Chinese researchers have built the world's fastest supercomputer, which is almost 50 percent more powerful than its predecessor.
The Tianhe-1A, based at the National University of Defense Technology (NUDT) in China, can perform over 2.5 thousand trillion floating point operations per second (petaflops).
Professor Jack Dongarra, who is a member of the Top500 supercomputer ranking project, told ZDNet UK on Friday that he had been present when NUDT tested the Tianhe-1A supercomputer at the National Supercomputer Center in Tianjin last week.
"I can verify this as of today," wrote Dongarra in an email on Friday. "I was at the supercomputer centre last week. They used the Linpack benchmark or a program we have called HPL. The benchmark was run last week. Rmax is 2.507 [petaflops] and Rpeak is 4.701 [petaflops]."
Dongarra is professor of electrical engineering and computer science at the University of Tennessee. Linpack benchmarks, which were developed by Dongarra, measure the floating-point computing power of high-performance systems. HPL is a software package that is an implementation of the Linpack benchmark package. Rmax is the maximum number of Flops measured using the Linpack benchmark, while Rpeak is the theoretical peak performance of the system.
The world's previous fastest supercomputer, the Jaguar Cray XT5 at the Oak Ridge National Laboratory in Tennessee, had an Rmax score of 1.759 petaflops and an Rpeak score of 2.331 petaflops.
The Chinese Tianhe-1A supercomputer uses 7,168 Nvidia Tesla M2050 GPUs and 14,336 Intel Xeon X5670 CPUs, as well as 2,048 NUDT FT1000 heterogeneous processors, according to Intel and Nvidia.
The use of GPUs instead of just CPUs has dramatically lowered the amount of energy used by the computer, said Nvidia. If the Tianhe-1A were built only with CPUs, it would need more than 50,000 CPUs and consume more than 12MW of power per hour. As it is, the Tianhe-1A consumes 4.04MW per hour, Nvidia said in a statement.
"The performance and efficiency of Tianhe-1A was simply not possible without GPUs," said Guangming Liu, chief of National Supercomputer Center in Tianjin, in the statement. "The scientific research that is now possible with a system of this scale is almost without limits; we could not be more pleased with the results."
An Nvidia spokesperson told ZDNet UK that to increase the processing speed, the GPUs used Nvidia's GF100 graphics architecture, and Cuda, a software and hardware architecture that enables the GPU to be programmed with a variety of high-level programming languages.
"Instead of programming dedicated graphics units with graphics APIs, the programmer could now write C programs with Cuda extensions and target a general purpose, massively parallel processor," said the spokesman in an email exchange.
Professor Hans Meuer of the University of Mannheim, who is part of the Top500 project, declined to say whether the Tianhe-1A was now the world's fastest supercomputer. Meuer said in an email that he would not make a pre-announcement about the Top500 list, which is due to be presented at the SC10 conference in New Orleans in November.