Why distro choice matters when it comes to game performance

[j4np0l]

I asked the author, and he says that the 10% difference between Ubuntu and Chakra means that for every 10fps you see under Chakra you would only see 9 under Ubuntu.

So, if I run RO at 80fps in Chakra, I would get 72fps in Ubuntu, doesn't seem a like a lot of difference to me. But I think sutch conclution should be made with the game running (who knows how well the NGD engine performs compared to the Engine used in the benchmarks).

[Malevolence]

Its not the distro you chose its ngd and openGL. I went to windows after using Linux since '96 and mac before that. Just to get decent fps on this game opengl is fail fail fail. 19 fps on Shad2 opengl. 100 fps on directx its a no brainer for me. My pedestal is like an elevator. I use what works.

[HidraA]

Ubuntu 10.04 (Pinguy) - Gnome
Ubuntu 10.10 Beta - Gnome
Linux Mint Rolling Release (Debian Based) - Gnome
Chakra 0.2.1 (Arch Linux Based) - KDE
Fedora 13 - Gnome
Mandriva 2010.1 - KDE
OpenSUSE 11.3 - KDE
PCLinuxOS 2010.07 - KDE
Sabayon 5.3 (Gentoo Based) - KDE

From my test there are a few mistakes on this.
-Video performance depends of version of OpenGL and Video drivers.
-Video performances depends of charge of video memory.
-Video performances depends of GPU performances.
-kernel version,precompiled vga drivers.

Bad in that post:
-all time gnome will charge less like kde video memory.(plasma server charge lots of memory)
-all time compiled vga drivers will work better like precompiled drivers.(because you will compile for one dedicated vga card and will not load multi modules).


Because Ro excenge net data with one server:
-In ro matter ro server code,how many daemons are lauched and watching network interface,if you run torrents or another kind of daeomons what charge network adapter with internet connection,pear to server connection,

That version of top cant be aplyed for ro,ubuntu has damn slow internet connection,fedora has best net connection but weack video performances.

Best performances got on Debian with compiled vga drivers(not precompiled),shader 3.0 and compiled kernel(where i disabled lots of network modules).like this jumped from 50-60 fps to 120-130 fps(before new update,balance wtv they call this)

[Znurre]

Originally Posted by HidraA

-Video performance depends of version of OpenGL and Video drivers.

NVidia replaces the whole video stack with their binary blob.
The same driver version is used in all tests

-Video performances depends of GPU performances.

Same hardware in all cases.

Bad in that post:
-all time gnome will charge less like kde video memory.(plasma server charge lots of memory)

Why is it then that all distros running KDE performs better than the GNOME ones? Take a closer look, even Fedora which is the most high performant distro among the GNOME ones is below average.

-all time compiled vga drivers will work better like precompiled drivers.(because you will compile for one dedicated vga card and will not load multi modules).

In the case of the official NVidia driver, the only thing you really compile is the interface that is communicating with the kernel.
It does not make much of a difference in performance whether it's pre-compiled or not.

That version of top cant be aplyed for ro,ubuntu has damn slow internet connection,fedora has best net connection but weack video performances.

First of all, what do you base this on?
Second, the benchmarks in that test are based on the Unigine benchmark suit.
Those are all offline, standalone games where only FPS matters for the outcome of the test.

As you can see, Fedora outperforms Ubuntu when it comes to video performance.
Personally I think it could have to do with the I/O scheduler in use, or maybe the number of background processes in a default setup of the respective distros or the fact that GTK is drawn using a software renderer while Qt uses Xrender which is partially hardware accelerated.

[Arafails]

Originally Posted by Znurre

GTK is drawn using a software renderer while Qt uses Xrender which is partially hardware accelerated.

Even back before it used Cairo for just about everything, GTK+ was generally accelerated using XRender also. Cairo, of course, is accelerated wherever possible.

[Znurre]

Originally Posted by Arafails

Even back before it used Cairo for just about everything, GTK+ was generally accelerated using XRender also. Cairo, of course, is accelerated wherever possible.

Thanks, didn't know that.
I just based my assumption on this: http://brainstorm.ubuntu.com/idea/9450/

[Arafails]

It's somewhat irrelevant when you consider that when you're playing a game (or running a 3D benchmarking program), the toolkit is not going to be doing much (no*one really does any large screen updates unless you're switching tabs, scrolling pages at a time, or playing solitaire anyway).

Memory consumption is far more interesting, so I did a test on my own PC. First, I installed KDE, which isn't to my taste, so I don't really use it. Then, I did clean log ins on Openbox (that known memory hog!), KDE, and Gnome, dumping the free -m results into a file for each. Then I opened up a terminal window, and did:

[fudje@blackberry ~]$ for i in memory-*; do echo "$i:"; cat "$i"; done
memory-gnome:
             total       used       free     shared    buffers     cached
Mem:          1501       1292        209          0          0        851
-/+ buffers/cache:        441       1060
Swap:         4604          1       4603
memory-kde:
             total       used       free     shared    buffers     cached
Mem:          1501       1487         13          0          0        955
-/+ buffers/cache:        532        969
Swap:         4604          1       4603
memory-openbox:
             total       used       free     shared    buffers     cached
Mem:          1501        984        516          0          0        758
-/+ buffers/cache:        226       1275
Swap:         4604          1       4603
[fudje@blackberry ~]$

Now the numbers we're interested in here are the used and free amounts in the -/+ buffers/cache section. The results speak for themselves - With Openbox 226MB is used, Gnome 441MB, and KDE 532MB. Make Gnome 467MB because I don't run nautilus 100% of the time, and that takes an extra 26MB or so.

That's hardly an insignificant difference – especially on my system with its amazing 2GB minus 512MB for the on board graphics, and another 10MB or so for what I guess is kernel overhead. But it clearly can't be the cause for the slowdown on Ubuntu, because you have to install the Kubuntu variant to even get KDE. So, what causes it?

Welll, a quick check shows that on the stock Debian kernel, known to be best suited to workstation or small server use and not low latency gamer's desktopping, the kernel's timer frequency is 250Hz, ‘a good compromise choice allowing server performance while also showing good interactive responsiveness,’ whereas the "generic" Ubuntu kernel is set to 100Hz, ‘a typical choice for servers, SMP and NUMA systems with lots of processors that may show reduced performance if too many timer interrupts are occurring,’ which is not really the best for gaming… especially accelerated 3D gaming which is somewhat interrupt driven. My self-rolled kernel is set to 1000Hz, the recommended frequency for low-latency desktop systems.
It's not really going to make up all the difference, but consider you often want more than 100 interrupts per second just for the one game. If you're using Ubuntu, do yourself a favour and switch to the 'linux-preempt' kernel

[Znurre]

Originally Posted by Arafails

...

Very interesting, thanks for the info
What still puzzles me though is why Fedora performs so bad compared to the other distros, since it's using 1000Hz as default.

Originally Posted by http://duopetalflower.blogspot.com/2009/06/running-laptop-long-time-without-it.html

4. 1000 Hz --> faster than any preset frequency (in kernel configuration) and gives an interrupt of 1 ms timer approx, and very good for playing games, watching HD (High Definition) movies, editing multimedia, but very bad for power saving in your laptop, and can heat laptop considerably, but very good as a host for running virtualization. This is the default kernel timer frequency used by Fedora and considered as a good virtualization host

Also, Arch Linux is using 300Hz by default (maybe this is patched in the Chakra kernel though).

Originally Posted by http://duopetalflower.blogspot.com/2009/06/running-laptop-long-time-without-it.html

3. 300 Hz --> faster than 250 Hz, gives good power saving, and very good media playback, it is very good for watching high quality videos, Arch Linux kernel uses this by default and watching movies in Arch Linux is a pleasure. Also since the timer frequency is more, Arch Linux feels faster than Ubuntu (and of course, Arch Linux is optimized for i686 and uses a low latency kernel (a hard preemptive kernel))

[Arafails]

As it happens, for most applications a 250Hz/300Hz kernel timer is sufficient. They are good frequencies for watching PAL/NTSC media respectively. It's also good enough to run the old physics Quake3 games at 125fps and still have enough interrupts to run background tasks without to much issue. It's certainly fast enough for Regnum Online, and in most cases you probably won't notice a difference between 250, 300, and 1000Hz (and this applies to Unigine too). It's only the slow timer on Ubuntu that's concerning.

As for Fedora, apart from the fact that its final score is still with 100 of the top, so not a huge difference, the only explanation that jumps to mind is that PulseAudio is set up in a way that normally successfully claims real time scheduling when it's not on most other systems (and will often whinge that it's not, albeit where you can't see).