NVIDIA GeForce RTX 3070 Ti Linux Performance

[mppix]

Yes and this is because Linux is the only mainstream OS that doesn't maintain a low level interface for graphics drivers which forces the drivers to be packaged along with the release of a Kernel.

Also other OS change ABI/API, they just coordinate it with the GPU vendors.

This isn't a good thing and its actually whats preventing Linux from being adopted in mainstream, neither OSX nor NVidia have this problem. Note that his has nothing to do with the license for the drivers, there is a reason why AMD on Linux has both open source and proprietary drivers, its because they latter they can support their hardware on older versions of the kernel.

Nvidia is certainly not the key issue preventing Linux from going mainstream.
Lack of specialized/proprietary software is a much bigger issue here (in part caused by a uniform software distribution system (hopefully flatpak/snap changes that).

This is also a misleadingly loose definition of the word "break". If you are compiling the kernel from source then sure, but if you are using a distribution that has package management you don't get this problem as an end user (and NVidia is historically very good time wise when it comes to supporting newer versions of the kernel with the blob).

'break' = non working as intended (you can use a car analogy if you don't understand it here).
Nvidia driver can break in many ways and usually without proper error messaging/handling: fully broken, partially broken, only compute broken, only en/decoding broken, display broken

According to whom is it he de-facto standard? NVidia is using an open API for hardware acceleration and up until recently they were the only serious graphics card manufacturer for Linux. There is VAAPI which is an open API developed by Intel for their hardware and then you have VDPAU which is an open API developed by NVidia for their hardware, there just as open as eachother so I don't think taking sides here is mature.

I don't care what API I use as long as my programs support it. VAAPI is picking up more support.
Also, I don't think VDPAU is develop by Nvidia.. it supports it..

...Rambling quote removed....

Well I guess if your definition of "babying them" means hitting "update system" on your Linux distribution /s

You need to stop making an issue out of a problem that doesn't exist and if you honestly think that this is a problem then there are probably another 100 hundred things that are much worse in the Linux landscape then NVidia blob being out of date from Linux tree for a minute period of time.

I think you simply are on a distro that simply manages this for you like Ubuntu or Manjaro. There are others like Fedora or Debian that will break the Nvidia driver every so often if you just "update system". To avoid it, you need to check compatibility at every kernel/driver/core upgrade.

[mdedetrich]

'break' = non working as intended (you can use a car analogy if you don't understand it here).
Nvidia driver can break in many ways and usually without proper error messaging/handling: fully broken, partially broken, only compute broken, only en/decoding broken, display broken

Yeah and again, as I said in the past 15 years with 15+ devices amongst 10 Linux distributions, using your definition of "break" and relative to other software its deceitful to say that NVidia drivers break all the time.

In fact if anything, the graphics driver that broke all the time was fglrx in the ATI days (and thats why I went to NVidia, because it worked)

I don't care what API I use as long as my programs support it. VAAPI is picking up more support.
Also, I don't think VDPAU is develop by Nvidia.. it supports it..

VDPAU was developed by NVidia, its even on the wiki page, quote below

VDPAU was originally designed by Nvidia for their PureVideo SIP block present on their GeForce 8 series and later GPUs.^[7]

On March 9, 2015, Nvidia released VDPAU version 1.0 which supports High Efficiency Video Coding (HEVC) decoding for the Main, Main 4:4:4, Main Still Picture, Main 10, and Main 12 profiles.^[8]

I think you simply are on a distro that simply manages this for you like Ubuntu or Manjaro. There are others like Fedora or Debian that will break the Nvidia driver every so often if you just "update system". To avoid it, you need to check compatibility at every kernel/driver/core upgrade.

The whole point of a distro is to manage this for you, so if a distro is not doing this then they are doing a terrible job at it.

[smitty3268]

Yes and this is because Linux is the only mainstream OS that doesn't maintain a low level interface for graphics drivers which forces the drivers to be packaged along with the release of a Kernel.

This isn't a good thing and its actually whats preventing Linux from being adopted in mainstream

Hmm? What evidence do you to support that? In the long list of things holding desktop linux back, that doesn't even crack the top 20 for me.

There are several other unix-like OS's that largely use the same userspace as linux but do provide more stable kernel interfaces. None of them have taken off on the desktop either.

[MadCatX]

Then let's just respectfully agree to disagree here!

"irqbalance" may have a usecase on very large & complex server & HPC system setups where latency is not of big concern, but a typical desktop PC is better off without it, in my humble opinion.

You're free to believe whatever you want. However, since there is no solid evidence to support your opinion about irqbalance, you probably shouldn't be suggesting it as a culprit of bad gaming performance.

Now, about CONFIG_HZ:
Unfortunately, there is only one benchmark I could find regarding this, and it's "only" in relation to KVM passthrough gaming (Linux host + Windows guest).
Still, the numbers are quite compelling:

As you can see, 1000 Hz kernel timer tick on Linux can make quite a difference in minimum framerates, which is arguably one of the most important aspects in videogames.

Minimum FPS can be quite misleading. If the idea of a benchmark is to show that a game will always run at playable FPS on a given system, then the minimum FPS value is the most important one. If you are benchmarking overall performance or performance difference between two setups, then the minimum FPS value is not a very good indicator. You've got to realize that the minimum FPS value can be caused literally by a single frame that just happens to get held up for whatever reason. If a benchmark renders a total of 10000 frames, a single-frame hiccup doesn't say very much. Average FPS or, even better, 99.9 percentile FPS are much better indicators of overall performance.