors and Nvidia V100 GPUs, connected by an HPE Slingshot network. Narwhal, a 12.8 peak petaflops system, will be installed at the Navy DSRC in Stennis Space Center, Mississippi; while Warhawk, a 5.1 petaflops peak system, is headed for the AFRL DSRC in Ohio.
All four upcoming systems — Jean, Kay, Narwhal and Warhawk — will address physics-based, AI, and ML applications for DoD users. Altogether, they will bring HPCMP’s aggregate peak capacity to 82 peak double-precision petaflops.