by Anton Shilov
11/18/2010 | 12:00 AM
During the SC10 industry trade-show the Lomonosov Moscow State University (MSU) and T-Platforms Group, a leading creator of supercomputers as well as supporting software and services, signed a memorandum of understanding with intention to collaborate on the development of next-generation supercomputer technology aiming for exascale levels of performance.
The agreement signed between T-Platforms and MSU paves way to developing original exascale technology in Russia and sets forth a long-standing technological collaboration. MSU brings to the partnership its vast and unique experience in the development of parallel algorithms, programming tools, and application software, as well as extensive expertise in testing and fine-tuning cutting edge hardware technology. T-Platforms will continue to develop original supercomputing equipment from components up to the system level, along with system management and monitoring software, designed to radically improve power efficiency, augment resilience, and improve manageability of next generation computers.
“It is of vital importance for Russia to compete successfully with other world leaders in the development of next-generation supercomputing technology, as speed of innovation and improved quality of life depend a great deal on this technology. The cooperation of Russian supercomputer industry leaders, both vendors and customers, is a fundamental condition required to successfully meet many challenges we face as we move forward toward exascale levels of computation,” said Victor Sadovnichy, rector of Lomonosov Moscow State University.
T-Platforms and the MSU have a successful track record of teamwork that has resulted in the two largest Russian supercomputers, Chebyshev (2008, 60TFLOPS), and Lomonosov, (2009, 350TFLOPS). The two organizations have partnered on techniques for improving software scalability with tremendous results. One example in particular is a thermal simulation at MSU that achieved 99% computational efficiency (real vs. peak performance) over 1000 nodes.
The two organizations did not reveal any actual solutions that can be utilized to build an ExaFLOPS supercomputer capable of performing at least a quintillion floating point operations per second (1018 FLOPS). Given that there are no world-class developers of processors in Russia at the moment and it is hardly possible that they will show up in the next ten years, it is highly likely that the first ExaFLOPS supercomputer in Moscow will use industry-standard chips. Many experts claim that exascale supercomputers will be heterogeneous and will utilize both traditional central processing units as well as special highly-parallel accelerators, such as AMD FireStream and Nvidia Tesla. As a result, the main tasks of the MSU and T-Platforms will be to create appropriate software, infrastructure and implementation of chips developed by current HPC market leaders, such as AMD, Intel and Nvidia.
“It has been demonstrated that the quality of technological breakthroughs achieved through computer simulations is in direct proportion to the performance level of available supercomputers. However, evolutionary development of today’s technology cannot provide substantial performance increases. We need a revolutionary leap to achieve this. All our efforts are now directed towards developing next decade technology, and the efficient use of this technology heavily depends on input from leading customers and software developers, such as Lomonosov Moscow State University," said Vsevolod Opanasenko, the chief executive officer of T-Platforms.