Announcement

Collapse
No announcement yet.

Linux 5.16 Graphics Performance In Great Shape For AMD Ryzen APUs

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

  • #11
    Originally posted by make_adobe_on_Linux! View Post
    Not for everyone... https://gitlab.freedesktop.org/drm/amd/-/issues/1455 - for many it is terrible
    the memory being 100% is the same with my desktop 6900 xt as well. it use to happen on windows too until amd came out with a newer driver for windows that allowed dynamic memory speeds based on usage. i've been eagerly waiting for them to apply that to linux too but overall i'm content enough for now. its only 10 watts difference at idle from my own testing. but i can understand those with laptops being upset by this as 10 watts on a laptop is pretty big.
    edit:
    i do want to say, linux amdgpu drivers do have some level of memory downclocking though. when the monitor is turned off / goes into sleep mode, the memory speed does drop down and power usage goes down by 10 watts according to my ups readout as well. at least from my own testing with a 6900 xt. and when i compared it to windows, it was the same.
    Last edited by middy; 09 January 2022, 11:45 PM.

    Comment


    • #12
      AMD APUs are really good, even though Ubuntu 21.10 currently has a serious bug for anyone updating to the latest kernel...

      SRU Justification Impact: This does not occur with linux-image-5.13.0-22-generic, but does with linux-image-5.13.0-23-generic. On startup, I get about a 60 second hang, with the following in the kernel dmesg: Jan 4 15:26:36 inspiron-3505 kernel: [ 34.160572] amdgpu 0000:04:00.0: amdgp : failed to write reg 28b4 wait reg 28c6 Jan 4 15:26:56 inspiron-3505 kernel: [ 54.189055] amdgpu 0000:04:00.0: amdgp : failed to write reg 1a6f4 wait reg 1a706 Jan 4 15:27:16 inspiron-3505 kernel: [ ...


      Supposedly fixed in the next kernel version though.

      Comment


      • #13
        Originally posted by zoltanp View Post
        Nice article and nice improvements!

        Has anybody tried to benchmark different desktop environments on AMD APUs ?
        I've experienced the situation that on a Ryzen 7 5700G, running The Witcher 3 game on GNOME3 + wine + dxvk got significantly more FPS that the same game on KDE + wine + dxvk. All this on Ubuntu 21.10.
        how many more fps you get with the new kernel?

        Comment


        • #14
          Originally posted by coder View Post
          Maybe those older laptops & Lenovo IdeaPad 5500U are using single-channel memory? If so, it could be that they're too bottlenecked on memory for the optimizations to show much improvement.
          I have a few of them around me, these laptops use soldered RAM... And seriously kick butt when it comes to graphics performance thanks to dual channel DDR4-3200. So, it's something else.

          Comment


          • #15
            Originally posted by make_adobe_on_Linux! View Post
            Not for everyone... https://gitlab.freedesktop.org/drm/amd/-/issues/1455 - for many it is terrible
            It's an APU : it uses system RAM. Of COURSE it always runs at 100% - system RAM really doesn't like changing clock speed on the fly.
            As to why performance varies from one DE to another, I've noticed something : DEs that use compositing heavily will hit the VRAM something fierce. And since many laptops default to 512 Mb of RAM for an AMD APU and it's not dynamically allocated, you can get some BIIIIG VRAM contention.

            Comment


            • #16
              Originally posted by loganj View Post

              how many more fps you get with the new kernel?
              I have only tested the stock kernel on Ubuntu 21.10, not any newer or custom kernel.

              As FPS change I have observed only that on the default GNOME3 desktop the game runs smoothly on ~high graphics settings (except texture quality on low), while on KDE the game has been shuttering with the same settings; it only got smooth when I configured ~low graphics settings.

              Comment


              • #17
                Originally posted by mitch074 View Post

                It's an APU : it uses system RAM. Of COURSE it always runs at 100% - system RAM really doesn't like changing clock speed on the fly.
                As to why performance varies from one DE to another, I've noticed something : DEs that use compositing heavily will hit the VRAM something fierce. And since many laptops default to 512 Mb of RAM for an AMD APU and it's not dynamically allocated, you can get some BIIIIG VRAM contention.
                What? I don't think you're understanding. There is no reason the memory clock should be pegged to 100%. It makes the laptop way warmer and at the end of that thread there you can see that running mpv with hw acceleration somehow fixes it until the laptop goes asleep or is rebooted, etc. So it clearly doesn't have to, and people with this GPU often saw more and less dramatic stuck memory clocks. So not only does it get much hotter, but it drains your battery faster. And these laptops already have a serious battery drain issue during sleep.

                Comment


                • #18
                  I have in mind, that the APUs had no or bad power management (automatically changing the TDP of CPU and GPU if one is not at full load). Didn't some Valve guys work on this recently, maybe that is a first result?

                  Comment


                  • #19
                    Originally posted by mitch074 View Post
                    I have a few of them around me, these laptops use soldered RAM... And seriously kick butt when it comes to graphics performance thanks to dual channel DDR4-3200. So, it's something else.
                    You have some identical to the specific ones he benchmarked?

                    Comment


                    • #20
                      Originally posted by mitch074 View Post
                      Of COURSE it always runs at 100% - system RAM really doesn't like changing clock speed on the fly.
                      I think Intel was recently touting the ability to do dynamic clockspeed adjustment of system RAM. I forget which product, however.

                      Anyway, so it's not the memory itself, but the CPU's memory controller that didn't traditionally support it.

                      Comment

                      Working...
                      X