Announcement

Collapse
No announcement yet.

Linux Appears To Lose Some Gaming Marketshare With Valve's October Stats

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

  • #51
    Originally posted by bridgman View Post
    Have you tried vanilla Ubuntu 16.10 ? That has pretty new drivers and seems to work well out of the box for most users.



    Can you be a bit more specific about the problems you are seeing ?

    Thanks,
    John
    Well, there are multiple problems, but the first major issue with all kernel/AMDGPU Pro releases is trying to get the dkms module to compile. I'm running Ubuntu 16.04, but have tried various kernels up to 4.8.5. I always start with patches I find on Arch Linux or Gentoo sites, and then scour the Internet for other patches hoping someone has fixed the myriad of compilation/stability problems. The basic procedure I use is as follows:

    1. Back up my root directory
    2. Install a kernel from the Ubuntu PPA.
    3. Try installing AMDGPU Pro without compiling the kernel with CIK enabled.
    4. That has always failed with dkms errors or failures to load the driver, so I search for any way to fix it without compiling the kernel.
    5. If I can get the dkms to compile I try getting amdgpu to load by banning Radeon, creating custom Xorg files, etc. I also try without the dkms module.
    6. If that doesn't work compile a kernel with CIK enabled.
    7. Repeat steps 4 and 5 again.

    On the few occasions I have gotten AMDGPU Pro to compile and load, I'll encounter various graphical errors. The most common one is the screen having streaks of pink, sometimes always and sometimes only while playing games under Wine. I try various versions of wine by the way, including wine-staging. Occasionally I'm simply met by a black screen and have to restore the root directory. I've also tried just running the driver without the dkms module but that always results in the black screen of death. And note that while I have gotten amdgpu to load with AMDGPU Pro a few times, I've never gotten it load by itself in the open source amdgpu configuration. My system always reverts to radeon.

    So the whole thing is a complete and utter mess. Thankfully I'm an embedded systems designer so I can understand and carry out the flabbergasting number of suggested fixes I come across, but at this point it's simply become a challenge to see if I can get anything to work at all. Basically when I have an extra eight hours or so I give it a try. I'd estimate I've given it a good 10 or 12 attempts so far, the last time with kernel 4.8.5 and AMDGPU Pro 16.40.

    Comment


    • #52
      Originally posted by muncrief View Post
      Well, there are multiple problems, but the first major issue with all kernel/AMDGPU Pro releases is trying to get the dkms module to compile. I'm running Ubuntu 16.04, but have tried various kernels up to 4.8.5. I always start with patches I find on Arch Linux or Gentoo sites, and then scour the Internet for other patches hoping someone has fixed the myriad of compilation/stability problems.
      Ahh, OK. Step #2 is the problem.

      The AMDGPU-PRO packages in 16.40 are designed for installation on a stock 16.04 kernel. The DKMS package replaces the built-in amdgpu driver module and installs one that is built with CIK support enabled (and DAL as well). All you should need is step #3.

      Any chance you can get back (or close) to a stock 16.04, at least in terms of graphics bits ?

      There are people running around the internet telling you that you need to rebuild the kernel when using AMDGPU-PRO, please ignore them.

      That said, if you *do* need to use a much newer kernel for some unrelated reason it's probably best to stay with the all-open stack for now. We are building up the kernel compatibility layer (KCL) that allows the DKMS package to compile against a number of different kernel versions, but first priority has been supporting older kernels to allow AMDGPU-PRO use with enterprise distros.
      Last edited by bridgman; 03 November 2016, 11:41 AM.
      Test signature

      Comment


      • #53
        Originally posted by anda_skoa View Post
        Only if your base value is 100 millions.

        For Tuxee's example of 1 million, 0.04% is 400.

        Cheers,
        _
        Oh... of course I can take a base of 100 users and say that 0.0004 user can be ignored.

        Cheers

        Comment


        • #54
          Originally posted by bridgman View Post

          Ahh, OK. Step #2 is the problem.

          The AMDGPU-PRO packages in 16.40 are designed for installation on a stock 16.04 kernel. The DKMS package replaces the built-in amdgpu driver module and installs one that is built with CIK support enabled (and DAL as well). All you should need is step #3.

          Any chance you can get back (or close) to a stock 16.04, at least in terms of graphics bits ?

          There are people running around the internet telling you that you need to rebuild the kernel when using AMDGPU-PRO, please ignore them.

          That said, if you *do* need to use a much newer kernel for some unrelated reason it's probably best to stay with the all-open stack for now. We are building up the kernel compatibility layer (KCL) that allows the DKMS package to compile against a number of different kernel versions, but first priority has been supporting older kernels to allow AMDGPU-PRO use with enterprise distros.
          First of all, thank you for trying to help, I really appreciate it. However I always attempt installing AMDGPU Pro with the stock 16.04 kernel first.

          And just to double check I did it again about half an hour ago after seeing your response. The current kernel is 4.4.0-45-lowlatency, and the AMDGPU Pro was amdgpu-pro-16.40-348864.tar.xz. With this iteration of kernel and driver the dkms compiles, but upon reboot there's simply a black screen. To make matters worse the fallback 4.4.0-45-generic kernel wouldn't boot either, so I had to restore my old root with rsync. By the way I've experimented with both the generic and low latency kernels in the past but this never affected any problems I encountered.

          In any case, restoring root only took about half an hour because after many years I have recovery down to an exact science, but for other users it would be disastrous. I didn't try to do any debugging this time because I need to get some work done, but as I said earlier it would have involved a lot of searching, reconfiguring, and recompiling.

          Keep in mind that I have an R9 390, which is a GCN 1.1 card. As far as I know CIK has to be enabled in the kernel to make it work, but as I said I've done that numerous times before and there are still various problems.

          Comment


          • #55
            Originally posted by muncrief View Post
            However I always attempt installing AMDGPU Pro with the stock 16.04 kernel first.
            Thanks. That's what I figured, but I didn't see you explicitly say it and there is a *lot* of bad advice washing around so I wanted to check.

            Originally posted by muncrief View Post
            And just to double check I did it again about half an hour ago after seeing your response. The current kernel is 4.4.0-45-lowlatency, and the AMDGPU Pro was amdgpu-pro-16.40-348864.tar.xz. With this iteration of kernel and driver the dkms compiles, but upon reboot there's simply a black screen. To make matters worse the fallback 4.4.0-45-generic kernel wouldn't boot either, so I had to restore my old root with rsync. By the way I've experimented with both the generic and low latency kernels in the past but this never affected any problems I encountered.
            OK... I'm pretty sure our testing uses the generic kernel (that's the stock one off a fresh install, isn't it ?). I'm wondering if there is something specific about your card/display the DAL code in the -pro driver isn't handling properly. Is there anything unusual about display connection, ie going through a KVM or anything ?

            What connection type & display res are you running ?

            Are there any other graphics cards installed in the system with displays connected, eg the integrated GPU on your CPU/APU ?

            If you do a fresh install and use the built in open source drivers does the display come up OK ?

            Originally posted by muncrief View Post
            In any case, restoring root only took about half an hour because after many years I have recovery down to an exact science, but for other users it would be disastrous. I didn't try to do any debugging this time because I need to get some work done, but as I said earlier it would have involved a lot of searching, reconfiguring, and recompiling.
            I'm glad you are able to recover but as you say it shouldn't be that hard. We install and test on a lot of different systems before releasing so I feel pretty confident saying that it should work on a fresh 16.04 install. I don't know of an easy way for you to roll back to a fresh install (entire graphics stack not just kernel) then drop the amdgpu-pro stack on top without having a separate partition though so don't feel good about asking you to try that... but if you had another system or HDD sitting around I guess I would ask.

            EDIT - and I realized I already asked above, clearly I need sleep

            Originally posted by muncrief View Post
            Keep in mind that I have an R9 390, which is a GCN 1.1 card. As far as I know CIK has to be enabled in the kernel to make it work, but as I said I've done that numerous times before and there are still various problems.
            The only "enabling CIK in the kernel" required is enabling in the amdgpu kernel driver build, and the DKMS package takes care of that.
            Last edited by bridgman; 03 November 2016, 10:46 PM.
            Test signature

            Comment


            • #56
              Originally posted by bridgman View Post

              OK... I'm pretty sure our testing uses the generic kernel (that's the stock one off a fresh install, isn't it ?). I'm wondering if there is something specific about your card/display the DAL code in the -pro driver isn't handling properly. Is there anything unusual about display connection, ie going through a KVM or anything ?

              What connection type & display res are you running ?

              Are there any other graphics cards installed in the system with displays connected, eg the integrated GPU on your CPU/APU ?

              If you do a fresh install and use the built in open source drivers does the display come up OK ?

              I'm glad you are able to recover but as you say it shouldn't be that hard. We install and test on a lot of different systems before releasing so I feel pretty confident saying that it should work on a fresh 16.04 install. I don't know of an easy way for you to roll back to a fresh install (entire graphics stack not just kernel) then drop the amdgpu-pro stack on top without having a separate partition though so don't feel good about asking you to try that... but if you had another system or HDD sitting around I guess I would ask.

              EDIT - and I realized I already asked above, clearly I need sleep
              I don't have anything unusual in my configuration that I know of. I'm running Xubuntu 16.04 on an FX-6300 with a Gigabyte GA-990FXA-UD5 Rev 3.0 motherboard and 16GB of DDR3. I have VMWare Player 12.5, and a KXStudio audio system with Jack 2, installed. Other than that I can't think of anything that could be related to the problems. Also I'm connected through DVI directly to a 27" LG Flatron LCD display, don't have any integrated graphics onboard, or any other graphics cards.

              In any case I'll try to do a clean install with the generic kernel and the driver only this weekend. Once again thank you for your help, until now I just assumed my card wasn't really supposed to work reliably yet. I'm more than happy to help debug the issue because, as I said in my original post, I've read of a lot of people having similar experiences to mine.

              Comment


              • #57
                I don't know if QA does any testing with Xubuntu but will find out. Didn't see anything in your configuration that should cause a problem but will see what we have in QA suite that is similar. Thanks !
                Test signature

                Comment


                • #58
                  Originally posted by bridgman View Post
                  I don't know if QA does any testing with Xubuntu but will find out. Didn't see anything in your configuration that should cause a problem but will see what we have in QA suite that is similar. Thanks !
                  Okay, I finally had a few days to try and get AMDGPU-Pro to work and discovered two things that were causing problems. One was easy, but the other evidently can't be fixed by the user.

                  First of all, the easy one. It's not enough to blacklist radeon, you must also uninstall xserver-xorg-video-radeon (sudo apt-get remove xserver-xorg-video-radeon). After that the system will reboot and run correctly. But here's the one that can't be fixed - If you have two displays and disable one the system will lock up thereafter.The only way to boot after that is to remove the ~/.config/xfce4/xfconf/xfce-perchannel-xml/displays.xml file and start again with two displays. By the way I have my main monitor on my DVI port, and the HDTV I used to use for gaming on the HDMI port.

                  Needless to say over two days I tried many different hardware and software configurations, including a plethora of xorg.conf files, but for some reason AMDGPU-Pro will not work unless both displays are enabled. It locks up immediately after entering your password if you don't.

                  Comment

                  Working...
                  X