Please write about NVidia's 2D performance problems

[Michael]

Hello,

Over the past two years the typical linux desktop has changed a lot.The XRender API has replaced the old X drawing model and composition introduced new visual possibilties.This also means that 2D drivers are now stressed much more and in different areas than 2-3 years ago.There are ongoing complains about poor 2D performance of NVidia GPUs, about 2 years ago it started with people complaining about slow text rendering with subpixel-antialiasing, but the more programs use the XRender api, the more complaints are posted.KDE4 which uses XRender a lot and also relies on more advanced feature is really slow (I would call it almost unuseable), also FireFox3 is no joy with nvidia's binary drivers. Often nouveau drivers with their EXA architecture offer better 2D performance than the binary drivers themself.There are also people calling nvidia to open-source their 2D driver, or at least provide specs to the nouveu project, because the think open-source projects could do it a lot better and are not that revenue-driven.Most of the discussions happen in the unofficial nvidia linux support forum:http://www.nvnews.net/vbulletin/forumdisplay.php?f=14It would be great if you could write a short arcticle about this topic, maybe it would change things to the better and make nvidia recognize that although people buy the cards because of 3D scores, they don't tolerate dog-slow 2D desktops.

Thanks
Regi

Anyone else interested? Anyone have their own experiences they would like to share on NVIDIA's 2D performance (or there the lack of)?

[Redeeman]

i believe you should write about it, they dont seem to give a damn about it themselves..

for instance, on 8xxx, openoffice is useless, im not kidding, its really really useless.

open OO impress, use one of its wizards to generate a sample presentation, and scrolling will litterally take SECONDS, it is really really useless.

Software rendering is faster.

[StringCheesian]

Am I the only one not having these problems? Firefox 3's scrolling performance on my 9600GT subjectively feels just as smooth as I remember Firefox 2 being on my old 7600GT.

I just tried "open OO impress, use one of its wizards to generate a sample presentation". Scrolling was smooth only while the entire presentation slide was visible. If I scroll such that part of it goes out of view and them back in, it does get a little choppy - like about 5 to 10 frames per second. But still nowhere near taking whole seconds...

Gentoo Linux
xorg-server version 1.3.0.0
nvidia binary driver version 173.14.09
Athlon 64 X2 4600+ overclocked to 2.52Ghz
nVidia GeForce 9600GT

[Redeeman]

well... i cannot comment as to whether it works for you, but all i can say is, that for the ~5 boxes i've seen with 8xxx hardware, they've all had the problems..

[snaury]

Anyone else interested? Anyone have their own experiences they would like to share on NVIDIA's 2D performance (or there the lack of)?

Yes, I have 8600M GS in my laptop and have an extremely bad performance with Ubuntu 8.04. First of all it's slow, especially text scrolling in gnome-terminal. When using Midnight Commander, trying to view/edit files is very painful. nvidia-settings -a InitialPixmapPlacement=2 (-a GlyphCache=1 doesn't work for some reason, there's no such setting in my case) helps with that, but somehow it manages to totally screw all the rest of my system, especially when I switch between windows. It's as if active window has cached fonts and everything is fast, but as soon as I switch to another window all the caches are dropped and nvidia starts rendering it cold. Note that all of this happens even without compiz. Compiz makes it worse. Compiz+Emerald make it unbearable, the delays are two and more seconds, and what's the most funny part, become noticeable without InitialPixmapPlacement=2.

The latest 173.14 and 177.13 drivers added another bug to the mix: in gnome-terminal (and in Midnight Commander) when you edit a file and move your cursor it often leaves "trails" in its old positions and not always draws itself in its new position. Editing files is even more pain, you don't even know where you have your cursor at the moment! Press Ctrl+L and suddenly you could notice it's in the wrong line/column. Enabling TripleBuffering helps, but not always. It's still happening sometimes.

Plus I've got an impression that 173.14/177.13 drivers have even more performance problems than 169.12 that is shipped with Ubuntu 8.04.

Additionally, for quite some time I was attributing the slowness to Firefox 3.0 and Ubuntu 8.04 (I had performance problems on windows with early firefox alphas) and have been thinking anything from "why the hell firefox on windows works so much faster than on windows" to "what the hell did they break in Ubuntu 8.04 this time to make Linux so unbearably slow". That was until I switched to nv driver, which had absolutely no performance problems as far as I've seen. Of couse it doesn't have 3d acceleration (and some dithering problems) though, which is a showstopper for me and forces me to get back to Vista.

Even worse is that a lot of people who scream "nvidia drivers work perfectly for me" might not even now they have the problem. They might think this is how it is supposed to work. You can spot performance problems only in the extremes on when comparing it with something else. And if they had neither it looks as if it's ok. :(

All this means that it might force me and some of my friends (who don't even care for Linux but might want to try it some time in the future) to avoid NVIDIA like a fire next time. Some already did and decided to wait for the next ATI for their next upgrade. Their reputation of good Linux support turned out to be fake and I hope that more and more people will realise that.

Plus what they don't realise is that their bad reputation on Linux causes their Windows users to reconsider buying from them next time. Just because they might want to try Linux. I can't imagine their marketing doesn't care about that.

[Kano]

Did you ever try something else besides from Ubuntu? I can not see problems with Kanotix where X is still Xorg 7.1.1 and Firefox - there called Iceweasel is 2.0.x by default.

[bacatta]

I've got no such problems with my 7600GT but I'm going to buy a 9600GT... is it affected by this regression ?

[Redeeman]

I've got no such problems with my 7600GT but I'm going to buy a 9600GT... is it affected by this regression ?

as far as i know, ALL 8xxx+ is

[Tsiolkovsky]

Anyone else interested? Anyone have their own experiences they would like to share on NVIDIA's 2D performance (or there the lack of)?
I'm absolutely interested what nVidia has to say about this. I become more aware of this problem whe I started testing KDE 4.1 beta 1. I couldn't figure out why my poor laptop with integrated ATI graphics and FOSS drivers is runing just fine and on my dual core machine with nVidia 7600GS everything is so slow. Digging further into this people also told me that KDE 4 is built on Qt 4 toolkit and Qt 4 relies much more on modern XRender architecture which appears to be dog slow on nVidia. Later I found out about a simple XRender Benchark (http://people.freedesktop.org/~zack/xrenderbenchmark.tar.bz2) and this one only proved how many times faster integrated ATI is compared to mVidia and their bad drivers. This has come to get out so more people is avare of this and nVidia starts doing something to correct this bad performance.

[chromis]

I have a 8600GTS running fine here. It runs on an old X server though: 6.8.2, and I don't run anything fancy. Mostly Gtk2 and some KDE3 applications (gwenview being to most 2D-graphics intensive I guess).

I have a mainboard with a K8M890 chipset, and a 3200+, pretty old stuff also.

Perhaps it's good to know what hardware the people have who experience these problems. There may be a relation to chipset/platform and these performance issues.

About X-render, I found this article on Phoronix: http://www.phoronix.com/scan.php?page=article&item=934&num=1 ; dated 3 december 2007. I can't find the benchmark with the E's though.

An old 2003 article on slashdot ( http://developers.slashdot.org/article.pl?sid=03/08/16/0034235 ) may also be of interest. I think it mentions the benchmark used in the phoronix article.

[chromis]

I found render_bench, the benchmark application with the E's. it's located right here: http://www.rasterman.com/files/render_bench.tar.gz ; these are my results:

EDIT: some extra info about the test system in question:

Nvidia board: 8600GTS
X Window System Version: 6.8.2,
driver version: 173.14.09
kernel: 2.6.24.2chromis
.config.: http://www.chromis.nl/vanalles/config.gz
chipset & cpu: VIA K8M890, AMD 939 3200+
running in 32 bit!

Available XRENDER filters:
nearest
bilinear
fast
good
best
Setup...
*** ROUND 1 ***
---------------------------------------------------------------
Test: Test Xrender doing non-scaled Over blends
Time: 0.068 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 0.080 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.224 sec.
*** ROUND 2 ***
---------------------------------------------------------------
Test: Test Xrender doing 1/2 scaled Over blends
Time: 0.031 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 0.031 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.110 sec.
*** ROUND 3 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 0.067 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 0.069 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 2.140 sec.
*** ROUND 4 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 0.066 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 0.069 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 1.466 sec.
*** ROUND 6 ***
---------------------------------------------------------------
Test: Test Xrender doing general nearest scaled Over blends
Time: 0.117 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 0.122 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 3.220 sec.
*** ROUND 7 ***
---------------------------------------------------------------
Test: Test Xrender doing general smooth scaled Over blends
Time: 0.118 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 0.122 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 5.597 sec.

[Max Spain]

It's also one of the tests in PTS, though I forgot which (maybe in x11perf).

[snaury]

Did you ever try something else besides from Ubuntu? I can not see problems with Kanotix where X is still Xorg 7.1.1 and Firefox - there called Iceweasel is 2.0.x by default.

Well, all I've been trying lately are latest Ubuntu and Sabayon: both must be using recent xorg, so it doesn't count. And even if it works under Xorg 7.1, it might be good for some people, but not for me. Xorg 7.3 was released in September 2007, that's 9 months for NVIDIA to adapt.

Also I just noticed something. There must have been something with the way I was installing the drivers, after latest attempts I found that I can finally set GlyphCache=1, though I can't say how much it helps.

However there's something else I found. When I was using firefox on 173.14.09 with InitialPixmapPlacement=2 and GlyphCache=1 I stumbed upon a relatively big picture that firefox had to scale down. This turned into almost ten seconds of work at which time gnome didn't even respond to my clicks. Just to check it I tried this with nv (everything was fast) as well as with InitialPixmapPlacement=1 (scaling was fast as well).

I don't know how firefox scales pictures (and what strange effect is at play here), but 5-10 seconds for 842x1032 is a little over the edge for me. :-/

[snaury]

I found render_bench, the benchmark application with the E's. it's located right here: http://www.rasterman.com/files/render_bench.tar.gz ; these are my results:

Looks like it's magnitudes slower for me:

Available XRENDER filters:
nearest
bilinear
convolution
fast
good
best
Setup...
*** ROUND 1 ***
---------------------------------------------------------------
Test: Test Xrender doing non-scaled Over blends
Time: 0.322 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 0.567 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.305 sec.
*** ROUND 2 ***
---------------------------------------------------------------
Test: Test Xrender doing 1/2 scaled Over blends
Time: 0.195 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 0.201 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.114 sec.
*** ROUND 3 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 0.557 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 0.568 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 3.674 sec.
*** ROUND 4 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 0.164 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 0.169 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 1.632 sec.
*** ROUND 6 ***
---------------------------------------------------------------
Test: Test Xrender doing general nearest scaled Over blends
Time: 0.314 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 0.322 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 2.801 sec.
*** ROUND 7 ***
---------------------------------------------------------------
Test: Test Xrender doing general smooth scaled Over blends
Time: 0.317 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 0.323 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 12.060 sec.

[borgus]

Here are my results on Gentoo with the latest NVIDIA drivers at the time of this writing:

NVIDIA Board: 8800 GTS 512MB (G92)
X Window System Version: 7.2 (X.Org V11, R0, Rel. 1.3)
Driver Version: 173.14.09
Kernel: 2.6.25-gentoo-r5
.config.: http://pastebin.com/f4152601c
Chipset, CPU: Intel P35, Intel C2D E6850
Running in 64-bit!

Available XRENDER filters:
nearest
bilinear
convolution
fast
good
best
Setup...
*** ROUND 1 ***
---------------------------------------------------------------
Test: Test Xrender doing non-scaled Over blends
Time: 0.058 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 0.089 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.252 sec.
*** ROUND 2 ***
---------------------------------------------------------------
Test: Test Xrender doing 1/2 scaled Over blends
Time: 0.028 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 0.035 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.090 sec.
*** ROUND 3 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 0.044 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 0.058 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 2.146 sec.
*** ROUND 4 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 0.050 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 0.052 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 0.751 sec.
*** ROUND 6 ***
---------------------------------------------------------------
Test: Test Xrender doing general nearest scaled Over blends
Time: 0.054 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 0.063 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 1.822 sec.
*** ROUND 7 ***
---------------------------------------------------------------
Test: Test Xrender doing general smooth scaled Over blends
Time: 0.061 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 0.063 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 6.888 sec.

Given chromis' results, that sounds pretty much like where it should be. So it's fast here too, if I'm interpreting it right.

[chromis]

I have altered my benchmark post, since I think it is wise to give some more detailed information about the system where the benchmark is run on.

For each posted benchmark, If we know the board type, the X window system version, the driver version, kernel version, .config of the kernel (you may get yours from /proc/config.gz if you have enabled that particular feature in your kernel), chipset & cpu and whether you run in 64-bit or 32-bit mode, then we may be able to find an actual cause.

Anyone a suggestion which metric would also be of interest?

That would be very nice, to have some evidence and a cause for this problem. :)

snaury,

Since your performance does seem indeed magnitudes slower, what is your particular configuration?

[Linuxhippy]

Well, I have 3 PCs with nvidia cards:
GF488Go Laptop
PC - GF5200
PC - GF6600

Because of nvidia's legacy drivers, I don't get any 2D enhancements for the first two cards, and even the 6600 won't be fully supported. (there are statements that some specific options will only be enabled on GF8xxx hardware).
Performance of the GF488Go and the GF5200 were quite ok, as long as I don't touch KDE4 or FireFox3. Maybe its because they are AGP, and thats friendlier to software-fallbacks.
On the 488 I found a bug leading to rendering-errors with all java-programs and was told that they won't fix it in a legacy driver.
The 6600 was never a lot of fun, especially considering how powerful this hardware could have been with proper drivers.

My new laptop is Intel-945GM based. With XAA it offers the best 2D experience I have ever seen, while EXA is currently getting better and better.

Yes, indeed, I would be happy to see some news, or maybe contacts with the development team.

lg Clemens

[sid350]

I have no problems with Palit Sonic+ GF8600GT

[lenrek]

I have no problems with Palit Sonic+ GF8600GT

I am using GeForce 8500 GT, no problem here as well.

But, I do like to read more about what others are facing.

[lenrek]

...
My new laptop is Intel-945GM based. With XAA it offers the best 2D experience I have ever seen, while EXA is currently getting better and better...

I know this is out of topic, but I can't help to comment. I never quite get XAA working for my Intel 945GM. In fact, if I only have the config of Driver "intel" in my xorg.conf file (the rest are VendorName, BoardName and BusID, which should not affect its performance), It enable EXA by default.

Since, by default the intel driver using EXA, we should assume EXA is better than XAA? No?

Anyway, if XAA works for you, I guess is fine.

[mlau]

I am using GeForce 8500 GT, no problem here as well.

But, I do like to read more about what others are facing.

8600GT and 8800GT. I have ~12 kde terminals open; after a few hours the
scrolling compiler output gets slower, which I can fix by setting
IPP=2 and GlyphCache=1 with nvidia-settings. Repeat every few hours
(with a cronjob).

I didn't notice any scrolling slowdowns in firefox or any other apps I
use.

[grigi]

On my 7800GT desktop the 2D graphics gradually slows down. It would take atleast 10 hours of heavy usage before the slowdown becomes noticeable, it only gets worse until I restart X.

The situation does not happen when using KDE3.5's compositing manager, and is worst when using Compiz.

Running X 7.2 with X server 1.3, and 169.something drivers (although this was the case since about the ~90 or ~100 drivers)

To get around the problem I try to restart X every day.

[lenrek]

Hmm... I decided to give it a try, based on what I read so far in this thread.

Currently, I am running total 26 Gnome-Terminalx in my desktop. 14 of them echoing "fortune -s" string, in every 1 seconds. 2 of them running 'top' command. So far, I have been running them about 18 hours. No slowness of any kind is noticeable. Neither does any of my applications (like firefox or OpenOffice Writer) has any problem.

Before I start doing the experiment, the system has been running for > 4 days (this desktop PC is also a tor router, so most of time, I let it run continuously, even when i am not using). My desktop is Dropline GNOME, but using Compiz-Fusion instead of Metacity. So far, no system freeze of any kind.


System info:
$> uname -simpro
Linux 2.6.25.10 i686 Intel(R) Pentium(R) D CPU 2.66GHz GenuineIntel GNU/Linux

$> glxinfo | grep renderer
OpenGL renderer string: GeForce 8500 GT/PCI/SSE2

$> head /var/log/Xorg.0.log

X Window System Version 1.3.0
Release Date: 19 April 2007
X Protocol Version 11, Revision 0, Release 1.3
Build Operating System: Slackware 12.0 Slackware Linux Project
Current Operating System: Linux orion 2.6.25.10 #1 SMP Sat Jul 5 23:36:01 SGT 2008 i686
Build Date: 09 May 2007


However, IIRC, it did freeze when GL's base screensaver was enabled (that happened a few weeks ago). But, is not something that I would concern much, since I am not that fancy of screensaver. I have not enabled screensaver since then.

Anyway, will let it run for a few more days...

[snaury]

X Window System Version 1.3.0
Well, to clarify, I'm on Ubuntu Hardy, which uses X.Org X Server 1.4.0.90, and gnome rendering is really slow here:

$ glxinfo|grep renderer
OpenGL renderer string: GeForce 8600M GS/PCI/SSE2

[jeffro-tull]

I've got a 6200 something-er-other in my tower. I bought it because I kept hearing how the drivers "Just Worked (TM)" in Linux. Since I'm not much on 3D, I didn't mind the ~1500 fps in glxgears. And before I started testing KDE4 last fall (and before openSUSE pushed Firefox 3 on me), the 2D stuff was just fine.

But when it comes to KDE4 and/or Firefox 3, my laptop (X1300 graphics) literally runs circles around my tower. There's such a huge performance discrepancy that I went back to KDE3.5.9 exclusively on my tower, and only use Firefox when there's a website that gives both Konqueror and Opera fits.

[BhaKi]

I'm using GeForce 7300 with KDE-3.5.9 and nVidia's proprietary driver.
I did face some initial problems with 2D. But after I made the following changes to my /etc/X11/xorg.conf, most of the problems went away:

Option "UseEvents" "on"
Option "OnDemandVBlankInterrupts" "on"
Option "TripleBuffer" "on"

[lenrek]

I stop the test, since I can't seem to face the same problem as you guys. Maybe is the Xorg version I am using, since mine is 1.3.0.

[Mota_boy]

I`m Using opensuse 11 with KDE3.5.9 and 8800gts ( g80 ),
2D and 3D is working with no problem at all, i know this is not a issue but i wanted tell that on compination ( Distro+desktop+drivercard ), what i use, works fine and i think many other users have working system too, why thiss so, not 100% sure.

[Kano]

I guess KDE 4 creates some problems and in some cases Firefox 3.

[Mota_boy]

Thats true kano, on opensuse i prefer much more KDE3 than 4, Firefox 3 have work fine, how it does work on other people is different thing..
Even i heard some problem on ubuntu/kubuntu people too, meaning KDE3 or 4 , situation, that 3 works better than 4.

[birdie]

I'm a semi-happy user of NVIDIA 8800GT GPU. While in Windows everything works like a breeze, in Linux everythings crawls to the complete freeze.

KDE 4.x window resizing and other operations look like a slide show for me. My CPU usage jumps to 100% on many web pages in Firefox 3, smooth scrolling is sometimes unbearable (Just open any slashdot.org big enough discussion thread).

A simple rotating Firefox 3 throbber consumes around 40% of 2600MHz Athlon 64 CPU.

There's really nothing to be said. Proprietary NVIDIA drivers do NOT support XRender acceleration on 8XXX and later GPUs - so there's really nothing to talk about. For the first time in my life I'm now thinking of dumping my NVIDIA GPU in favour of ATI solution - since I'm eager to upgrade to KDE 4.1 and I ... can not.

[Kano]

I dont think you loose anything with the stable KDE 3.5. You still can run KDE 4 apps - you don't have to use it as desktop.

[birdie]

I dont think you loose anything with the stable KDE 3.5. You still can run KDE 4 apps - you don't have to use it as desktop.

I'm not interested in running a couple of KDE4 applications in KDE3.5.x - it sounds a bit ridiculous, I want to migrate to KDE4 altogether. Besides there's a burden of running simultaneously a set of qt3/kde3 and qt4/kde4 libraries.

I do want to hope that NVIDIA will do something about that in the nearest future, although the prospects of this are quite unfortunate taking into consideration their kinda mishap with GTX2XX GPUs and $200 million loss related to faulty notebook parts.

[mgc8]

I have two current nVidia machines (bought specifically after the horrible experience from Ati):

1. Core2Duo laptop w/ 8600M GT
- Debian Sid, with all the updates, including nvidia-drivers 173.14.09.
- KDE 3.5.9, Firefox 3.0
- compiz
=> things run ok, although some actions like maximizing a window are painfully slow, scrolling on the other hand could be better but is quite decent. The latest drivers seem slower than the 171 version, which had other problems. All-in-all, the desktop is usable but feels slower than it should be.

2. Core2Quad desktop with a GTX 260
- Ubuntu 8.10 II, with all the updates, including nvidia-driver 177.something
- KDE 4.1beta, Firefox 3.0
- compiz
=> everything is silky smooth, much better than on the laptop, including compiz effects, maximizing, scrolling, etc. I have them side-by-side so I can compare. Maybe the latest drivers are better is it because of the faster chip? Anyway, it seems not *all* KDE4/FF3 & nVidia combos are bad, in my case it's the older version having problems :)

This being said, I would very much like to see proper FOSS support from nvidia, at least on their older cards, instead of those "legagy" drivers which are starting to become quite numerous these days (do we really need dozens of "legacy" driver versions?)

Oh, and for a little offtopic -- I also have another computer with Intel G35 on-board. It's been the most unstable one of the bunch when it comes to playing videos, and I'm not even running compiz or anything 3D on it... so much for having alternatives :-/

[mpcd]

I've been following this thread and it seems to offer some interesting things to try, notably the new GTX200 series beta drivers and the PixmapCacheSize switch.

http://www.nvnews.net/vbulletin/showthread.php?t=115916

[energyman]

this was with a 8600GT on gentoo. CPU AMD X2 6000+, 4gb ram. nf520 board.

with kde 4.1 beta/rc/final:
Resizing windows: very slow
scrolling: jerky and slow.
after a right click on the desktop it could take up seconds until the menu appeared - and then it wasn't even usable - that was a second later.
kmenu appearing slowly
when closing a window a grey, half transparent shadow of it remained on the desktop (only after some hours of usage).
when scrolling through pictures in gwenview and one was manually set to 100% there was a big chance the next picture would show garbage at first.

with AMD HD3870:
resizing windows still slow
scrolling, faster but a lot of times not perfect.
desktop menu pops up instantly - and is usable from the first moment
kmenu pops up instantly and is usable from the first moment
no shadows when a window is closed
picture viewing in gwenview without any problems.
watching videos not as good as with nvidia.

I tried: xrender, opengl+texture from pixmap, sharedmen, KWIN_NVIDIA_HACK=1, nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1, different xorg.conf settings.

And nothing really helped. If there were improvements somewhere - like scrolling, resizing or menus became worse.

[Xanikseo]

I think I may have found the magical solution to the 2D performance problems. Look what I got with gtkperf with 1000 iterations on my 8600gts:

gtkperf -c 1000
GtkPerf 0.40 - Starting testing: Sat Aug 9 20:30:55 2008

GtkEntry - time: 0.11
GtkComboBox - time: 8.94
GtkComboBoxEntry - time: 7.30
GtkSpinButton - time: 1.08
GtkProgressBar - time: 0.81
GtkToggleButton - time: 5.00
GtkCheckButton - time: 3.70
GtkRadioButton - time: 3.99
GtkTextView - Add text - time: 23.73
GtkTextView - Scroll - time: 3.37
GtkDrawingArea - Lines - time: 1.96
GtkDrawingArea - Circles - time: 2.20
GtkDrawingArea - Text - time: 2.12
GtkDrawingArea - Pixbufs - time: 4.03
---
Total time: 68.34


These options are taken from loads of different places (mainly different posts on NVnews), and suddenly I struck gold with a special combination. This may look a little overkill, but I'm not sure which options are useless, and I don't think I will try to find out; if it ain't broke, don't fix it. Here they are:

Enter this into the terminal:
nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1 -a SyncToVBlank=0
And I KNOW some of you will complain about bad performance with that, I got some problems too until I added this to my xorg.conf:

Option "PixmapCacheSize" "2500000"
Option "UseEvents" "on"
Option "OnDemandVBlankInterrupts" "on"
Option "backingstore" "true"

I hope it works for you! Everything happens instantly, all problems have gone, I didn't realise how fast my computer could be!

[energyman]

been there, done that, no real improvement.

[Xanikseo]

Really? Have you tried those exact options? If so, that's annoying, perfect here. Then again, I didn't have windows taking seconds to maximise before.. oh well.


What are your specs?

[energyman]

were - amd x2 6000+ 4gb ram, 8600gt.

[etymxris]

nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1 -a SyncToVBlank=0

Just that by itself helps me a lot, haven't restarted X to try the other settings yet. Is there an easy way to get the settings applied every time I start X?

[Xanikseo]

The easiest way is to add a session to the sessions applet in the preferences sub-menu, and use that line as a command. It would run that command as soon as you log in.

[Kano]

As user:

mkdir -p ~/.kde/Autostart
echo nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1 > ~/.kde/Autostart/nvidia.sh
chmod +x ~/.kde/Autostart/nvidia.sh

You could discuss if you really want SyncToVBlank=0, because scaled youtube videos look much better with vsync on with latest flash. That only works with 177.x . Also the vsync option is stored in .nvidia-settings-rc, so you could start

nvidia-settings -l

too in order to load all settings from the configfile (the first two options are not stored there as they are not accessable via the gui).

[oyvind]

I've got an 8600GT (purchased just recently), and even with all tweaks (initial pixmap placement, glyphcache and 177.X beta drivers), I'm definitely not happy with 2D. So I really hope nvidia means business when they say they're working on improving it. :(

[adamj]

I think Firefox under Vista is silky smooth for on an 8800 GTX. It doesn't have an embarassing 4 second pause when rescaling images or zooming in Google maps, either.

[gummybearx]

just wanted to sound off... I have a 7900gt in opensuse 11 64bit and 173.14.12 drivers, and experience no problems with 2d or 3d. im using c-f as well, no memory leaks :-)

[Linuxhippy]

The last two driver releases added excellent RENDER support, even more complete than many open-source drivers.

The last driver without proper RENDER accaleration remains fglrx for now :-/

[kraftman]

Did you ever try something else besides from Ubuntu? I can not see problems with Kanotix where X is still Xorg 7.1.1 and Firefox - there called Iceweasel is 2.0.x by default.

Of course. I tried many distros and 2D performance was always awfull with nvidia binary blobs (GF 6800XT AGP). I don't have big experience with other cards but on Intel card that I had in laptop 2D performance was perfect (including flash animations!!). If someone thinks that his nvidia card is not affected probably CPU does dirty work which should be done by GPU. There are some improvements with the latest drivers, but mainly in Firefox performance. KDE4 is still dog slow for me. 2D is faster in some situations when I use vesa!

EDIT:

There are performance problems with 5xxx series cards on win xp too. Scrolling in firefox is sometimes horrible.

[unimatrix]

Yes, I have 8600M GS in my laptop and have an extremely bad performance with Ubuntu 8.04. First of all it's slow, especially text scrolling in gnome-terminal. When using Midnight Commander, trying to view/edit files is very painful. nvidia-settings -a InitialPixmapPlacement=2 (-a GlyphCache=1 doesn't work for some reason, there's no such setting in my case) helps with that, but somehow it manages to totally screw all the rest of my system, especially when I switch between windows. It's as if active window has cached fonts and everything is fast, but as soon as I switch to another window all the caches are dropped and nvidia starts rendering it cold. Note that all of this happens even without compiz. Compiz makes it worse. Compiz+Emerald make it unbearable, the delays are two and more seconds, and what's the most funny part, become noticeable without InitialPixmapPlacement=2.

The latest 173.14 and 177.13 drivers added another bug to the mix: in gnome-terminal (and in Midnight Commander) when you edit a file and move your cursor it often leaves "trails" in its old positions and not always draws itself in its new position. Editing files is even more pain, you don't even know where you have your cursor at the moment! Press Ctrl+L and suddenly you could notice it's in the wrong line/column. Enabling TripleBuffering helps, but not always. It's still happening sometimes.

Plus I've got an impression that 173.14/177.13 drivers have even more performance problems than 169.12 that is shipped with Ubuntu 8.04.

Additionally, for quite some time I was attributing the slowness to Firefox 3.0 and Ubuntu 8.04 (I had performance problems on windows with early firefox alphas) and have been thinking anything from "why the hell firefox on windows works so much faster than on windows" to "what the hell did they break in Ubuntu 8.04 this time to make Linux so unbearably slow". That was until I switched to nv driver, which had absolutely no performance problems as far as I've seen. Of couse it doesn't have 3d acceleration (and some dithering problems) though, which is a showstopper for me and forces me to get back to Vista.

Even worse is that a lot of people who scream "nvidia drivers work perfectly for me" might not even now they have the problem. They might think this is how it is supposed to work. You can spot performance problems only in the extremes on when comparing it with something else. And if they had neither it looks as if it's ok. :(

All this means that it might force me and some of my friends (who don't even care for Linux but might want to try it some time in the future) to avoid NVIDIA like a fire next time. Some already did and decided to wait for the next ATI for their next upgrade. Their reputation of good Linux support turned out to be fake and I hope that more and more people will realise that.

Plus what they don't realise is that their bad reputation on Linux causes their Windows users to reconsider buying from them next time. Just because they might want to try Linux. I can't imagine their marketing doesn't care about that.

Exactly the same in my case. I'm SOOO glad somebody pointed this out.

I'm reviving this thread, because this urgently needs to get resolved. It's driving me mad, and giving me dark thoughts... like moving back to Windows...

[mgc8]

You revive an year-old thread to post what accounts to a "mee too"?! Why? You could at least specify which configuration gives you the trouble, otherwise that was a pretty useless post.

If you are still using Ubuntu 8.04 as in that thread, understand that it's almost two years old and both X.Org and the nVidia drivers have progressed immensely since. I don't have any problems with any of my 4 nVidia-based computers, all running up-to-date versions of software (KDE 4.2) and drivers. If you do have these problems with recent versions, posting in the nVidia forums would certainly be better since you'd've found a possibly rare bug.

[unimatrix]

You revive an year-old thread to post what accounts to a "mee too"?! Why? You could at least specify which configuration gives you the trouble, otherwise that was a pretty useless post.

If you are still using Ubuntu 8.04 as in that thread, understand that it's almost two years old and both X.Org and the nVidia drivers have progressed immensely since. I don't have any problems with any of my 4 nVidia-based computers, all running up-to-date versions of software (KDE 4.2) and drivers. If you do have these problems with recent versions, posting in the nVidia forums would certainly be better since you'd've found a possibly rare bug.

I have had these problems with EVERY ubuntu version and with every nvidia driver version.
I have posted to nvnews: http://www.nvnews.net/vbulletin/showthread.php?t=135027 ( even videos showing the problem ) and ubuntuforums: http://ubuntuforums.org/showthread.php?t=1257784 and http://ubuntuforums.org/showthread.php?t=1196716
Most of those are firefox-oriented, because the problem is most visible there, but obviously it's system-wide.

And I have tried absolutely everything. Phoronix is my last hope, that is why I am now posting here.

[mgc8]

Now that's more like it... however, it is still not clear to what extent your problems are identical to what you quoted. After looking through the linked threads, it seems you ran browser benchmarks that show difference under Linux (which is to be expected) yet the actual "feel" of the system as a whole was not unuseably slow, just "not right".

I understand that "speed" of an interface is a very subjective matter, but what this thread was discussing was dead-slow behaviour of applications -- like a window that is impossible to scroll because of the speed, or an app taking many seconds to maximize or minimize. Your experience on the other hand seems more about things being slower than in windows, taking longer to show up or load or having slower benchmark numbers; while that is indeed a real problem with Linux atm., it affects all video configurations to some extent (see the recent "Intel graphics regressions" articles on this very site) not just nVidia and as such deserves a sepparate discussion.

The 185/190 driver revisions have solved most issues that people were experiencing, so if you still have extremely slow behaviour then it certainly is a bug, perhaps related to a certain board/bios/etc. configuration, and the thread on NV forums should help fix that. By the way, did you try the same tests with the nv driver (or nouveau) to get a direct comparison?

As to the general slowness of a Linux Desktop vs. the same system running *win, basically you can blame the X System -- however, keep in mind that for all it's weaknesses, you can use X in ways that are simply impossible with other OS'es: try running Firefox on your server with the window displaying on a thin client for example... It's all about tradeoffs.

[unimatrix]

Now that's more like it... however, it is still not clear to what extent your problems are identical to what you quoted. After looking through the linked threads, it seems you ran browser benchmarks that show difference under Linux (which is to be expected) yet the actual "feel" of the system as a whole was not unuseably slow, just "not right".

I understand that "speed" of an interface is a very subjective matter, but what this thread was discussing was dead-slow behaviour of applications -- like a window that is impossible to scroll because of the speed, or an app taking many seconds to maximize or minimize. Your experience on the other hand seems more about things being slower than in windows, taking longer to show up or load or having slower benchmark numbers; while that is indeed a real problem with Linux atm., it affects all video configurations to some extent (see the recent "Intel graphics regressions" articles on this very site) not just nVidia and as such deserves a sepparate discussion.

The 185/190 driver revisions have solved most issues that people were experiencing, so if you still have extremely slow behaviour then it certainly is a bug, perhaps related to a certain board/bios/etc. configuration, and the thread on NV forums should help fix that. By the way, did you try the same tests with the nv driver (or nouveau) to get a direct comparison?

As to the general slowness of a Linux Desktop vs. the same system running *win, basically you can blame the X System -- however, keep in mind that for all it's weaknesses, you can use X in ways that are simply impossible with other OS'es: try running Firefox on your server with the window displaying on a thin client for example... It's all about tradeoffs.

First of all I have to say I'm so glad to get a reply from someone who actually knows what they're talking about.

I will be replying to your last paragraph because that is the case here. That is basically the whole problem - The same system is being slower and less responsive on Xorg than on *win. I suspected Xorg was to blame. I have tried it with nv too, and it wasn't any better. Possibly even worse. (but seriously, nv is a piece of s**t, even vesa has better 2d rendering performance)
So my question is, is there anything that can be done about it?
If not, then I'm afraid Linux will never be a successful desktop OS. It is just too frustrating to have all these tiny delays when minimizing/maximizind, opening/closing windows and especially the scrolling lag. I'm seriously thinking of switching back to Windows, because this is really getting on my nerves. (Not a threat, just being honest here. I love linux, but I simply cannot live with this.)

[nanonyme]

It is just too frustrating to have all these tiny delays when minimizing/maximizind, opening/closing windows and especially the scrolling lag.How long are they? Sub-second pauses or longer? Sub-second pauses are plenty frustrating too, just wondering if you're experiencing the same I am.

[unimatrix]

How long are they? Sub-second pauses or longer? Sub-second pauses are plenty frustrating too, just wondering if you're experiencing the same I am.

Yes, definitely sub-second.

A nice example would be unminimizing a window in Metacity. You can literally see how the window border appears with a delay, after the window content.
Or scrolling the music collection in Rhythmbox. Scroll clearly lags behind the mouse cursor.

[nanonyme]

Right. That'd be similar to what I've seen on a different card and drivers too so might be X itself. I found compositing helped a bit in some latency cases but made some others a bit slower.

[unimatrix]

Right. That'd be similar to what I've seen on a different card and drivers too so might be X itself. I found compositing helped a bit in some latency cases but made some others a bit slower.

Yes, same in my case. Compositing greatly reduces CPU usage in plenty of cases (in result things run smoother), but also reduces performance in other cases. See my GtkPerf test results: http://www.nvnews.net/vbulletin/showpost.php?p=2083922&postcount=26

EDIT: Since this is most likely a problem with X, I really wish I could try out Wayland. Anyone know if it works with nvidia drivers?

[unimatrix]

I made a video of a messagebox appearing: http://dl.getdropbox.com/u/121855/pub/blip.avi
See how some parts of it appear with an annoying delay? It's not that bad, but it's annoying.

Here's the video in slow motion: http://dl.getdropbox.com/u/121855/pub/blip_slow_motion.avi
(If it doesn't work in VLC, thank Avidemux).

PS: No compiz in this video. Compiz hides this with its animations that not many of us want because they represent even larger delays.

[nanonyme]

No compiz in this video. Compiz hides this with its animations that not many of us want because they represent even larger delays.It's different even without the animations though. If they're using double-buffering (which I'd assume they do), they can draw the window completely in backbuffer, then flip it into frontbuffer and user only sees it a) not drawn b) completely drawn.

[mgc8]

I suspected Xorg was to blame. I have tried it with nv too, and it wasn't any better. Possibly even worse.

Well, in that case this is not necessarily a problem with the nVidia drivers, it is something that has different causes. Discussing it in this thread is not really appropriate since it would make people blame one part of the system when others are responsible.

So my question is, is there anything that can be done about it? If not, then I'm afraid Linux will never be a successful desktop OS

Everyone has different expectations when it comes to using a computer, for some it's applications that matter, for others the eye candy, some value responsiveness while others place number crunching performance on top.

Up to this stage, GNU/Linux as an OS was never seriously developed for the Desktop, there are only some attempts from distros like Ubuntu or Suse to improve performance, but that's it. The kernel was originally only optimized for server use, nowadays there is serious work going on to bring desktop features like KMS and better desktop-optimized scheduling into it, but it's a slow process... Likewise, X.Org as the only display option right now has a design based around client-server fuctionality, not lightning-fast desktop use. It can do things that are impossible in other OS'es, but at the same time it introduces difficult-to-solve performance problems. Furthermore, the whole driver stack is undergoing significant changes on the ATI and Intel sides, so there are plenty of bugs and regressions to be solved there.

That being said, as far back as I remember (and that's around 10 years), 2D acceleration was always poor in Linux and never on par with either *win or MacOSX. I never gave it much thought because I care more for using a free OS than snappy text-boxes, but it is a nagging problem and one that gives newcomers a feel of "slowness" that they associate with the whole OS... Perhaps this should be addressed more seriously by the X.Org/Qt/GTK/etc. devs, there are certainly ways of improving it without switching to a whole different display system like Wayland (which AFAIK does not work on nVidia yet). I'd start bugging them on mailing lists and bug trackers, but wait until the whole KMS/GEM/Gallium thing stabilizes, there's little hope of any kind of improvement until then.

[unimatrix]

Well, in that case this is not necessarily a problem with the nVidia drivers, it is something that has different causes. Discussing it in this thread is not really appropriate since it would make people blame one part of the system when others are responsible.

That's why I'd like to get to the core of this issue. To blame the right part.
The thing that bothers me is, many people claim they do not have this problem. Just today I spoke to somebody on IRC who uses similar software. He's using the 180 nvidia driver with default xorg.conf settings. That is what lead me to believe that it is a problem with nvidia. But on the other hand, my older computer (Celeron 2GHz) has an ATI card in it and the situation is even worse. So it's really confusing and I would love to get to the bottom of this.


Everyone has different expectations when it comes to using a computer, for some it's applications that matter, for others the eye candy, some value responsiveness while others place number crunching performance on top.

And I firmly believe that Linux has the potential of doing all of that :)


I never gave it much thought because I care more for using a free OS than snappy text-boxes, but it is a nagging problem and one that gives newcomers a feel of "slowness" that they associate with the whole OS...

Exactly. This has to be fixed. It is most likely one of the greatest obstacles in desktop Linux adoption.
It's why people claim Linux is slower than Windows, when in reality it's not - for example Mathematica on Linux calculates twice as fast than on Windows on my computer.


Perhaps this should be addressed more seriously by the X.Org/Qt/GTK/etc. devs, there are certainly ways of improving it without switching to a whole different display system like Wayland (which AFAIK does not work on nVidia yet). I'd start bugging them on mailing lists and bug trackers, but wait until the whole KMS/GEM/Gallium thing stabilizes, there's little hope of any kind of improvement until then.
Yes, well ATI and Intel cards have a very promising future. But what about nvidia? AFAIK the GPL nature of KMS/GEM/Gallium stuff doesn't allow them to implement it in their closed-source driver.

[nanonyme]

Yes, well ATI and Intel cards have a very promising future. But what about nvidia? AFAIK the GPL nature of KMS/GEM/Gallium stuff doesn't allow them to implement it in their closed-source driver.Well, let's see. They could just start supporting Nouveau with specs and manpower like AMD/ATi and Intel do for the respective openource driver projects.

[energyman]

Well, let's see. They could just start supporting Nouveau with specs and manpower like AMD/ATi and Intel do for the respective openource driver projects.


but that would be reasonable. Oh noes!