Nvidia today formally introduced its K20 GPU family, supposedly the highest performing and most efficient accelerators ever built. If you recall, these new GPUs were featured in Titan, the world's fastest supercomputer that was unveiled by the U.S. Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL) a couple of weeks ago.
Each of Titan's 18,688 nodes sports a Tesla K20X GPU accelerator, helping the system reach a performance record of 17.59 petaflops in the LINPACK benchmark.
According to Nvidia, the K20X provides tenfold application acceleration when paired with leading CPUs. It offers 3.95 teraflops single-precision and 1.31 teraflops double-precision peak floating point performance. The standard K20 isn't far behind, offering single- and double-precision peak floating performance of 3.52 teraflops and 1.17 teraflops, respectively.
They're also energy efficient. Nvidia claims Tesla 20X accelerators deliver three times higher energy efficiency than previous generation GPU accelerators, further widening the efficiency advantage over CPUs. In large part because of the heavy reliance on Tesla 20X accelerators, Titan was able to achieve 2,145.77 megaflops of performance per watt, Nvidia says.
The Tesla K20 family is shipping today and available for order from leading server manufacturers.