2 Replies Latest reply on Nov 18, 2016 1:04 PM by Tugrul_8192bit

    FPGA embedded in for opencl parallel code and gaming acceleration

    Tugrul_8192bit

      Is Intel on a plan to release a cheap desktop or mobile SoC with a fpga module to be harnessed by opencl compilers in future?

       

      It would be very cool thing to run n-body simulations, game physics, fast encryption and similar embarrassingly parallel codes on this module with %100  area efficiency(unlike a normal cpu with %2 area ALU).

       

      Just an extreme condition:

       

      Lets assume I'm playing doom-95 or doom-2016 games, CPU recognizes its repeated pattern and puts its template in fpga fabric and that fpga starts working like a fat optimized core and performs 50x faster than other cores but does only this job. Maybe not only opencl codes or host codes, but graphics pipeline too! When I'm not gaming, it could be turned into a fast cache maybe?

       

      Doable?

       

      Thank you for your time.

       

       

      Doom accelerator ---> triple the fps! so developers don't need to tune inverse square root maybe Fast inverse square root - Wikipedia  and focus on other aspects of game

       

      Cpu-physx expander ---> still cpu physx but turned into 128 mini cores that perform at a gpu-level fps. If running code is multiplication-heavy, they become more multiplicative cores than a balanced traditional general purpose core.

       

      Opencl work stealer ---> learns and forks any opencl workload on runtime dynamically. Does only what code has so no hiccups because of other softwares.