Skip to main content

GPU with AI cores

Comments

6 comments

  • Rick Sammartino Community moderator

    I've never seen any mention of AI cores from ON1. These are the official specs,

    0
  • Matt Dean

    From On1's website:
    "With our 2023 generation of products, we take advantage of some of these new technologies including Vulkan and DirectML"

     

    From https://www.intel.com/content/www/us/en/developer/articles/technical/effective-deployment-ai-workloads-with-xe.html
    "DirectML efficiently executes the individual layers of your inference model on the GPU (or on AI-acceleration cores, if present)"

    0
  • Brian Lawson Community moderator

    If those AI cores are anything like Apple's Neural chip, they are for creating the models used by the AI and not for putting that AI to work.

    1
  • Don Maclean

    They probably mean Compute Units rather than AI cores.

    0
  • Carl Traub

    I've seen rumors that AI cores might be included in some forthcoming higher end laptop APUs, but I am not aware that they are currently available in any consumer grade CPU/GPU/APU from AMD.

    Use of any such cores may also require coding changes specific to the AMD AI cores or drivers, and it doesn't seem that On1 focuses very much on AMD hardware.  Up until a few months ago, the only AMD GPUs listed on their supported hardware page were 5+ years old.

    0
  • Matt Dean

    The RX 7000 series cards have AI Accelerator Cores, yes they are compute cores but are either dedicated or switchable based on the need, which I don't know. I also don't know what types of workloads can utilize the cores. I've never seen any explanation about whether they are only used for creating the models like Brian mentioned with the Apple chips.

    From AMD's own site:

    Break through new levels of performance with up to 24 GB of GDDR6 memory and blazing fast clock speeds for an incredible gaming experience. Radeon™ RX 7000 Series graphics feature advanced AMD RDNA™ 3 compute units, with second-generation raytracing accelerators and new AI accelerators to deliver remarkable performance while maximizing graphical fidelity.

    Since PR uses DirectML which has support for the AMD AI tech, then it should work even if PR was not programmed to specifically take advantage of AMD AI tech. I am not a programmer, so I could be wrong. Also, if the AI tech is only for creating models, then it doesn't matter and wouldn't be used whether it's an AMD or Nvidia card.

    0

Please sign in to leave a comment.