A new supercomputer, the Chinese Tianhe-1A, was unveiled at HPC 2010 in China and has set a new performance benchmark at 2.507 petaflops. It has become the fastest in the world and has taken the title from the Cray XT5, made in US, which had a performance of 1.756 petaflops. The new supercomputer uses 7,168 NVIDIA Tesla M2050 GPUs and 14,336 CPUs combined, which allow much more operations to be performed in parallel.
By using GPUs, the system us able to conserve more energy, therefore the Tianhe-1A uses 4.04 megawatts. Without using the GPUs, the system would need 12 megawatts for the same effect. The difference of power would be sufficient for 5,000 homes for a whole year, according to data from NVIDIA. The performance of the new supercomputer has been measured using the LINPACK benchmark.
Designed by the NUDT (National University of Defense Technology) in China, the Tianhe-1A is fully operational in Tianjin, at the National Supercomputer Center. It will be used as an open access system and will serve for large scale scientific computations.
NVIDIA said that its GPUs power now two of the first three fastest supercomputers in the world. NVIDIA CUDA-based Tesla GPU is made specifically for applications where deep and massively parallel computations are required. They help in tsunami and hurricane modelling, cancer research, drug discovery, car design and many other very complex calculations.
Posts
