Announcement

Collapse
No announcement yet.

AMD Submits Initial AMDGPU Graphics Driver Improvements Ahead Of Linux 5.1

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

  • Marc1n
    replied
    Any chances we will get proper reset mechanism for VM's for Polaris and Vega cards? It's silly that i have to reboot my host every time i run VM because when i try to launch it second time - vm hangs. Also, support for proper amdgpu driver unloading would be usefull so my Polaris GPU can be switched between host an VM.
    Now, it's time to test if 5.0-rc4 kernel fixes problems with kms on my Ryzen APU introduced in 4.20 series...

    Leave a comment:


  • tildearrow
    replied
    Originally posted by Marc Driftmeyer View Post
    No one truly cares as 99% of the work is for Vega 10 or newer.
    It's more like the inverse... People do not care much about Vega, and tha

    *display freezes for 10 seconds*

    t

    *display freezes for 10 seconds*

    's

    *display freezes for 10 seconds*

    *display freezes for 10 seconds*

    wh

    *enough of this freezing! reboots system because they don't even have a proper GPU recover mechanism*

    Originally posted by Marc Driftmeyer View Post
    No one truly cares as 99% of the work is for Vega 10 or newer.
    It's more like the inverse... People do not care much about Vega, and that's why it has some bugs that really disturb me. For example,- wait, I'll be back.

    *turns off monitor*

    *takes a cup of coffee, then turns on monitor*

    *no picture*

    *argh, screw this. When this happens there is no way out, other than rebooting, and therefore losing my work!*

    Leave a comment:


  • L_A_G
    replied
    Originally posted by davidbepo View Post
    this should improve performance, right?
    DCC is a technique for improving effective memory bandwidth and from what I can tell this is improvement to how it can be used. AMD's lower end cards are obviously going to be the main recipients of any performance benefit due to the higher end ones having massive memory bandwidth thanks to HBM and the massively wide buses that go with it (2048 lanes when planar memory very rarely goes beyond 384, 256 being the most common bus width). Not that there aren't cases where even HBM equipped cards become memory bandwidth constrained, 4k gaming being the obvious use case.

    Leave a comment:


  • mibo
    replied
    I'm looking forward to NAVI hoping for day1 Linux driver support (that rx590 episode hopefully woke up everybody involved).

    Leave a comment:


  • davidbepo
    replied
    - Support for Delta Color Compression (DCC) on scanout surfaces.
    this should improve performance, right?

    Leave a comment:


  • LeJimster
    replied
    Originally posted by monte84 View Post

    Well its nice to be no one
    Another no one reporting in. =)

    Leave a comment:


  • monte84
    replied
    Originally posted by Marc Driftmeyer View Post
    No one truly cares as 99% of the work is for Vega 10 or newer.
    Well its nice to be no one.

    Worthless post is worthless.

    Leave a comment:


  • Marc Driftmeyer
    replied
    No one truly cares as 99% of the work is for Vega 10 or newer.

    Leave a comment:


  • schmidtbag
    replied
    That's actually a lot of cool changes.
    I forget - have we seen benchmarks of PowerPlay vs the old DPM?

    Leave a comment:


  • mihau
    replied
    PCIe bandwidth usage would be particularly useful for debugging external GPU bottlenecking. I don't know how many eGPU on Linux users there are, but I'm glad to see this feature added. I actually don't think the Windows driver exposes this (nvidia has had it since forever on both OSes, though).

    Leave a comment:

Working...
X