Author Topic: NVIDIA Inside Kepler slide  (Read 1703 times)

0 Members and 1 Guest are viewing this topic.

Stefan

  • Global Moderator
  • Hero Member

  • Offline
  • *****

  • 2912
    • View Profile
NVIDIA Inside Kepler slide
« on: June 18, 2012, 06:08:17 PM »
Download here

This slide contains more comprehensive infos about Kepler like:

Quote
New Instruction: SHFL

Data exchange between threads within a warp

    Avoids use of shared memory
    One 32-bit value per exchange



ATOM instruction enhancements

   Added int64 functions to                       2 – 10x performance gains
   match existing int32                                Shorter processing pipeline

                                                       More atomic processors
    Atom Op        int32    int64
                                                       Slowest 10x faster
    add              x        x
                                                       Fastest 2x faster
    cas              x        x

    exch             x        x

    min/max          x        X

    and/or/xor       x        X