The new November TOP500 list was released this morning with a supercomputer called Titan atop of it. Titan resides in Oak Ridge National Laboratory in Oak Ridge, Tennessee and has wrested the top spot from the Lawrence Livermore National Laboratory’s Sequoia system. The Titan supercomputer was measured at a record performance of 17.59 petaflops.
NVIDIA is particularly proud with the new Titan supercomputer news as it is “armed with 18,688 NVIDIA Tesla K20X GPU accelerators” made up of 261,632 cores (14 cores per accelerator), which deliver 90 per cent of the performance of this supercomputer. Another 18,688 AMD 16-core Opteron 6274 CPUs provide the Cray XK7 design Titan supercomputer a total of 560,640 cores. The NVIDIA K20X GPUs are doing an impressive job by any measure.
NVIDIA says that the Tesla K20X provides “the highest computing performance ever available in a single processor, it provides tenfold application acceleration when paired with leading CPUs. It surpasses all other processors on two common measures of computational performance - 3.95 teraflops single-precision and 1.31 teraflops double-precision peak floating point performance.” In a supercomputer it needs to be paired with a general purpose CPU to work its magic.
NVIDIA have a big lead over AMD in applying GPU computing to supercomputers says ComputerWorld. However AMD recently claimed that its Fire SM10000 card delivers 1.48 teraflops peak double-precision performance which bests the NVIDIA K20’s 1.31 teraflops. At this time the parallel programming tools for NVIDIA supercomputer solutions are much more mature.
Looking more at the TOP500 list we can see there is a definite trend in utilising Accelerator/Co-Processor technology. Six months ago 58 of the top supercomputers were using GPU/CPU and now the number has grown to 62. Intel is still the dominant processor supplier and supplies processors that are in 76 per cent of TOP500 systems. The UK is the country with the most supercomputers in Europe.