AMDXDNA Driver For Ryzen AI Now Ready To Appear In The Linux Kernel

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

  • JRepin
    replied
    Originally posted by Errinwright View Post
    Will video upscaling be supported for media use-cases?
    I think Video Processing Engine (VPE) / Video Core Next (VCN) in AMD APUs/GPUs are more efficient and specialized for this use-case and I would say already supported by the kernel and other media libraries/frameworks/players.

    Leave a comment:


  • EliasOfWaffle
    replied
    Originally posted by Adarion View Post
    A bit peculiar to put this under DRM, but anyway, nice to have. Not that I have one of these CPUs, yet. And yes, it would have been nice at launch day, but hopefully people will now have a nicely reviewed driver. Now it would of course be fine if we had means to see what calculations could be offloaded to this ASIC and software that would be using it once available.
    DRM is means Direct Rendering Manager, it's a kernel subsystem like a infrastructure for writing drivers sharing common interface, abstractions and code, DRM is used for graphics resources like gpu, DC but its too used for offloading to heterougeneous hardwares.

    Accel is a new subsystem design to be more focused with NNA (Neural Network Accelerators) but it's in DRM subsystem.
    Linux DRM doesn't means about some restriction tecniques for protection content, i know, the name looks much but is not the same thing.

    Leave a comment:


  • nuetzel
    replied
    Maybe Zen 6 comes with NPU.
    It's time.

    Leave a comment:


  • Ferrum Master
    replied
    Originally posted by loganj View Post
    is there even an app that use this (or any AI) on linux? something similar to windows at least.
    You can use tensorflow libs for PhotoPrism as pretty basic thing, if you manage it get them working. I kinda avoid some piece of software that haven't managed to put delete all buttons where they should be, basically everywhere.

    Leave a comment:


  • David Huang
    replied
    Originally posted by oleid View Post
    Sadly, llama.cpp doesn't support NPUs, yet, AFAIK.
    There's a Xilinx fork of llama.cpp in the RyzenAI-SW repo that supports running on NPU. Unfortunately its rather PoC-style rather than a mature implementation, and its device kernel was not open source, so it's hard to contribute to the project and improve the quality.

    Leave a comment:


  • loganj
    replied
    is there even an app that use this (or any AI) on linux? something similar to windows at least.

    Leave a comment:


  • Errinwright
    replied
    Will video upscaling be supported for media use-cases?

    Leave a comment:


  • oleid
    replied
    Sadly, llama.cpp doesn't support NPUs, yet, AFAIK.

    Leave a comment:


  • euduvda
    replied
    Originally posted by Adarion View Post
    And yes, it would have been nice at launch day, but hopefully people will now have a nicely reviewed driver. Now it would of course be fine if we had means to see what calculations could be offloaded to this ASIC and software that would be using it once available.
    I have one of these and I'm curious for what uses it will really have!

    Leave a comment:


  • Michael
    replied
    Originally posted by Adarion View Post
    A bit peculiar to put this under DRM, but anyway, nice to have. Not that I have one of these CPUs, yet. And yes, it would have been nice at launch day, but hopefully people will now have a nicely reviewed driver. Now it would of course be fine if we had means to see what calculations could be offloaded to this ASIC and software that would be using it once available.
    It's under DRM because the accelerator "accel" subsystem is under the DRM umbrella due to commonality with GPUs/drivers.

    Leave a comment:

Working...
X