Fedora 23: KDE vs. Xfce vs. GNOME vs. LXDE vs. MATE

[Del_]

No, of course not. It affects the game play, just like the browser with the facebook tab does, just like any other running application affects it. We have three things to look at: 1) above 60 fps: In this area it doesn't matter if (for whatever reason we "slow down" the game). We render the 60 fps the user can at max see. 2) significantly below 60 fps: the system is not able to render the game. Yes disabling compositing might give you a few more frames, but the situation stays the same, the system is not able to render the game. In this case the environment needs to be adjusted so that the game can be rendered (close facebook for example). Situation 3) the game would be able to render at 60 fps, but KWin slows it down so that it doesn't render at 60 fps. That would be the interesting case to look at, because that's the one which would need fixing. But that the benchmark cannot show. To get this data we would have to look into the timing KWin does on rendering. We would need to know how often we missed the frame. Sorry the way how phoronix benchmarks are made, that's just not possible. And sorry also in KWin we don't have such an infrastructure yet. I'm considering to add one for the DRM backend on Wayland, because there it becomes interesting.

I see your points, thanks for sharing. My perspective is a bit different, it goes like this. 1)Some games run smoothly on all my set-ups, no issues there facebook or not. 2) Demanding games typically have graphic options I need to turn off to get decent performance, here every fps counts 3) Some games really need to run without slowdowns, because it ruins gameplay (think Counter Strike Global Offensive where you are on an online team), here every fps counts.

Getting means to do better benchmarking on wayland sounds very interesting, looking forward to see how it plays out. In the meantime, phoronix is the only place I find valuable information on the current state of graphics in linux land.

systemsettings -> Display and Monitor -> Compositor -> second last checkbox.

Ah, there it is. Thank you!

[mgraesslin]

Are you claiming that compositing has no effect for sub-60fps gaming?

No, of course not. It affects the game play, just like the browser with the facebook tab does, just like any other running application affects it. We have three things to look at: 1) above 60 fps: In this area it doesn't matter if (for whatever reason we "slow down" the game). We render the 60 fps the user can at max see. 2) significantly below 60 fps: the system is not able to render the game. Yes disabling compositing might give you a few more frames, but the situation stays the same, the system is not able to render the game. In this case the environment needs to be adjusted so that the game can be rendered (close facebook for example). Situation 3) the game would be able to render at 60 fps, but KWin slows it down so that it doesn't render at 60 fps. That would be the interesting case to look at, because that's the one which would need fixing. But that the benchmark cannot show. To get this data we would have to look into the timing KWin does on rendering. We would need to know how often we missed the frame. Sorry the way how phoronix benchmarks are made, that's just not possible. And sorry also in KWin we don't have such an infrastructure yet. I'm considering to add one for the DRM backend on Wayland, because there it becomes interesting.

Already looked five times without finding it.

systemsettings -> Display and Monitor -> Compositor -> second last checkbox.

[Del_]

No sorry, the benchmarks do not show that. You can trust me on that, I know my code.

Are you claiming that compositing has no effect for sub-60fps gaming?

Of course the setting still exists.

Already looked five times without finding it.

Excellent performance, for whom? For productivity users or for gamers?

For me, I fit both categories. For my kids, they actually also fit both, and they like their music player and whatnot running while they game.

[Del_]

Double post

[mgraesslin]

Because expecting the end-users to shift+alt+f12 for games is the way to go.

No we don't expect end-users to know that. Because of that we gave games a way to indicate that this should happen automatically.

[mgraesslin]

The relevant questions are whether the performance differences matter, and whether they represent real sub-60fps situations. It does and they do.

No sorry, the benchmarks do not show that. You can trust me on that, I know my code.

It seems system settings no longer allows me to turn off compositing for full screen windows

Of course the setting still exists.

Thanks for reminding me of the Shift+Alt+F12 shortcut, but honestly, we need to have a default set-up that gives excellent performance. Performance matters.

Excellent performance, for whom? For productivity users or for gamers? Sorry to say, but I don't optimize for gamers. I just don't think it's worth the effort to do so on X11, because well you can optimize the shit out of X, X stays X. If gamers care about each fps, there are solutions to that: run a dedicated X and don't do anything else with the system. You don't want your facebook tab in browser to reload and waste CPU cycles in Flash. Do what a game console does: do nothing else than it. We cannot compete with a PlayStation 4 when a whole DE is running. Why should we try at all. Let's better make sure that a realistic usage scenario works fine.

[drSeehas]

"... FirePro V7900 GPU ..."
The FirePro V7900 is NOT a GPU. It is an entire graphics card.

[edoantonioco]

We consider this a useless option for Plasma. If one wants to run games: turn of desktop effects completely. Alt+Shift+F12 to do so, use a kwin rule to set it automatically for a window or use a "game script" (see http://kde-look.org/content/show.php...content=156659). Even games can indicate to KWin that they want desktop effects disabled by a simple property.

Because expecting the end-users to shift+alt+f12 for games is the way to go. Well, I guess than with wayland that won't matter (afaik compositing does not affect fps on it), so I guess than until that moment is when kde will have a chance to really improve how user-friendly it can become.

[Del_]

Sure, one should have 60 FPS. Not one frame more. Now please look at the benchmark result in this benchmark: all values are significantly above the 60 fps threshold. So given that: I don't care about it. The only relevant question is whether we don't get the 60 FPS when we should get it. This benchmark doesn't answer the question and I haven't seen any on Phoronix yet which would answer that question (sure there were benchmarks where it didn't hit the 60, but then all test candidates were significantly below 60). I have no data on which I could say that we do not get the 60 FPS when we should. So from my perspective everything is fine.

I am afraid you are not being very constructive. Benchmarks should try to be representative. Designing them so that they capture the break-point of 60fps on every set-up is not practically feasible, and I am sure you understand that. The relevant questions are whether the performance differences matter, and whether they represent real sub-60fps situations. It does and they do. I am running Debian testing on my main machines here, and upgrading to Plasma 5.4 has been a bit on the painful side, but that is another story. It seems system settings no longer allows me to turn off compositing for full screen windows, and from the thread I assume it is on by default. Thanks for reminding me of the Shift+Alt+F12 shortcut, but honestly, we need to have a default set-up that gives excellent performance. Performance matters.

[mgraesslin]

I have a big ass PC, it cost me a lot of money, and when I play games in it, especially heavy games (and now in Linux we have a lot of then), I sure want every single FPS I can get, because I'm not playing these games at 150 FPS, I'm getting below 60 FPS in a lot of then, so every FPS does count. Even Indie games, that I play in my work machine with a APU, need some optimization to get playable FPS.

Sure, one should have 60 FPS. Not one frame more. Now please look at the benchmark result in this benchmark: all values are significantly above the 60 fps threshold. So given that: I don't care about it. The only relevant question is whether we don't get the 60 FPS when we should get it. This benchmark doesn't answer the question and I haven't seen any on Phoronix yet which would answer that question (sure there were benchmarks where it didn't hit the 60, but then all test candidates were significantly below 60). I have no data on which I could say that we do not get the 60 FPS when we should. So from my perspective everything is fine.