Announcement

Collapse
No announcement yet.

LWJGL 3.3 Released For This Popular Java Library - OpenCL 3.0 Added, New Bindings

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by jonix View Post

    Wait, that mean I can run minecraft on my Raspberry Pi now?

    That is interesting
    I was running Java Minecraft on a Pi last year. However there are a bunch of steps to take to get it to run, there are guides.

    And the performance is abysmal because the GPU is so weak (it hasn't made the same gains in the SoC as the CPU has since the device was released).

    Comment


    • #12
      Originally posted by Developer12 View Post
      I don't really care that they're running the latest version of CUDA, or especially OpenCL. (How are games written in java even making use of these anyway? How are they exposed?)

      (Edit: oh dear god. https://github.com/LWJGL/lwjgl3/blob...HelloCUDA.java They're embedding C code into the java app and compiling it.)
      Disclaimer I am not a CUDA dev. That "C code" looks like device kernel (used to create Nvidia PTX IR). You probably can't use CUDA without specifying it to the driver. Sure they could have hidden it away in a seperate file, but IMO it's better to show the nasty part in a single file... at least for basic sample code. AFAIK there's no Java -> PTX ASM. I doubt you could even use the latest C or C++ standard in the PTX compiler. The same goes for OpenCL (take a look at mining software).



      IMO GPU compute is nasty on all platforms in all languages. Hopefully SYCL will change this in the future.

      Comment


      • #13
        Originally posted by Jabberwocky View Post

        Disclaimer I am not a CUDA dev. That "C code" looks like device kernel (used to create Nvidia PTX IR). You probably can't use CUDA without specifying it to the driver. Sure they could have hidden it away in a seperate file, but IMO it's better to show the nasty part in a single file... at least for basic sample code. AFAIK there's no Java -> PTX ASM. I doubt you could even use the latest C or C++ standard in the PTX compiler. The same goes for OpenCL (take a look at mining software).

        https://docs.nvidia.com/cuda/nvrtc/i...p__compilation

        IMO GPU compute is nasty on all platforms in all languages. Hopefully SYCL will change this in the future.
        Not sure what I was expecting to be honest. Probably a set of basic primitive operations like pytorch/numpy will provide, and a way to string them together.

        Dunno how feasible that would actually be to implement though. I guess you could have a java object get passed around and keep track of the operators it went through, then use that to build a pipeline behind the scenes.

        Comment


        • #14
          Originally posted by Alexmitter View Post

          So, when you spend thousands for your closed instruction set closed hardware "open"POWER workstation, your wish is to play Minecraft on it?
          TBH even the biggest mainframes couldn't handle a big Forge server :P

          Comment

          Working...
          X