NVIDIA Linux Gaming Performance For Wayland vs. X.Org On Ubuntu 22.04

[pinguinpc]

Reading these conversations between Volta and Birdie is genuinely hilarious. Y'all need to just chill out. Seriously.

and now this

[Sin2x]

I'm sure you've never managed large complex infrastructures because systemd is a godsend on them.

Funny thing is -- managing complex infrastructure in the cloud is what I do for a living. And systemd doesn't play any role there at all, k8s does.

Everything else from your list is not true as well: this very benchmark shows clearly that Xorg gives more FPS in games; GCC produces faster binaries and doesn't require literal tens of gigabytes of memory and disk space to produce them like LLVM, and traditional inits (why choose obsolete SysV? there's runit, openRC, s6 -- pick your __modern__ poison) give you control which binary nature of systemd could never give you by defintion.

[skeevy420]

*nom* *nom* *nom*

[ColdDistance]

Haiku and Google Fuchsia have real chances of becoming native open source platforms for games but Linux has blown its chances completely.

With crappy and permissive licenses like MIT and Apache? No. Their proprietary derivatives would be gaming platforms.

[treba]

For anyone wondering: Nvidia still misses some APIs to run at optimal speed on Wayland. As this is fullscreen games, on X11 "unredirect" is used while on Wayland "direct scanout" would be used with all APIs in place (happens with Mesa drivers, which is one reason Wayland is faster on AMD as previous benchmarks have shown). It should be noted that unredirected/uncomposited X11 can show tearing while Wayland with scanout does not.

[birdie]

Funny thing is -- managing complex infrastructure in the cloud is what I do for a living. And systemd doesn't play any role there at all, k8s does.

Everything else from your list is not true as well: this very benchmark shows clearly that Xorg gives more FPS in games; GCC produces faster binaries and doesn't require literal tens of gigabytes of memory and disk space to produce them like LLVM, and traditional inits (why choose obsolete SysV? there's runit, openRC, s6 -- pick your __modern__ poison) give you control which binary nature of systemd could never give you by defintion.

On Systemd

I've no idea how k8s is related to startup scripts and their management as we don't yet use it but we extensively use systemd and its features and we are quite happy with what systemd has in store:

One command to start multiple services with intricate dependencies? ☑
Automatically restart services if they fail? ☑
Easily check the service status? ☑
Tons of features in terms of controlling what the service is allowed to access? ☑
And a billion other features.

SysVinit gives me no control whatsoever, it's just crap. A basic utility which can only do one thing: run scripts from a certain run level one by one. Zero control whatsoever, zero accountability, nothing. I last used it in CentOS 6 and it was painful. Systemd streamlines and unifies a ton of things SysVinit didn't even begin to tackle.

You continue to disparage systemd and it tells me one thing: you've never actually used it or its features. It was unstable around six years ago but nowadays it's rock solid

On Wayland

There already have been benchmarks where Wayland is in the lead.

Again, I'm not a fan of Wayland, in fact just like you I think it's a ton of wasted effort but I can give credit where its due. Wayland is a lot more simpler (read efficient) in terms of graphics rendering but simplicity comes at a great cost and that is the Wayland compositor has to implement a ton of features which the Xorg server offered by default.

On LLVM

You completely ignored the licensing issue. There are multiple cases where LLVM produces binaries which are faster than those built with GCC. Lastly almost each release GCC sees quite serious performance regressions which I've not noticed with LLVM. It certainly doesn't help that GCC has become so complicated basically five dozen people in the entire world understand and can contribute to its code base.

Trying to lump all the new features/tech reeks of being a Luddite and I sometimes am but you still need to remain objective cause it's easy to lose track of something new and better which measurably improves your quality of life.

[skeevy420]

With crappy and permissive licenses like MIT and Apache? No. Their proprietary derivatives would be gaming platforms.

You're too limited in thought. The automotive industry would love an OS they could all use as a base and then reskin and tweak to their own needs without worrying about who is violating what or who. I suppose a car or plane needs better integration between kernel and userspace than Linux and GPL would allow...in regards to revealing trade secrets, State secrets, etc. Most licenses don't have those kinds of exceptions.

[birdie]

Reading these conversations between Volta and Birdie is genuinely hilarious. Y'all need to just chill out. Seriously.

Your name checks out, Lbib_ass. Not sure if it was intentional or it's meant to be pronounced differently. I see nothing hilarious in being called a Microsoft/NVIDIA/Intel shill or a joke and just being outright insulted and mocked at. Must be something wrong with you if you enjoy people being publicly denigrated but it's not my problem. It's yours and Volta's issue.

It takes five seconds to find my numerous contributions to Open Source (in the range of several hundreds). Volta's contributions other than being a raging a$$hole? Nothing so far.

[ColdDistance]

You're too limited in thought. The automotive industry would love an OS they could all use as a base and then reskin and tweak to their own needs without worrying about who is violating what or who. I suppose a car or plane needs better integration between kernel and userspace than Linux and GPL would allow...in regards to revealing trade secrets, State secrets, etc. Most licenses don't have those kinds of exceptions.

What's the reason to release code if the only consequence is to be parasitized by everybody? The consequence of your thinking will be to that all software will turn proprietary.

[Sin2x]

On Systemd

I've no idea how k8s is related to startup scripts and their management as we don't yet use it but we extensively use systemd and its features and we are quite happy with what systemd has in store:

One command to start multiple services with intricate dependencies? ☑
Automatically restart services if they fail? ☑
Easily check the service status? ☑
Tons of features in terms of controlling what the service is allowed to access? ☑
And a billion other features.

SysVinit gives me no control whatsoever, it's just crap. A basic utility which can only do one thing: run scripts from a certain run level one by one. Zero control whatsoever, zero accountability, nothing. I last used it in CentOS 6 and it was painful. Systemd streamlines and unifies a ton of things SysVinit didn't even begin to tackle.

You continue to disparage systemd and it tells me one thing: you've never actually used it or its features. It was unstable around six years ago but nowadays it's rock solid

OpenRC gives you all the same features and adheres to KISS, while systemd does not. Which exact features are you missing to force you to use systemd with its mess of a code that you are never going to be able to audit, its ugly and unwieldy binary log format and its totalitarian wish to be the "one world to rule them all"?

Diversity? Competition? Bazaar? No, never heard of either.

On Wayland

There already have been benchmarks where Wayland is in the lead.

I don't see the links.

On LLVM

You completely ignored the licensing issue. There are multiple cases where LLVM produces binaries which are faster than those built with GCC. Lastly almost each release GCC sees quite serious performance regressions which I've not noticed with LLVM. It certainly doesn't help that GCC has become so complicated basically five dozen people in the entire world understand and can contribute to its code base.

Trying to lump all the new features/tech reeks of being a Luddite and I sometimes am but you still need to remain objective cause it's easy to lose track of something new and better which measurably improves your quality of life.

Why should I care about licensing issues, I'm not a corporation that prefers non-copyleft licenses for obvious reasons -- while you certainly do sound like a shill for IBM/RedHat. I care about software quality and GCC is superior.


> There are multiple cases where LLVM produces binaries which are faster than those built with GCC.

Where are the links to those as well?