China is unveiling a new supercomputer on Thursday that incorporates thousands of graphics chips and can reach a sustained performance of 2.5 petaflops, making it one of the fastest systems in the world.
Located at the National Supercomputing Center in Tianjin, the Tianhe-1A supercomputer has 7,168 Nvidia Tesla M2050 GPUs (graphics processing units), each with 448 processor cores. It also has 14,336 six-core Intel Xeon CPUs. The supercomputer was built by China's National University of Defense Technology and is "the fastest system in China and in the world today," Nvidia claimed.
Besides a sustained performance of 2.5 petaflops measured by the Linpack benchmark, it has a theoretical performance of 4.669 petaflops when all the GPUs are operational, according to an Nvidia spokesman. The benchmarks were provided by the National Supercomputing Center in China, he said.
China has been moving up the supercomputing ranks in recent years. The last Top500 list of the fastest supercomputers, issued in June, lists the Nebulae supercomputer in Shenzhen as the world's second fastest. That system also combines Intel CPUs and Nvidia GPUs.
The fastest supercomputer in the world according to the June list is the Jaguar system at the U.S. Department of Energy's Oak Ridge National Laboratory, which can deliver 1.76 petaflops of sustained performance.
The Tianhe-1A was announced two weeks before the release of the next Top500 list, so it is too early to say if it will be the fastest system on the list. China is looking to boost its computing resources and is doing a lot with hybrid supercomputer designs, said Nathan Brookwood, principal analyst at Insight 64.
A number of supercomputers have combined GPUs with CPUs to boost performance. GPUs are specialised co-processors that are faster than traditional CPUs at executing certain tasks, like those used in scientific and mathematical applications.