SAN FRANCISCO--The Tokyo Institute of Technology’s Global Scientific Information Computing Center (GSIC) has won the the supercomputing industry’s highest honor, the Gordon Bell Prize, with its Nvidia GPU-accelerated Tsubame 2.0 supercomputer. The award typically recognizes achievements by researchers using parallel computing to push the scientific envelope.
The Tsubame 2.0, made up of HP ProLiant SL390 servers powered by Intel’s six-core Xeon 5650s, gets an extra boost from Nvidia’s Tesla GPU cards, which take care of the vector math, freeing up large portions of the CPUs.
The Tsubame 2.0 uses just 1.2MW of power to reach its 2.0 petaflops of performance, making it 3.4 times more efficient than a similar system at Los Alamos National Laboratory using an x86-only Cielo system from Cray. It has already been voted the most efficient supercomputer on the Green500 list when it comes to pushing past petaflops levels.
Professor Takayuki Aoki’s research group at the GSIC won the award for “Special Achievement in Scalability and Time-to Solution” with the committee particularly emphasizing the achievement of reaching 2.0 petaflops of performance on a practical research application in single precision.
The application itself was designed to simulate the behavior of metal alloy microstructures called dendrites. The simulation allowed researchers to identify lighter, stronger metal materials which could be used for the development of more fuel-efficient cars. Previous attempts to simulate complex dendrite microstructures were apparently limited by available performance.
Aoki said the project was “one of many” his team was working on which harnessed the GPU to achieve a performance boost.
The $10,000 award was presented by the Association for Computing Machinery in conjunction with the Institute of Electrical and Electronics Engineers.