by Anton Shilov
10/29/2010 | 02:44 PM
On Thursday the National University of Defense Technology (NUDT) in China revealed its improved supercomputer called Tianhe-1A that achieves peak performance of 2.5PetaFLOPS as measured by the LINPACK benchmark. The system is the fastest in China and in the world at the moment. What is noteworthy is that Nvidia Corp.'s Tesla powers the new supercomputer.
Tianhe-1A epitomizes modern heterogeneous computing by coupling massively parallel GPUs with multi-core CPUs, enabling significant achievements in performance, size and power. The system uses 7168 Nvidia Tesla M2050 compute cards and 14 336 central processing units (CPUs); it would require more than 50 000 CPUs and twice as much floor space to deliver the same performance using microprocessors alone.
According to Nvidia, a 2.507PetaFLOPS system built entirely with CPUs would consume more than 12 megawatts (MWs). Thanks to the use of GPUs in a heterogeneous computing environment, Tianhe-1A consumes only 4.04MW, making it 3 times more power efficient.
"The performance and efficiency of Tianhe-1A was simply not possible without GPUs. 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," said Guangming Liu, chief of National Supercomputer Center in Tianjin.
Tianhe-1A was designed by the National University of Defense Technology (NUDT) in China. The system is housed at National Supercomputer Center in Tianjin and is already fully operational. The Tianhe-1A supercomputer will be operated as an open access system to use for large scale scientific computations.
"GPUs are redefining high performance computing. With the Tianhe-1A, GPUs now power two of the top three fastest computers in the world today. These GPU supercomputers are essential tools for scientists looking to turbocharge their rate of discovery," said Jen-Hsun Huang, president and chief executive officer of Nvidia.
Nvidia Tesla C2050 and C2070 computing processor boards are single-chip cards with 3GB and 6GB (respectively) of on-board GDDR5 memory (with ECC enabled, user available memory will be 2.625GB for a C2050 and to 5.25GB for a C2070) with 384-bit interface operating at 1.5GHz. Nvidia Tesla 2000-series deliver up to 515GFLOPS of double-precision peak performance and are based on the GF100/T20 chips with 448 stream processors operating at 1.15GHz.
Nvidia Tesla 2000-series cards and compute modules are now available from all major suppliers of systems for high-performance computing, including Cray, Dell, HP, IBM, SGI and others.