Announcement

Collapse
No announcement yet.

CUDA on AMD?

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

  • CUDA on AMD?

    so, you guys with your skillz have enabled cuda on cpu for hybrid rendering... any chance you could do the same for AMD gpus? a custom driver or something...? im sure its probably illegal even if possible. cant imagine nvidia being very happy.


    however given that there is a nice dual 16gb amd gpu which would probably match a 1080TI in cuda, while running in the slower opencl, and which costs only $999, it seems running cuda it would beat the pants off the nvidia options.

    somebody needs to do something, nvidia has such a stranglehold on gpus (particularly for compute) that they can (and do) charge what they want... which kills the value proposition of higher end cards.

  • #2
    Why not just improve and port more features to OpenCL instead? Is it more difficult than coding for CUDA? Or maybe it's slower than CUDA and not worth coding for it? I've looked at a few AMD compute cards and on paper they have pretty nice specs - core count in the 4 thousands and everything so they should, in theory, be pretty fast for compute, no?
    Aleksandar Mitov
    www.renarvisuals.com
    office@renarvisuals.com

    3ds Max 2023.2.2 + Vray 7 Hotfix 1
    AMD Ryzen 9 9950X 16-core
    96GB DDR5
    GeForce RTX 3090 24GB + GPU Driver 566.14

    Comment


    • #3
      We share almost the same major code paths between our OpenCL and CUDA engines. It is not about the API

      Best,
      Blago.
      Last edited by savage309; 11-06-2018, 02:19 PM.
      V-Ray fan.
      Looking busy around GPUs ...
      RTX ON

      Comment


      • #4
        Originally posted by Alex_M View Post
        Why not just improve and port more features to OpenCL instead?
        OpenCL is pretty much dead for the future. Even Apple - the company that originally proposed the standard - is now recommending Metal instead. I've heard that AMD is working on their own compute programming stack too.

        Is it more difficult than coding for CUDA?
        Yes. OpenCL does not support all of the features that we need and as a result we don't support everything on OpenCL.

        Or maybe it's slower than CUDA and not worth coding for it?
        Yes to that too. On NVidia cards, OpenCL runs considerably slower than CUDA. We never managed to get good performance out of AMD hardware either. And the vast majority of our customers have NVidia cards anyways.

        Best regards,
        Vlado
        I only act like I know everything, Rogers.

        Comment

        Working...