The US Department of Defense will install seven new supercomputers at the Air Force Research Laboratory (AFRL) and Navy DoD Supercomputing Research Centers (DSRCs), with a combined capacity of 14 petaflops.

The purchase of the HPE SGI 8600 systems brings the total computing power of the DoD’s High Performance Computing Modernization Program (HPCMP) to 47 petaflops. 

That’s a lot of Skylake

The AFRL DSRC in Dayton, Ohio, will receive four HPE SGI 8600 systems based on Intel Xeon Platinum 8168 (Skylake) processors:

  • A single system with 56,448 Skylake cores, 24 Nvida Tesla P100 GPUs, 244 terabytes of memory and 9.2 petabytes of usable storage;
  • A single system with 13,824 Skylake cores, 58 terabytes of memory and 1.6 petabytes of usable storage;
  • Two systems, each with 6,912 Skylake cores, 30 terabytes of memory and 1 petabyte of usable storage.

The Navy DSRC at Stennis Space Center, Mississippi, will receive three HPE SGI 8600 systems based on the same Xeon Platinum:

  • Two systems, each with 35,328 Skylake cores, 16 Nvida Tesla P100 GPUs, 154 terabytes of memory and 5.6 petabytes of usable storage;
  • A single system with 7,104 Skylake cores, four Nvida Tesla P100 GPUs, 32 terabytes of memory and 1 petabyte of usable storage.

The HPE SGI 8600 is a liquid cooled, high-density clustered computer system originally announced in June 2017. 

The seven systems for the DoD, which are set to enter production service in the second half of 2018, mark HPCMP’s last investment of the fiscal 2017. The Department of Defense Appropriations Act of 2017 set  the program’s budget at $222.19 million.

In a 2016 presentation, Dr Reed Mosher of the US Army’s Information Technology Laboratory (ITL) explained the DoD HPCMP footprint at the time:

DoD HPCMP
– DoD

In addition to AFRL and the Navy DSRC, the program includes the Army Research Lab in Maryland, the Engineer Research and Development Center in Mississippi and the Maui High Performance Computing Center in Hawaii.

The presentation outlines the HPCMP’s ambitions to have a 100 petaflop system by 2025, a cognitive production system in 2026, an exaflop system in 2031 and a 10 exaflop system and a quantum pilot in 2036. Finally, in 2040, it hopes to have a quantum production system.

Last year, AFRL announced that it would partner with IBM on a ‘brain-inspired’ high performance computing system powered by a 64-chip array of the IBM TrueNorth Neurosynaptic System. The company claims that the system’s pattern recognition and sensory processing power will equal the performance of 64 million neurons and 16 billion synapses, while the processor will require just 10 watts of power.