According to NVIDIA GPU idea GP100 - senior decision in a line of media Pascal architecture - finds primary use in acceleration calculation . One of the first new GPU will receive a French research centers, plans to complete the deployment of a supercomputer based on Pascal-generation accelerators to the middle of this year.
3DCenter NVIDIA published materials, which contain information about the performance level of supercomputer GP100 in operations on floating point numbers. So, with the double precision processor can perform four trillion operations per second (teraflops 4), which is three times exceeds the capabilities of Tesla K20 accelerator based GPU GK110. With the number of single precision index increased to 12 teraflops, three times greater than those measured GK110, and almost twice - GM200.
Also, the slide makes it clear that the GP100 will be equipped with memory type HBM2, the capacity of which amounts to 1 Tbit / s. Related Products :
|