Ubuntu 20.04 Gaming Performance Across Desktops, X.Org vs. Wayland

[birdie]

Who ever invented a fairy tale that Wayland should have more performance?

To be honest at least in theory Wayland (based DEs) should work faster since applications are essentially working with KMS near directly however it looks like Xorg is optimized enough to avoid a significant performance loss despite its many more layers of abstraction.

Also, full screen applications using OpenGL/Vulkan work somewhat similarly under Wayland/Xorg, so it might explain their similar performance but I'm not an expert and it would be interesting to listen to more knowledgable people.

[Volta]

KDE out-performs Gnome on wayland. And both are pretty far behind Xorg performance. Does this mean that all those who have been screaming "Xorg is dead" in the comments, and "No one is working on Xorg - they all hate it and want to kill it" - have been misrepresenting the facts?

No, it's matter of time when software will stop relying on legacy stuff like X. Can you name a single native Wayland application being tested in this benchmark? When comes to Xorg:

https://en.wikipedia.org/wiki/X_Window_System#Release_history

Last release is 2012. Sorry, but it's a dead horse.

[oiaohm]

12 years in. This is embarrassing.

Yes it is that you still have conformational bias.

https://www.phoronix.com/scan.php?page=article&item=ubuntu-2004-desktops&num=3

Just go and look at portal. Now we have games that the best option to play them is Wayland not X11 as well.

The fact of the matter is that Xorg still works beautifully for 99% of people out there

Total crap. Portal and other examples like it where Wayland beats Xorg are turn up more and more often now. X11 xorg on bare metal is coming the best case less than 40 percent of the time.

while Wayland still has teething issues even though it was meant to be THE BEST GRAPHICS SYSTEM IN THE WORLD EVAR. But that will surely change - we just have to wait a few more years. Ten will be enough I guess. Right? 22 years of development must surely be enough!

The X11 performance issues are not fixable they are in the base protocol so where wayland starts winning there is no coming back for X11.

What Wayland fans fail to realize is that you don't swap something which is working fine for absolute most people out there with something which still has a ton of issues and corner cases and requires special graphics drivers and which doesn't work well or doesn't work at all with NVIDIA.

Really birdie if you drop your conformational bias for one min and go and look at the numbers carefully. Other than the 2 examples that are really horrible for wayland all the other benchmarks are in wayland favour. So is X11 x.org server working fine for most people any more the answer like it or not is coming no.

And to add insult to injury Wayland requires a ton of work/code from the environments which want to work with it that's why XFCE/IceWM developers have simply ignored it so far.

Yes doing the ton of work to make Wayland work well at KDE and Gnome as resulted in Xfce and IceWM general performance advantage going by by even in X11 mode.

Performance optimisation takes a ton of work either way.

[evil_core]

I'm glad it has been testesd using Radeon VII.
But it's strange that some engines has been tested only using 1080p, others only 1440p, while some gained also 2160p and different antialiasing settings.
I personally think that benchmark would be more interesting when all games would be tested using identical sets of resolutions.

ET:Legacy is a shitty engine. Anybody playing True Combat Elite knows that, so I don't think it matters much (I liked this game, but Counter Strike framerate was much more smooth and predictable[especially with smoke, etc])

And about Xorg vs Wayland flamewar, it's stupid, se please stop that!
non-composited Xorg will e always faster when window rendering, while fullscreen, unredirected rendering should be faster on wayland, because of lower overhead (and then it shouldn't matter much Xorg vs Wayland, because it's using same Mesa driver stack)
The problem is that all native current wayland games don't bypass compositor(it's frameless. monitor-sized, centered windows), so you are benchmarking compositor, not Xorg/wayland (and differences mainly occurs, because different codepathes for Xorg and wayland in those compositors). Soon there should be extensions for fullscreen apps for wayland and later implementation in SDL2, and for dynamically linked SDL2 apps you will be able to see difference, but for most commercial games shipping outdated libs you know what it mean(you have also to wait for distros to pick those stable releases in their next stable distros releases. It will take years for most users)

[andyprough]

Originally posted by birdie View Post

The fact of the matter is that Xorg still works beautifully for 99% of people out there

Just go and look at portal. Now we have games that the best option to play them is Wayland not X11 as well.

Seems like you've made birdie's point - far fewer than 1% of GNU/Linux computer users are playing Portal on Wayland. So the 99% number probably stands as an accurate assessment.

[You-]

Phoronix: Ubuntu 20.04 Gaming Performance Across Desktops, X.Org vs. Wayland

Last month we provided some early benchmarks looking at the Ubuntu 20.04 X.Org vs. Wayland gaming performance under GNOME 3.36, but now that Ubuntu 20.04 LTS has been officially released, here is a look at the AMD Radeon Linux gaming performance across a wide variety of desktops on both X.Org and Wayland where supported.

http://www.phoronix.com/vr.php?view=29124

It also depends on what you measure. If you are measuring something designed to bypass Wayland/X11, then as long as that is done well, there should be little difference.

The same tests in windowed mode will be interesting - see the performance hit when X11 is in use compared to when a Wayland compositor is used.

The other issue is that compositors have been designed to be compatible with both X11 and Wayland. There is a question whether this prevents optimisations or ideas which break compatibility but would result in much improved Wayland performance.

[Michael_S]

Most of the core Wayland contributors are long time Xorg contributors. So this isn't a bunch of outsiders challenging the established technology, this is the leading experts with the established technology announcing that it sucks and a replacement is needed.

If the game is based on Xorg the statistic in this comparison is not congruent. The comparison is congruent comparing games based on Xorg with the same games based on Wayland.

But a lot of the games tested were not ported to run natively on Wayland. That's not Michael Larabel's fault or the Wayland developers' fault.

[andyprough]

Soon there should be extensions for fullscreen apps for wayland

Again, you are making birdie's point - we are 12 years into this project, it literally does not work at all with the vast majority of graphics cards or desktops that are currently in use, and you are saying "well, soon, there will be this extension that will fix one of these little problems...". When birdie says "22 years of development must surely be enough!", it's a fair point.

[franglais125]

andyprough and birdie : as every time this is tested, the usual caveat follows.

This is *not* Xorg vs Wayland. This is XOrg vs Xwayland.

We are essentially testing what the penalty of an additional layer can be. In some cases, Wayland even turns out to be faster. For the 2 cases where there is an obvious bug, it looks very dramatic, but hopefully that's addressed soon.

We are yet to see the actual difference Xorg vs Wayland (not claiming it is much faster, simply interested in seeing that.)

[TumultuousUnicorn]

Oh, Sway in a benchmark! Thank you Michael!