By Matthew Dublin
The National Center for Supercomputing Applications announced last week the roll-out of a new 153 teraflop supercomputer dubbed Forge. The new system has a "hybrid" architecture that combines GPUs and CPUs, and is set to replace NCSA's previous hybrid system called Lincoln.
Forge will combine 18 Dell blades that contain 36 nodes of dual-socket/eight-core AMD processors, Nvidia GPUs — including eight Fermi GPU units for each node, for a total of 288 —and will use an InfiniBand interconnect fabric. The system will have 700 terabytes of file system space with an I/O bandwidth its designers hope will surpass 16 GB per second at full operational performance.
The new cluster will be housed in a cutting-edge 20,000-square-foot machine room at the University of Illinois' National Petascale Computing Facility.
Forge's predecessor, Lincoln, was NCSA's first foray into large-scale hybrid core computing and its track record so far has validated the concept of hybrid-core cluster computing for life sciences research. Klaus Schulten's research team at the University of Illinois at Urbana-Champaign, used Lincoln to run NAMD to study the organization and function of proteins and protein complexes within cells. Schulten's team found that two of Lincoln's GPUs were equivalent of 24 of Lincoln's CPU cores, while eight of its GPUs were equivalent to 96 CPU cores. And the University of Utah's Thomas Cheatham, who is studying how proteins behave in solutions and how drugs interact with them, was able to accelerate AMBER on the system, achieving 15 times speedup per node.
As of July 1st, the new system will be allocated through the National Science Foundation Translational Research in the Academic Community process.