Announcement

Collapse
No announcement yet.

KDE On Wayland: "The Biggest Thing Needed Now Is Adoption By 3rd Party Apps"

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

  • Originally posted by oiaohm View Post

    I see mSparks cannot read.

    How is CPU running in powersave with a max of 5.07Ghz going to compete with performance with CPU maxing out at 5.3 Gpz.
    1. it crashed at 5.3GHz
    2. "as normally with P-State powersave the top performance states (and in turn peak power draw) is still being hit"
    3. ignoring stability issues is a you problem. You should fix that.
    Last edited by mSparks; 26 September 2023, 07:45 PM.

    Comment


    • Originally posted by mSparks View Post
      1. it crashed at 5.3GHz
      2. "as normally with P-State powersave the top performance states (and in turn peak power draw) is still being hit"
      3. ignoring stability issues is a you problem. You should fix that.
      Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

      Where does it say crashed please quote exactly.

      Xeon one from 2020 has high clock no higher power draw. Yes powersave and performance on Xeons mostly are just a clock boost.

      Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

      Yes the last page clocks is important. Basically in performance your CPU average clock with Intel is sitting higher than the Max clock-speed that you can do in powersave.

      This 2021 test does not say crashes or stability issues. Lot more power draw. Some chips don't draw more power with Intel just you find that adverage clock is faster than you can do in ondemard or powersave with performance enabled.

      You really need to start reading this stuff on the Intel chips benchmarks with ondemard vs powersave vs performance. Then remember lot of distributions come out with powersave is about 70 percent power save 29% ondepard and about 1 percent performance when you look at distribution defaults.

      ondemard, powersave and performance all have different max CPU clock limits with Intel and that was first noticed on Xeons in 2020 then in consumer chips 2021 most likely had been around longer than that.

      Comment


      • Originally posted by oiaohm View Post

        Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

        Where does it say crashed please quote exactly.

        "That higher power consumption is seeming to cause stability issues for at least some workloads."

        Do you really think linking an article article that outright states that cpu is an exception to the rule that powersave and performance modes hit the same frequencies at high loads, and when it did performance mode was crashing says anything other than normally powersave and performance mode are identical under load.

        This is _very_ different to windows powersave mode that typically cripples performance at half what the cpu is capable of and shortens battery life as a result.
        Last edited by mSparks; 26 September 2023, 08:44 PM.

        Comment


        • Originally posted by mSparks View Post
          "That higher power consumption is seeming to cause stability issues for at least some workloads."
          But is that in the 2020 Xeon benchmarks showing the same clock thing those had perfect stability and no extra power usage.

          Also here a good question is that in fact abnormal for that chip.

          2021 Intel cpus enable performance mode is a kick me sign be it Windows or Linux some of the time. Yes you get higher perform yes you get more crashes and this is dependent if you won or lost the silicon lottery. Yes have windows reports of the oddness back with 2018 CPUs.

          Sample size of 1 has it problems mSparks. Yes massively increasing CPU wattage because you enable performance mode can make the 5-40% gain pointless because now your cooling solution can collapse in a heap as it get saturated with massive spikes in heat.

          Yes powersave, ondemard and powerformance with Intel results in your cpu running at different temperatures with different power draw and different clockspeeds.

          You quoted that performance most could be unstable. So why are you happy that compare benchmark has been done in performance mode for any test. Performance mode with Intel is basically like vendor approved overclocking with all the problems.

          Comment


          • Originally posted by oiaohm View Post

            Yes you get higher perform yes you get more crashes
            you dont get more crashes for the same cpufreq, its purely an artifact of a cpu being allowed to clock to high and powersave detecting load isnt decreasing if it increases the frequency further.

            There are some applications where performance mode is useful. games isnt one of them. for games performance mode and powersave modes are identical.

            examples where going to perf mode can be useful include low latency servers, when requests need to complete in the shortest time possible but the requests never actually load the server enough to send it to a higher powerstate. Others include the RT systems when behaviour needs to be 100% deterministic (and generally there it would be perf mode with an underclocked cpu)
            Last edited by mSparks; 26 September 2023, 09:49 PM.

            Comment


            • Originally posted by mSparks View Post
              There are some applications where performance mode is useful. games isnt one of them. for games performance mode and powersave modes are identical.
              Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite


              Was not true in 2019 and has not been true every time since it been tested but the difference was small. With the gap getting wider between performance and powersave.

              Performance is normally classed as not advantage in games due to the extra crashing it can introduce.

              You need to catch up with modern Intel CPUs. If all his tests were done with powersave I would believe the number. Like the second set of tests could be higher because the laptop was cooler and with performance max clock is unlocked so the system will go to thermal limit.

              Reason why since 2020/21 you system with intel can come unstable with performance is that you can overheat everything if your system cooling is not good enough and the cpu you have just happens to go like the one I point to say hell yes I can pull another 100-200 more watts of power because you asked for performance.

              Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

              And its not only Intel that picked up this behavior. AMD has as well in their mobile. Yes 2023 AMD the difference between powersave, ondemard and performance when gaming is very clear.

              Yes AMD pstate powersave say hello to only 20% of the performance of performance mode in games with ondemard not catching up to performance yes still 10 percent down..

              Comment


              • Originally posted by oiaohm View Post
                Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite


                Was not true in 2019 and has not been true every time since it been tested but the difference was small. With the gap getting wider between performance and powersave.

                Performance is normally classed as not advantage in games due to the extra crashing it can introduce.

                You need to catch up with modern Intel CPUs. If all his tests were done with powersave I would believe the number. Like the second set of tests could be higher because the laptop was cooler and with performance max clock is unlocked so the system will go to thermal limit.

                Reason why since 2020/21 you system with intel can come unstable with performance is that you can overheat everything if your system cooling is not good enough and the cpu you have just happens to go like the one I point to say hell yes I can pull another 100-200 more watts of power because you asked for performance.

                Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

                And its not only Intel that picked up this behavior. AMD has as well in their mobile. Yes 2023 AMD the difference between powersave, ondemard and performance when gaming is very clear.

                Yes AMD pstate powersave say hello to only 20% of the performance of performance mode in games with ondemard not catching up to performance yes still 10 percent down..
                Its really simple.
                In a game getting 64 fps in X11 with performance mode enabled and vsync disabled on a typical 60Hz monitor
                In performance mode with vsync enabled you will get 60fps
                with power save mode and vsync enabled you will get 60fps
                On wayland in performance mode vsync enabled you will get 30fps
                On wayland in power save mode vsync enabled you will get 30fps

                Its not that complicated.
                For games Wayland sux monkey nuts and performance v power save mode is identical, hence his shock nobaru went with wayland, and performance mode not being important.
                Last edited by mSparks; 27 September 2023, 07:26 AM.

                Comment


                • Originally posted by mSparks View Post
                  Its really simple.
                  In a game getting 64 fps in X11 with performance mode enabled and vsync disabled on a typical 60Hz monitor
                  In performance mode with vsync enabled you will get 60fps
                  with power save mode and vsync enabled you will get 60fps
                  On wayland in performance mode vsync enabled you will get 30fps
                  On wayland in power save mode vsync enabled you will get 30fps

                  Its not that complicated.
                  For games Wayland sux monkey nuts and performance v power save mode is identical, hence his shock nobaru went with wayland, and performance mode not being important.
                  Really explain gamescope.
                  SteamOS session compositing window manager. Contribute to ValveSoftware/gamescope development by creating an account on GitHub.

                  Its not a 50 percent hit.

                  Remember you can run gamescope ontop of X11 baremetal. Its still using the Wayland protocol. Magic right that the differences you are talking about go away. There is a issue with Nvidia drivers when a Wayland compositor setup on bare metal. The Nvidia x.org user space DDX driver is doing something different to get from 4 swap chain to 2 swapchain. Force X11 to use mode setting driver on Nvidia hardware see the same performance hit Wayland compositors are suffering form.

                  His benchmarking was not using locked to vsync. What you described would be a 50% hit that not the case.



                  Nobara is fairly clear they want GBM/DMABUF to work. They are not going to do the X11 work around forever.

                  Something to remember Nobara started AMD hardware solution with the Nvidia bits tacked on latter and you do feel it.

                  Comment


                  • Originally posted by oiaohm View Post

                    Really explain gamescope.
                    SteamOS session compositing window manager. Contribute to ValveSoftware/gamescope development by creating an account on GitHub.

                    Its not a 50 percent hit.

                    with vysnc on a normal (none VRR) you have two fps options. 60, or 30.
                    if you are at 64fps and take a 5% hit you stay at 60, if you take a 10% hit you fall to 30.

                    gamescope makes sense just for "write their own display server for the steam deck that forces games fullscreen even when they are set to windowed mode"

                    Im sure higher fps was also an expectation and they are as surprised as everyone else that wayland compositors can barely match x11 display servers at best.

                    TBH though it shouldnt be that surprising, X11s 40+ years of optimisations to run well on the weakest of microcontrollers was never going to be the easy competition it was made out to be.

                    Comment


                    • Originally posted by mSparks View Post
                      with vysnc on a normal (none VRR) you have two fps options. 60, or 30.
                      if you are at 64fps and take a 5% hit you stay at 60, if you take a 10% hit you fall to 30.
                      ​Except that not how it behaves at all.
                      https://www.reddit.com/r/skyrimmods/..._fps_never_60/ does not matter if you are Linux or Windows by the way. You don't straight fall to 30 because you cannot meet 60 vsync. Yes 57-58 you have odd frames here and there being displayed twice with 60vsync.

                      Fun problem some games miss gauge the vsync meaning even without overhead you end up with these screw ups. Yes skyrim is one of those programs that mucks up it frame pacing so you have vsync on and need to tell skyrim limiter off " EnableFPSLimit=false" so it does 60fps instead of 57-58 when in 60vsync.

                      Basically claiming instant drop to 30 means you have no clue how this really works.

                      Also you have a horrible one that some VRR applications intentionally set their limit 3-5 frames under the hz to give VRR the means to double up frames. Yes you can now see a reason why you don't see VRR monitors that are 60hz. Frame doubling and Frame skipping is a common thing to attempt to counter latency issues.

                      Originally posted by mSparks View Post
                      gamescope makes sense just for "write their own display server for the steam deck that forces games fullscreen even when they are set to windowed mode"
                      Gamescope replaced steamcompmgr on the steamos/steamdeck. Gamescope come out of the libcapsule work about legacy application support also funded by Valve.

                      It did not make sense for Valve to be making two compositors.

                      gamescope is also auto used when you install different legacy games on your normal X11 desktops from Steam. Gamescope is not restricted to steamdecks or to wayland solutions.

                      It's getting game frames through Wayland by way of Xwayland, so there's no copy within X itself before it gets the frame.
                      Gamescope documentation has some super gems. Xwayland does not add overhead because it does not in fact make copies. Yes the gamescope developer worked out very quickly that Xwayland on top of Wayland compositor adds basically zero latency. Yet xnest and xephyr that they were first using with legacy application support was adding a hell lot of latency because those both add copies.

                      Originally posted by mSparks View Post
                      Im sure higher fps was also an expectation and they are as surprised as everyone else that wayland compositors can barely match x11 display servers at best.
                      Not really. Issues there is only so fast you can go. X11 bare metal without compositor, Wayland Compositor and KMS/DRI(the direct) with working drivers the difference is margin of error because you end up hitting the hardware limit.

                      Driver defects with X11 bare metal/Wayland Compositor and KMS/DRI can all result in poor performance. We have seen cases where the intel driver is bad with X11 yet fine with Wayland and KMS/DRI as well.

                      Originally posted by mSparks View Post
                      TBH though it shouldnt be that surprising, X11s 40+ years of optimisations to run well on the weakest of microcontrollers was never going to be the easy competition it was made out to be.
                      Except X11 composite extension is not 40 years old and was not designed to be optimized so resulting in X11 server + X11 compositor performing badly. Of course until 2022 this was presumed to be all X11 composite extension. Some of why X11 compositors perform so bad is the graphics drivers themselves. Ok fixing the graphics driver issues only make X11 compositor sux slightly less not completely fix the problem. Fix the X11 compositor performance problem is rewrite the X11 composite extension and all X11 compositors or go Wayland compositor.


                      Wayland first developer was focused on fixing the X11 Compositor over head issue.

                      Comment

                      Working...
                      X