Today Nvidia unveiled accelerator Tesla K40, designed for the use in high-performance computers, servers and supercomputers. At the core of Tesla K40 GPU GK110B lies the maximum number of cores CUDA. Apparently, this is the same revision that is used to create the GeForce GTX 780 Ti .
Tesla K40 performance with single precision - 4,29 TFLOPS against 3,95 TFLOPS in K20, while the double-precision performance increased from 1.31 to 1,43 TFLOPS. In addition to this amount of memory is 12 GB (vs. 6 GB in K20X), and the capacity increased from 250 GB / s to 288 Gb / s.
Tesla K40 is now available for Nvidia servers partners , and in one of the universities of Austin (Texas, USA) in January 2014 will be launched supercomputer Maverick, based on the new accelerators.