US Government plans 2 new 150 Petaflops supercomputers
The Department of Energy Awards $425 Million (339 M€) for Next Generation Supercomputing Technologies, of which $325 are for two new 150 Pflops supercomputers called Summit (in Oak Ridge) and Sierra (in Lawrence Livermore) for scientific and military purposes. The project is called CORAL - for Collaboration of Oak Ridge, Argonne, and Lawrence Livermore.
Both CORAL awards announced today leverage the IBM Power Architecture, NVIDIA’s Volta GPU and Mellanox’s Interconnected technologies to advance key research initiatives for national nuclear deterrence, technology advancement and scientific discovery. Oak Ridge’s new system, Summit, is expected to provide at least five times the performance of ORNL’s current leadership system, Titan. Livermore’s new supercomputer, Sierra, is expected to be at least seven times more powerful than LLNL’s current machine, Sequoia. Argonne will announce its CORAL award at a later time.
▻http://www.energy.gov/articles/department-energy-awards-425-million-next-generation-supercomputing-technolo
Today’s most powerful US supercomputer is the Cray Titan (560 640 cores) with theoretical peak performance of 27.11 Pflops and 17.59 Pflops actual.
Number one worldwide is the Chinese Tianhe-2 (3 120 000 cores ) with 54.9 Pflops theoretical peak and 33.86 Pflops actual.
CORAL will use a hybrid architecture, using Nvidia’s Volta GPU, about 4x more bandwidth (1 Tbps) than the current Nvidia Kepler architecture used by the Titan. Volta has stacked DRAM which is directly mounted on the GPU (reduces footprint and latency).
Hybrid architecture means the combination of both CPU and GPUs; GPUs offer many more simultaneous threads than CPUs, and can be spawned without soliciting the CPU (dynamic parallelism).
This is extensively used in the Cray XK7 Titan:
▻https://www.olcf.ornl.gov/computing-resources/titan-cray-xk7
Titan hybrid architecture employs 16-core AMD Opteron 6274 CPUs (2.2 Ghz) in conjunction with Nvidia Testa K20X GPUs (Kepler architecture) to improve energy efficiency while providing an order of magnitude increase in computational power over Jaguar (=Cray’s previous supercomputer, which was upgraded to the Titan). Titan uses 18,688 CPUs (299,008 cores in total) paired with an equal number of GPUs. It has a total system memory of 710 Tb. The system runs Cray’s UNICOS/lc, which is a Cray’s SUSE-based Linux Environment (CLE).
Cost : 100 million dollar.
The Tianhe-2:
▻http://en.wikipedia.org/wiki/Tianhe-2
The Top 500 classification of supercomputers:
▻http://www.top500.org/lists
Tranformation of the previous Jaguar into the Titan:
▻https://www.youtube.com/watch?v=S8Y77efFW-I