updated 05:15 pm EDT, Thu October 28, 2010
Tianhe-1A supercomptuer fastest at 2.5 petaflops
A new Chinese supercomputer, the Tianhe-1A was revealed today at HPC 2010 in China has set a new performance benchmark, at 2.507 petaflops. In doing so, it has become the fastest supercomputer in the world, taking the title from the US-based Cray XT5, which managed a performance of 1.756 petaflops. The Tianhe-1A uses 7,168 NVIDIA Tesla M2050 GPUs and 14,336 CPUs, which combined lets it calculate much more in parallel.
The use of the GPUs helps the system conserve energy, as the Tianhe-1A uses just 4.04 megawatts. Without the GPUs, it would require 12 megawatts of power for the same effect. The difference would power 5,000 homes for one year, according to NVIDIA. The performance was measured using the LINPACK benchmark.
Tianhe-1A, designed by the National University of Defense Technology (NUDT) in China, is fully operational at the National Supercomputer Center in Tianjin. The supercomputer will be used as an open access system to use for large scale scientific computations.
NVIDIA added that its GPUs now power two of the three fastest supercomputers in the world.
NVIDIA CUDA-based Tesla GPUs are made specifically for environments where deep, massively parallel math is required. They currently help in drug discovery, hurricane and tsunami modeling, cancer research, car design and other complex calculations and computing.