Intel Graphics Card's Full Potential

quadomatic · 2009-10-25 22:48:04

Still getting the same performance even with mesa-git...gonna install intel-dri-git and libdrm-git and see what happens...if there still isn't any improvement then maybe I need to use xf86-video-intel-git, or a newer kernel

What kernel and xorg-server are you all using? I'm using xorg-server 1.6.3.901-1 and kernel26 2.6.31.5-1

Last edited by quadomatic (2009-10-25 22:51:48)

quadomatic · 2009-10-26 01:29:46

OMG...after going through the hell of upgrading xorg-server to 1.7, and not installing the necessary hotplugging drivers, thereby losing keyboard and mouse support for 2 hours, I found a significantly increased glxgears fps measure. I know it's not a benchmark, but if it changes on MY system then it means something.

It was at 550 to 600, now it's at 750 to 800.

So, just to say, here's a record of what I did to improve my performance:

1. Upgraded kernel to 2.6.31.5-1
2. Upgraded Mesa to mesa-git (had to install dri2proto-git, glproto-git, and delete some glew files)
3. Upgraded to intel-dri-git, libdrm-git
4. Upgraded to xf86-video-intel-git (had to modify pkgbuild to tell it dependencies were intel-dri-git and libdrm-git, not *-newest)
5. Upgraded xorg-server to 1.7.1-1 (BE SURE to install the necessary xorg upgrades, like xf86-input-evdev and xf86-input-synaptics for hotplugging and what not, or you'll reboot and find your keyboard and mouse don't work)

I got a sizable increase in GLXgears, compiz flows considerably more smoothly, and urban terror performs considerably better (though it could certainly be better...I'm not really getting 30 fps yet, but then again I am running it while compiz is on).

Good luck to anyone trying to get an extra boost!

Last edited by quadomatic (2009-10-26 01:47:36)

1LordAnubis · 2009-10-26 03:57:59

Another thing I've noticed is from this thread: http://bbs.archlinux.org/viewtopic.php?id=81429

I just made a gmabooster.sh file, did a chmod +x, then did ./gmabooster.sh 400 and my gma has gotten better performance in some apps. When you reboot the clock speed is reset to normal, but when I suspend it gets set to something else... I've received no ill effects in performance or temperature, but I would be careful on what hardware you try this on. It should be ok on an intel 945 gm though.

The script:

#!/bin/sh

case "$1" in
    250) val=31 ;;
    400) val=33 ;;
    200) val=34 ;;
    *)
        echo "Illegal argument! Possible values are: 200, 250, 400." >&2
        exit 1
    ;;
esac

setpci -s 02.0 f0.b=00,60
setpci -s 02.0 f0.b=$val,05

goran'agar · 2009-10-26 10:50:08

quadomatic wrote:

1. Upgraded kernel to 2.6.31.5-1
2. Upgraded Mesa to mesa-git (had to install dri2proto-git, glproto-git, and delete some glew files)
3. Upgraded to intel-dri-git, libdrm-git
4. Upgraded to xf86-video-intel-git (had to modify pkgbuild to tell it dependencies were intel-dri-git and libdrm-git, not *-newest)
5. Upgraded xorg-server to 1.7.1-1 (BE SURE to install the necessary xorg upgrades, like xf86-input-evdev and xf86-input-synaptics for hotplugging and what not, or you'll reboot and find your keyboard and mouse don't work)

I have more or less your config with all those -git packages.
Did you just enable the testing repo and issue a "pacman -S xorg-server" ?

[Edit]
Now that I think about it I can't see how this can even work unless you upgraded your xorg-server package from 1.7.X also from testing, otherwise the server wouldn't be able to load the intel driver compiled for a different ABI.

Can you please confirm?

Last edited by goran'agar (2009-10-26 14:40:01)

quadomatic · 2009-10-26 14:39:20

I upgraded xorg-server to 1.7.1-1 from testing

Also, I just unsuspended this morning and I don't have real great performance anymore...maybe I just need a reboot

stefanwilkens · 2009-10-26 17:00:52

I would, once again, like to refer you all to this:
http://wiki.cchtml.com/index.php/Glxgea … _Benchmark

To supply the main conclusion:

So to summarize, glxgears only tests a small part of what you typically see in a 3D game. You could have glxgears FPS performance increase, but your 3D game performance decrease. Likewise, you could have glxgears performance decrease and your 3D game performance increase.

The extremely limited scope of OpenGL functions makes Glxgears completely unreliable and should NOT be used as a benchmark to determine desktop / game performance.

Due to the popular use of glxgears and the everlasting frustration, some of us have started this:
http://wiki.archlinux.org/index.php/Benchmarking
http://bbs.archlinux.org/viewtopic.php?pid=643939

I would very much like for all of you to join in the discussion and making this wiki article complete

much better, isn't it

quadomatic · 2009-10-26 20:05:10

stefanwilkens wrote:

I would, once again, like to refer you all to this:
http://wiki.cchtml.com/index.php/Glxgea … _Benchmark
To supply the main conclusion:
So to summarize, glxgears only tests a small part of what you typically see in a 3D game. You could have glxgears FPS performance increase, but your 3D game performance decrease. Likewise, you could have glxgears performance decrease and your 3D game performance increase.
The extremely limited scope of OpenGL functions makes Glxgears completely unreliable and should NOT be used as a benchmark to determine desktop / game performance.
Due to the popular use of glxgears and the everlasting frustration, some of us have started this:
http://wiki.archlinux.org/index.php/Benchmarking
http://bbs.archlinux.org/viewtopic.php?pid=643939
I would very much like for all of you to join in the discussion and making this wiki article complete
http://unigine.com/download/heaven.jpg
much better, isn't it

Thanks for the info...tho I think unigine benchmarks are a bit too much for intel

Slack · 2009-10-26 22:11:26

omg, with xorg 1.7.1 from testing, I have finally a good performance with compiz (intel drivers, uvdev and synaptics from testing too)

quadomatic · 2009-10-26 22:59:27

Cool, I'm glad it works

Unfortunately I still get xorg crashes and nasty crashes where the system just freezes, probably more violent ones than I had before. I'll log in, and it'll instantly take me back to login, or things like that...

I'm guessing my problem probably lies elsewhere. I posted the backtrace in another thread but I guess I'll put it here too:

Backtrace:
0: /usr/bin/Xorg (xorg_backtrace+0x28) [0x45cf48]
1: /usr/bin/Xorg (0x400000+0x59ea9) [0x459ea9]
2: /lib/libpthread.so.0 (0x7f2a709f7000+0xe790) [0x7f2a70a05790]
3: /lib/libc.so.6 (__select+0x13) [0x7f2a6fb479f3]
4: /usr/bin/Xorg (WaitForSomething+0x1ba) [0x45f82a]
5: /usr/bin/Xorg (0x400000+0x463c2) [0x4463c2]
6: /usr/bin/Xorg (0x400000+0x2202c) [0x42202c]
7: /lib/libc.so.6 (__libc_start_main+0xfd) [0x7f2a6faa19ed]
8: /usr/bin/Xorg (0x400000+0x21bd9) [0x421bd9]
Fatal server error:
Caught signal 3 (Quit). Server aborting

steve___ · 2009-10-27 01:41:44

Did you do anything else to the computer? Like upgrade any hardware recently? I'm starting to wonder if your 'xorg/intel' problem is a red herring. Is it possible to swap out your memory modules with different ones? If that is not possible can you swap out the hdd and install arch refresh? And see if you get the crashes.

quadomatic · 2009-10-27 02:57:23

I haven't upgraded any hardware. This is a relatively new laptop...though I definitely feel like i've been having serious issues.

I've had some hard drive corruption, though that's likely due to the crashing...I have to hard power off and then I sometimes have problems afterwards, but e2fsck generally fixes it. However, I've had my windows vista partition corrupted as well...not sure what the deal is there.

Also, there will be times when I boot up and my bios will complain about there being something wrong with my ram, and it'll tell me to reseat the memory. I have no idea why this happens, but it gives me some options when this happens, and if I go into the bios setup and then leave and try booting again it's all good.

I can't really swap out any memory, and I can't pop in a new hard drive either, since I don't have another one lying around (it's a laptop...don't have any extra laptop memory or laptop drives). I can't even install a new OS off of the disc drive. The disc drive is shot...damn self-loading drive broke within a couple months.

Here's the output from a bonnie benchmark on my hard drive:

Version 1.96 ------Sequential Output------ --Sequential Input- --Random-
Concurrency 1 -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
aneesh-laptop 8G 221 93 42831 18 20268 5 1154 79 54412 7 140.1 6
Latency 84498us 2500ms 4172ms 253ms 290ms 798ms
Version 1.96 ------Sequential Create------ --------Random Create--------
aneesh-laptop -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 11416 36 +++++ +++ 24789 63 22807 70 +++++ +++ 27675 70
Latency 74620us 2381us 14389us 472us 230us 545us
1.96,1.96,aneesh-laptop,1,1256587432,8G,,221,93,42831,18,20268,5,1154,79,54412,7,140.1,6,16,,,,,11416,36,+++++,+++,24789,63,22807,70,+++++,+++,27675,70,84498us,2500ms,4172ms,253ms,290ms,798ms,74620us,2381us,14389us,472us,230us,545us

Not sure if that says anything about the drive that's important, but there it is.

I think I was having crashing/slowness issues in Ubuntu as well though, on this same laptop, prior to installing arch, so I'm not sure it's the hardware's fault...but then again, that would seem to point to a hardware problem. It seems like it'd be more a driver issue of some kind.

I'm using KDE4 now. It seems like points where GNOME would've crashed, KDE simply stalls for a quick sec and goes on. I'm not sure though. I've only been using it for a few hours now. Usually it takes a day or two to know for sure.

I'll see if I can try running memtest86+, maybe I'll find a problem there.

Last edited by quadomatic (2009-10-27 03:02:01)

quadomatic · 2009-10-27 04:27:21

I ran memtest86, turns out one of my memory sticks was completely screwed. I ran it once with both sticks in, and it came up with a ton of errors, so I popped one of em out and tested again and a full pass came up clean.

I then popped out that stick and put in the other one. Dell bios told me there was a read write error on the memory and limited the memory to 186MB of RAM O_o

I ran memtest with that single stick and came up with a ton of errors again, so I popped the other stick back in, and memtest came up with no errors again (only ran for a few seconds though...but I had finished a full pass with that stick already so w/e). So I guess the culprit was that stick of ram.

However, I still got a xorg crash. I tried logging in after booting and then I instantly had the login screen reload. I looked at my log file, and it looks like I got a crash:

Backtrace:
0: /usr/bin/Xorg (xorg_backtrace+0x28) [0x45cf48]
1: /usr/bin/Xorg (0x400000+0x59ea9) [0x459ea9]
2: /lib/libpthread.so.0 (0x7fcc4d0b4000+0xe790) [0x7fcc4d0c2790]
3: /lib/libc.so.6 (__select+0x13) [0x7fcc4c2049f3]
4: /usr/bin/Xorg (WaitForSomething+0x1ba) [0x45f82a]
5: /usr/bin/Xorg (0x400000+0x463c2) [0x4463c2]
6: /usr/bin/Xorg (0x400000+0x2202c) [0x42202c]
7: /lib/libc.so.6 (__libc_start_main+0xfd) [0x7fcc4c15e9ed]
8: /usr/bin/Xorg (0x400000+0x21bd9) [0x421bd9]
Fatal server error:
Caught signal 3 (Quit). Server aborting

I've only been running on the single stick for a little while though. I've yet to see a crash outside of that single crash upon first login (and I had those login crashes since upgrading to Xorg 1.7, but not before...before it would just be sponatneous crashes back to login while just using my comp)

steve___ · 2009-10-27 04:38:37

My guess is the bad memory module has written corupt data and that caused the latest crash. I'd fsck the drive. Hopefully things are back on track now

Last edited by steve___ (2009-10-27 04:39:15)

quadomatic · 2009-10-27 05:39:18

OMG THAT's why I've been having drive corruption then?

GAWD, stupid memory module...

quadomatic · 2009-10-27 06:00:04

Alright (I know we've gone WAY off topic) but I used e2fsck on all my partitions. My root and var partitions seemed to be fine, but my home partition was SCREWED. SOOOOO many problems came up.

Xorg still dies after the first login and reloads the login right after, but I guess it's not a huge deal since I'm not having any other issues for now.

quadomatic · 2009-10-27 06:14:52

Back to the discussion of intel gfx, here are my benchmarks done in phoronix test suite with glmark. Included is a bunch of info about my system when the test was run.

I suggest everyone run benchmarks like this. PLEASE, install phoronix-test-suite, and then do "phoronix-test-suite install glmark", then run glmark tests by doing "phoronix-test-suite run glmark"

System info:

Installed: intel-dri-git, libdrm-git, glproto-git, dri2proto-git, xf86-video-intel-git, xorg-server-1.7.1-1

Info according to glmark:

Processor: Intel Core 2 Duo CPU T6400 @ 1.99GHz (Total Cores: 2)
Motherboard: Dell XPS M1330
Chipset: Intel Mobile PM965/GM965/GL960 + ICH8M
System Memory: 2000MB
Disk: 250GB WDC WD2500BEVT-7
Graphics: Intel Mobile GM965/GL960 IGP 256MB
OS: Arch, Kernel: 2.6.31-ARCH (x86_64)
Desktop: KDE 4.3.2
Display Server: X.Org Server 1.7.1
Display Driver: intel 2.9.0
OpenGL: 2.1
Mesa 7.7-devel
Compiler: GCC 4.4.1
File-System: ext3
Screen Resolution: 1280x800

Vertex Array test, avg 43 fps
http://global.phoronix-test-suite.com/i … 2588-18031

Vertex Buffer test, avg 58 fps
http://global.phoronix-test-suite.com/i … 30680-2191

Texture Filter, Linear, avg 49 fps
http://global.phoronix-test-suite.com/i … 1920-26803

Texture Filter, mipmapped, avg 60 fps
http://global.phoronix-test-suite.com/i … 16153-5522

GLSL per vertex, avg 45 fps
http://global.phoronix-test-suite.com/i … 7614-32008

GLSL per pixel, avg 44 fps
http://global.phoronix-test-suite.com/i … 27948-8412

I tried UrbanTerror, and it runs at like 21 fps...but I am in kde. KDE doesn't seem to have as good performance as gnome...I think gnome got 25 to 30. KDE seems a bit more stuttery too. Not sure though.

Last edited by quadomatic (2009-10-27 06:44:10)

stojic · 2009-10-27 07:38:48

These are the results for Intel GMA 4500MHD.

Packages I have installed (all are current ones from extra)

intel-dri 7.5.1-2
libdrm 2.4.13-1
glproto 1.4.10-1
dri2proto 2.1-1
xf86-video-intel 2.8.1-1
xorg-server 1.6.3.901-1

My wm is openbox.

GLMark info and test results (all tests were run at 1280x800):

Processor: Intel Core 2 Duo CPU P8400 @ 2.26GHz (Total Cores: 2)
Motherboard: LENOVO 2714A22
Chipset: Intel Mobile 4 Series Chipset Hub + ICH9M
System Memory: 1968MB
Disk: 250GB HITACHI HTS54322
Graphics: Intel Mobile 4 Series Chipset IGP (rev 07)

OS: Arch
Kernel: 2.6.31-ARCH (i686)
Display Server: X.Org Server 1.6.3.901 (1.6.4 RC 1)
Display Driver: intel 2.8.1
OpenGL: 2.1 Mesa 7.5.1
Compiler: GCC 4.4.1
File-System: ext3
Screen Resolution: 1280x800

Vertex array            FPS: 179
Vertex buffer object        FPS: 183
Texture Filtering - Linear    FPS: 138
Texture Filtering - Mipmapped    FPS: 189
GLSL Per Vertex Lighting    FPS: 128
GLSL Per Pixel Lighting        FPS: 128

quadomatic · 2009-10-27 14:17:39

OMG those benchmarks are good.....hm

I feel like I should try testing in openbox real quick...I do have it set up. I'll try it later and post the results.

quadomatic · 2009-10-28 20:02:23

I ran a couple benchmarks again in openbox. I ended up getting 62 fps avg in the vertex array and buffer benchmarks, so not a HUGE improvement, but sizeable.

1LordAnubis · 2009-10-28 20:23:47

I had similar memory problems on my alienware laptop, apparently I got memory corruption if i moved the ram anywhere but the bottom two slots. This caused a lot of seemingly random crashes....

stojic · 2009-11-01 13:17:19

I noticed some improvements with recent upgrades, so I decided to retest.

Packages:

intel-dri 7.6-2
libdrm 2.4.15-1
glproto 1.4.10-1
dri2proto 2.1-1
xf86-video-intel 2.9.1-1
xorg-server 1.7.1-1

GLMark test results with comparison to previous results (all tests were run at 1280x800, info same as before):

Vertex array                  FPS: 206    +15%
Vertex buffer object          FPS: 210    +15%
Texture Filtering - Linear    FPS: 179    +30%
Texture Filtering - Mipmapped FPS: 213    +13%
GLSL per vertex lighting      FPS: 164    +28%
GLSL per pixel lighting       FPS: 164    +28%

So there is an upgrade with some improvements finally, ten or fifteen upgrades like this and it would be running like a mid-range NVIDIA card .

@quadomatic:I think you should be getting at least half of these numbers on GMA X3100, maybe your cpufreq governor is set to ondemand or something so it slows things down? I run these tests with performance governor, results (and performance in games too) are a bit lower if I use other governors (CPU probably scaling all the time and slowing things down).

edit: This is section Device from my xorg.conf, that Option causes some games to always run with vsync on so I disabled it:

Section "Device"
        Identifier  "x4500"
        Driver      "intel"
        Option      "SwapbuffersWait"        "False"
EndSection

Last edited by stojic (2009-11-01 13:24:42)

quadomatic · 2009-11-09 00:59:01

I apparently do have cpufrequtils installed, and it's configuration file is set to ondemand...but my rc.conf isn't set to load that module. Also, no process with the string cpu or freq in the name is running on my system. So, I don't think there's actually any scaling going on, so I don't think it actually affects my performance in any way.

Still though, there's definitely a huge disparity between our benchmarks...

Last edited by quadomatic (2009-11-09 01:01:33)

fluffylime · 2009-11-09 15:47:39

zuargo wrote:

:: intel-dri conflicts with intel-dri-git. Remove intel-dri-git? [Y/n] n
error: unresolvable package conflicts detected
error: failed to prepare transaction (conflicting dependencies)
:: intel-dri: conflicts with intel-dri-git
==> ERROR: Pacman failed to install missing dependencies.
Error: Makepkg was unable to build xf86-video-intel-git package.
If I want to install the xf86-video-intel-git package I must remove the intel-dri-git package, hence, I can't install both packages (intel-dri-git and xf86-video-intel-git) on the system.
Do you have installed both packages on your system?
If your answer is yes:
How did you do that?
Thanks in advance

The thing is that xf86-video-intel-git wants mesa>=7.5, but mesa-git from AUR provides mesa without version number. To solve this, I've changed 'mesa>=7.5' to 'mesa' in the depends of xf86-video-intel-git.

P.S. In my case yaourt complained about dri2proto, but I think root of the problem is the same.

Last edited by fluffylime (2009-11-09 15:51:55)

linuxgod · 2010-01-25 19:55:24

How could i run phoronix testing suite??
I have GMA X4500 and willing to test it!!

but i get error like this

$ sudo phoronix-test-suite install glmark
PHP Warning:  require(): open_basedir restriction in effect. File(/usr/share/phoronix-test-suite/pts-core/library/pts-functions.php) is not within the allowed path(s): (/srv/http/:/home/:/tmp/:/usr/share/pear/) in /usr/share/phoronix-test-suite/pts-core/phoronix-test-suite.php on line 33
PHP Warning:  require(/usr/share/phoronix-test-suite/pts-core/library/pts-functions.php): failed to open stream: Operation not permitted in /usr/share/phoronix-test-suite/pts-core/phoronix-test-suite.php on line 33
PHP Fatal error:  require(): Failed opening required '/usr/share/phoronix-test-suite/pts-core/library/pts-functions.php' (include_path='.:/usr/share/pear') in /usr/share/phoronix-test-suite/pts-core/phoronix-test-suite.php on line 33

Arch Linux

#51 2009-10-25 22:48:04

Re: Intel Graphics Card's Full Potential

#52 2009-10-26 01:29:46

Re: Intel Graphics Card's Full Potential

#53 2009-10-26 03:57:59

Re: Intel Graphics Card's Full Potential

#54 2009-10-26 10:50:08

Re: Intel Graphics Card's Full Potential

#55 2009-10-26 14:39:20

Re: Intel Graphics Card's Full Potential

#56 2009-10-26 17:00:52

Re: Intel Graphics Card's Full Potential

#57 2009-10-26 20:05:10

Re: Intel Graphics Card's Full Potential

#58 2009-10-26 22:11:26

Re: Intel Graphics Card's Full Potential

#59 2009-10-26 22:59:27

Re: Intel Graphics Card's Full Potential

#60 2009-10-27 01:41:44

Re: Intel Graphics Card's Full Potential

#61 2009-10-27 02:57:23

Re: Intel Graphics Card's Full Potential

#62 2009-10-27 04:27:21

Re: Intel Graphics Card's Full Potential

#63 2009-10-27 04:38:37

Re: Intel Graphics Card's Full Potential

#64 2009-10-27 05:39:18

Re: Intel Graphics Card's Full Potential

#65 2009-10-27 06:00:04

Re: Intel Graphics Card's Full Potential

#66 2009-10-27 06:14:52

Re: Intel Graphics Card's Full Potential

#67 2009-10-27 07:38:48

Re: Intel Graphics Card's Full Potential

#68 2009-10-27 14:17:39

Re: Intel Graphics Card's Full Potential

#69 2009-10-28 20:02:23

Re: Intel Graphics Card's Full Potential

#70 2009-10-28 20:23:47

Re: Intel Graphics Card's Full Potential

#71 2009-11-01 13:17:19

Re: Intel Graphics Card's Full Potential

#72 2009-11-09 00:59:01

Re: Intel Graphics Card's Full Potential

#73 2009-11-09 15:47:39

Re: Intel Graphics Card's Full Potential

#74 2010-01-25 19:55:24

Re: Intel Graphics Card's Full Potential

Board footer