Hi!

Obviously X11 is significantly slower than the Windows UI. Why is that?
I have a GeForce2 MX board and read that the Linux drivers are
outperformed by their Windows counterparts (are they?). By architecture
X11 is certainly slower than Windows' UI but at least in my case the
difference is that dramatic that I really have problems to explain it.

Or is it that the CPU does most of the UI work while it's the GPU for
Windows/DirectX? Is it a matter of VRAM (I only have few windows and 1
desktop open when running Windows)? Would a GeForceFX 5200 noticably
speed up my desktop?

Timo

On Tue, 06 Jul 2004 21:57:53 +0200, Timo Nentwig staggered into the
Black Sun and said:
> Obviously X11 is significantly slower than the Windows UI.

Got hard data? Without some numbers, this statement is troll bait.

> I have a GeForce2 MX board and read that the Linux drivers are
> outperformed by their Windows counterparts (are they?)

Funny, people have reported higher framerates on Linux than on Windows
when running the evil binary-only nVidia X server and kernel module and
games that are native on Linux and Windows (like RTCW.)

> By architecture X11 is certainly slower than Windows' UI

More troll bait, unless you explain what you mean more precisely.

X servers run in userspace. Windows display drivers run in kernel
space. Running in kernel space reduces context switches, which
translates into speed boosts. The acceleration features X can take
advantage of may lag behind the acceleration features available on a
graphics card, too. Also, the X developers don't always have the full
specs to the graphics cards they're developing for, while the Windows
display driver writers always have full specs. That can make things
slower for X.

> in my case the difference is that dramatic that I really have problems
> to explain it.

Which X server are you using? For speed, nvidia > nv > VESA . You
should be using the evil binary-only nVidia X server and kernel module
if you want the fastest possible performance out of your graphics card.
I use that X server on my desktop with a comparatively ancient TNT2 card
and I've never had any problems with the speed of X.

> Or is it that the CPU does most of the UI work while it's the GPU for
> Windows/DirectX?

If you're using an accelerated X server (nv or nvidia, not VESA) then
the CPU should be minimally involved in drawing junk on the screen.

> Is it a matter of VRAM?

No.

> Would a GeForceFX 5200 noticably speed up my desktop?

Probably not, though it might make your 3D games run faster. Answer my
questions above and give specific details about any speed problems
you're having (like "It takes N seconds to open a new Nautilus window."
"Playing DVDs fullscreen with xine causes frame dropping.") and someone
will be able to offer suggestions.

--
Matt G|There is no Darkness in Eternity/But only Light too dim for us to see
Brainbench MVP for Linux Admin / mail: TRAP + SPAN don't belong
http://www.brainbench.com / Hire me!
-----------------------------/ http://crow202.dyndns.org/~mhgraham/resume

> Obviously X11 is significantly slower than the Windows UI. Why is that?
> I have a GeForce2 MX board and read that the Linux drivers are
> outperformed by their Windows counterparts (are they?). By architecture
> X11 is certainly slower than Windows' UI but at least in my case the
> difference is that dramatic that I really have problems to explain it.
Actually, most of the "slowness" in X comes from the different toolkits
(mostly GTK and QT) which don't use the X protocol very effectively and
not from X itself. Also, these toolkits have to simulate some effects
which X doesn't directly support (e.g. transparent windows) in software
instead of using hardware acceleration. One can hope that now after the
XFree fork, some more work will be done to enable the toolkits to
produce eye candy with better performance. Thus, at this time, on most
machines, X11 programs using GTK or QT will feel slower than the windows
GUI. However, this difference should not be "dramatic" on most
reasonably new machines if everything is configured properly.
Could you give some hardware specs of your machine ? Please also write
which distro and X server you are using and whether you installed
nvidia's binary video drivers.
Michal

Michał Kosmulski wrote:
> Could you give some hardware specs of your machine ? Please also write

Athlon XP 2200+, 1GB RAM, GeForce2 MX 400

> which distro and X server you are using and whether you installed

SuSE 9.1, XFree 4.3

> nvidia's binary video drivers.

Yes, NVIDIA drivers.

Dances With Crows wrote:
> On Tue, 06 Jul 2004 21:57:53 +0200, Timo Nentwig staggered into the
> Black Sun and said:
>
>>Obviously X11 is significantly slower than the Windows UI.
>
> Got hard data? Without some numbers, this statement is troll bait.

Nope. But just try Opera on Windows and on Linux. In this case the
difference is extreme.

> Funny, people have reported higher framerates on Linux than on Windows
> when running the evil binary-only nVidia X server and kernel module and
> games that are native on Linux and Windows (like RTCW.)

I read that too, but I refer to 2D/desktop performance. I assume that
it's Microsofts sophisticated DirectX utilization what makes their UI so
fast, i.e. the work is done by the GPU rather than the CPU.

> Which X server are you using? For speed, nvidia > nv > VESA . You

XFree 4.3 and NVIDIA drivers.

> Probably not, though it might make your 3D games run faster. Answer my

Are there serious 3D games for Linux?

> questions above and give specific details about any speed problems
> you're having (like "It takes N seconds to open a new Nautilus window."
> "Playing DVDs fullscreen with xine causes frame dropping.") and someone
> will be able to offer suggestions.

I don't have concrete benchmark data but everything feels much slower,
esp. redrawing of windows. On Windows you click and it's simply there. A
good example is opening/closing tabs or the sidebar in Opera. Or
eclipse, of course which made it neccessary to upgrade from an Athlon
1GHz to an Athlon XP 2200+ just in order to use it (and it's still slow).

> Athlon XP 2200+, 1GB RAM, GeForce2 MX 400

> Yes, NVIDIA drivers.

Hmmm. Pretty strange. With these specs you should be getting really
decent performance. I don't have any idea why it's as slow as you write.

Michal

Hallo!

>> Athlon XP 2200+, 1GB RAM, GeForce2 MX 400
>
>
>> Yes, NVIDIA drivers.
>
>
> Hmmm. Pretty strange. With these specs you should be getting really
> decent performance. I don't have any idea why it's as slow as you write.

I heard that the new sheduler in 2.6 together with a reniced XServer
(see result of "top") can degrade performance. Also disabled hard disk
DMA might reduce system performance. Perhaps there is a problem with AGP
access? Do you use the kernel or the XServer AGP driver? Can you see if
AGP 4x is used?

I have an X* 1800+ together with 512 MB and the same grafics card and
performance is good and I can play halflife under Wine in amazing speed :-)

--
Gruß...
Tim.

Timo Nentwig wrote:
> Hi!
>
> Obviously X11 is significantly slower than the Windows UI. Why is that?
> I have a GeForce2 MX board and read that the Linux drivers are
> outperformed by their Windows counterparts (are they?). By architecture
> X11 is certainly slower than Windows' UI but at least in my case the
> difference is that dramatic that I really have problems to explain it.

X11 may feel slow at some points, but that on machines with a lot less CPU
power than what you have. I got quite good preformance on an AMD Duron 700MHz
woth 512MB ram and GeForce2MX-400 and got enough fps on quake3 for me and
think I got even 50fps on Allied Assult running on WineX.


> Or is it that the CPU does most of the UI work while it's the GPU for
> Windows/DirectX?

As far as I know, DricetX isn't used for any desktop tools in Microsoft
Windows, so it wouldn't affect things, but as others has pointed out that some
of the cool effects in KDE/Gnome/Gnome2 is done in software instead of
hardware, but that will most likely to change in XFreee86 v5 or in the xorg fork.

> Is it a matter of VRAM (I only have few windows and 1
> desktop open when running Windows)? Would a GeForceFX 5200 noticably
> speed up my desktop?

I can say that the 2D stuff will not make any bigger changes between a GF2MX
and a GF4Ti, but for 3D stuff there is a major difference, but if you aren't
into games, then I do suggest to increase the amount of RAM in your system, it
will make the system to feel swift, regadles of your operating system.


//Aho

On 2004-07-07, Timo Nentwig wrote:

> I read that too, but I refer to 2D/desktop performance. I assume that
> it's Microsofts sophisticated DirectX utilization what makes their UI so
> fast, i.e. the work is done by the GPU rather than the CPU.

No. With an accellerated server like nVidia's or the nv server the grunt
work is done by the GPU just like in Windows. The speed difference may be
due to the fact that Microsoft runs the graphics drivers in kernel space
space rather than user space. This improves performance but at the risk
of a driver problem being able to crash the whole kernel rather than just
user processes.

--

-John (john@os2.dhs.org)

Hallo!

> No, the thing that _really degrades_ performance is that there are
> performance bottlenecks resulting from questionable implementation
> techniques in common graphics libraries like GTK and Qt that make it
> easy for application performance to "suck." (Look at the GTK release
> notes and see how often they mention performance improvements
> surrounding redraws...)

Is there a list of things one should do and one should not do? Best
practices or similar so one can make it right from the start?

--
Gruß...
Tim.

Tim Teulings wrote:
>> No, the thing that _really degrades_ performance is that there are
>> performance bottlenecks resulting from questionable implementation
>> techniques in common graphics libraries like GTK and Qt that make it
>> easy for application performance to "suck." (Look at the GTK release
>> notes and see how often they mention performance improvements
>> surrounding redraws...)
>
> Is there a list of things one should do and one should not do? Best
> practices or similar so one can make it right from the start?

There's no quick answer, alas...

Note that efforts on getting X libraries "right" have been ongoing for
years.
--
"cbbrowne","@","cbbrowne.com"
http://www.ntlug.org/~cbbrowne/advocacy.html
"I've seen estimates that 10% of all IDs in the US are phony. At least
one-fourth of the president's own family has been known to use phony
IDs." -- Bruce Schneier CRYPTO-GRAM, December 15, 2001

Centuries ago, Nostradamus foresaw when John Thompson would write:
> On 2004-07-07, Timo Nentwig wrote:
>
>> I read that too, but I refer to 2D/desktop performance. I assume that
>> it's Microsofts sophisticated DirectX utilization what makes their UI so
>> fast, i.e. the work is done by the GPU rather than the CPU.
>
> No. With an accellerated server like nVidia's or the nv server the grunt
> work is done by the GPU just like in Windows. The speed difference may be
> due to the fact that Microsoft runs the graphics drivers in kernel space
> space rather than user space. This improves performance but at the risk
> of a driver problem being able to crash the whole kernel rather than just
> user processes.

The "kernel part" _isn't_ what would improve performance. Indeed, it
is entirely likely having it in the kernel would degrade performance
somewhat because every interaction with the graphics system requires
the context switch of crossing the threshold between user processes
and kernel mode.

No, the thing that _really degrades_ performance is that there are
performance bottlenecks resulting from questionable implementation
techniques in common graphics libraries like GTK and Qt that make it
easy for application performance to "suck." (Look at the GTK release
notes and see how often they mention performance improvements
surrounding redraws...)

People used to bash X because Netscape Navigator was so slow; the
_true_ problem was that there was plenty of blame to distribute
liberally:

- X has its "coding challenges" that make it challenging for
developers to write efficient code;

- Motif has much the same, and adds in deeply entrenched
bugs-as-features;

- Netscape were not innocent, either; while they suffered from the
challenges of trying to write code that worked well atop Motif,
this provided enough distractions that Navigator had misfeatures
leading to it being a bloated slow pig of an application.

Some idiots would, of course, attribute this _totally_ to "X not being
in the OS kernel". That's what idiots are for... Drawing stupid
conclusions not based on any actual evidence...
--
If this was helpful, rate me
http://www3.sympatico.ca/cbbrowne/internet.html
"...I'm not one of those who think Bill Gates is the devil. I simply
suspect that if Microsoft ever met up with the devil, it wouldn't need
an interpreter." -- Nicholas Petreley, InfoWorld, Sept 16, 1996