NVidia’s new Tesla K20 GPU units power the world’s fastest supercomputer – for now
NVidia has launched the Tesla K20 and K20x GPUs, which provide up to 1.3 Teraflops of processing power, and are the basis of the Titan supercomputer, which has today been confirmed as the world’s fastest, and may be the world’s most energy efficient supercomputer.
The Titan machine, with a reported top speed of 20 Petaflops of scientific computing performance, is at the top of the latest TOP500 list of the world’s fastest supercomputers, with a measured performance of 17.59 Petaflops on the widely-used Linpack benchmark, beating the 16.32 Petaflops of the previous holder, IBM’s Sequoia system
“Kepler will be the next big leap forward in the HPC [high performance computing] world,” said Sumit Gupta, general manager of NVidia’s Tesla line of accelerators for supercomputers, emphasising the new units’ performance and energy efficiency.
“We have entered the era of accelerated computing,” he said, where supercomputers all use GPU attached processors to achieve high speeds. In this world, there are two players – NVidia and Intel, he said, before claiming that NVdia’s Tesla is well ahead of Intel’s Xeon Phi.
NVida’s new flagship Tesla processor, the K20X, offers 1.31 Teraflops, while the K20 runs at 1.16 Tflops – but they can be used cleverly to get even better performance, said Gupta, in “single precision” mode, which uses four bytes of memory.
Running not-so-accurate single precision calculations, the K20X goes at 3.95 Tflops, and scientists can use this in “mixed precision” working to get fast, accurate results, said Gupta. “In mixed precision, you run the application as much as possible in single precision mode, then go to double precision when you are converging on a result.”
The Titan system uses K20X GPUs to carry out 90 percent of its calculations, and achieves an energy-efficiency of 2000 MegaFlops per Watt of power. “We also expect it to beat Sequoia on energy efficiency,” said Gupta. “The K20X uses 235W of power at its peak load, while the K20 uses 225W”.
As well as Titan, the new processors will be in a machine at the Swiss National Supercomputer Centre. “We are taking advantage of NVidia GPU architectures to significantly accelerate simulations in such diverse areas as climate and meteorology, seismology, astrophysics, fluid mechanics, materials science, and molecular biophysics,” said Dr. Thomas Schulthess, director of the centre and professor of computational physics at ETH Zurich.
“The K20 family of accelerators represents a leap forward in computing compared to NVidia’s prior Fermi architecture, enhancing productivity and enabling us potentially to achieve new insights that previously were impossible.”
Do you know all about Green IT? Try our quiz!