Bug 521080 - intel driver very slow on 82865G
Summary: intel driver very slow on 82865G
Keywords:
Status: CLOSED INSUFFICIENT_DATA
Alias: None
Product: Fedora
Classification: Fedora
Component: xorg-x11-drv-intel
Version: rawhide
Hardware: All
OS: Linux
low
medium
Target Milestone: ---
Assignee: Adam Jackson
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks: 505485
TreeView+ depends on / blocked
 
Reported: 2009-09-03 14:29 UTC by David
Modified: 2018-04-11 19:17 UTC (History)
11 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of: 504048
Environment:
Last Closed: 2009-11-08 02:08:17 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)

Description David 2009-09-03 14:29:07 UTC
+++ This bug was initially created as a clone of Bug #504048 +++

Description of problem:
the intel graphics performance is very poor on f11 on this 82865G system:
http://www.smolts.org/client/show/pub_45e71f52-1171-43c0-9c36-aacc87834e13


Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1. Run gtkperf
2.
3.
  
Actual results:
With no options to gtkperf I got these results:

intel driver KMS 1680x1050
GtkEntry - time:  0.47
GtkComboBox - time:  6.78
GtkComboBoxEntry - time:  3.31
GtkSpinButton - time:  2.38
GtkProgressBar - time:  1.16
GtkToggleButton - time:  1.50
GtkCheckButton - time:  0.45
GtkRadioButton - time:  1.21
GtkTextView - Add text - time:  3.39
GtkTextView - Scroll - time:  3.55
GtkDrawingArea - Lines - time:  1.76
GtkDrawingArea - Circles - time: 44.60
GtkDrawingArea - Text - time: 15.80
GtkDrawingArea - Pixbufs - time:  1.38
 ---
Total time: 87.74

intel driver KMS 1280x1024:
GtkEntry - time:  0.49
GtkComboBox - time:  6.90
GtkComboBoxEntry - time:  4.83
GtkSpinButton - time:  2.43
GtkProgressBar - time:  1.16
GtkToggleButton - time:  2.04
GtkCheckButton - time:  0.46
GtkRadioButton - time:  1.17
GtkTextView - Add text - time:  3.42
GtkTextView - Scroll - time:  3.85
GtkDrawingArea - Lines - time:  1.65
GtkDrawingArea - Circles - time: 45.98
GtkDrawingArea - Text - time: 16.29
GtkDrawingArea - Pixbufs - time:  1.05
 ---
Total time: 91.73



intel driver nomodeset 1680x1050 :
GtkEntry - time:  0.36
GtkComboBox - time:  3.87
GtkComboBoxEntry - time:  2.58
GtkSpinButton - time:  1.33
GtkProgressBar - time:  0.66
GtkToggleButton - time:  1.02
GtkCheckButton - time:  0.29
GtkRadioButton - time:  0.88
GtkTextView - Add text - time:  2.55
GtkTextView - Scroll - time:  2.25
GtkDrawingArea - Lines - time:  1.51
GtkDrawingArea - Circles - time: 14.21
GtkDrawingArea - Text - time: 11.44
GtkDrawingArea - Pixbufs - time:  1.02
 ---
Total time
43.97


vesa driver KMS (1280x1024??):
GtkEntry - time:  0.14
GtkComboBox - time:  1.77
GtkComboBoxEntry - time:  1.29
GtkSpinButton - time:  0.30
GtkProgressBar - time:  0.20
GtkToggleButton - time:  0.33
GtkCheckButton - time:  0.12
GtkRadioButton - time:  0.28
GtkTextView - Add text - time:  1.01
GtkTextView - Scroll - time:  0.53
GtkDrawingArea - Lines - time:  0.61
GtkDrawingArea - Circles - time:  1.43
GtkDrawingArea - Text - time:  1.31
GtkDrawingArea - Pixbufs - time:  0.10
 ---
Total time:  9.42

I didn't try the vesa driver with nomodeset.

Expected results:
I would hope that the intel driver would outperfom
the vesa driver, at least at the same resolution.
Instead, it was about 10 times slower.

Additional info:

I'm not sure if it's related or not, but scrolling in firefox
is painfully slow as well.

A proprietary application the uses gtk and cairo runs about 20x
slower than I'm used to.  I'm going from distant memory, but
back before the intel driver was too broken to use on this system
for several fedora releases, I think the same application ran
with the i810 driver much faster than it does now with the intel
driver.

--- Additional comment from selinux on 2009-06-04 11:38:37 EDT ---

Adding to this:

I have a Thinkpad X200 with "Intel Corporation Mobile 4 Series Chipset Integrated Graphics Controller (rev 07)"/"Intel Mobile Intel® GM45 Express Chipset"

List below gtkperf results running both KMS and "nomodeset".  The "nomodeset" run is about 15 seconds faster.... (114 vs. 129).

Here are results running gtkperf with KMS boot/1280x1024:

[tbl@tlondon ~]$ gtkperf -c 500 -a
GtkPerf 0.40 - Starting testing: Thu Jun  4 08:24:02 2009

GtkEntry - time:  0.88
GtkComboBox - time:  6.24
GtkComboBoxEntry - time:  4.40
GtkSpinButton - time:  3.57
GtkProgressBar - time:  2.38
GtkToggleButton - time:  2.10
GtkCheckButton - time:  0.65
GtkRadioButton - time:  2.08
GtkTextView - Add text - time: 10.62
GtkTextView - Scroll - time:  6.83
GtkDrawingArea - Lines - time: 25.69
GtkDrawingArea - Circles - time: 45.33
GtkDrawingArea - Text - time: 16.05
GtkDrawingArea - Pixbufs - time:  2.53
 --- 
Total time: 129.37

Quitting..
[tbl@tlondon ~]$

For comparison, here is gtkperf output running with "nomodeset nopat" boot/1280x1024:

[tbl@tlondon ~]$ gtkperf -c 500 -a
GtkPerf 0.40 - Starting testing: Thu Jun  4 06:28:27 2009

GtkEntry - time:  0.87
GtkComboBox - time:  7.17
GtkComboBoxEntry - time:  5.22
GtkSpinButton - time:  3.92
GtkProgressBar - time:  1.73
GtkToggleButton - time:  1.92
GtkCheckButton - time:  0.76
GtkRadioButton - time:  2.15
GtkTextView - Add text - time:  9.18
GtkTextView - Scroll - time:  4.60
GtkDrawingArea - Lines - time: 23.35
GtkDrawingArea - Circles - time: 33.16
GtkDrawingArea - Text - time: 16.75
GtkDrawingArea - Pixbufs - time:  3.46
 --- 
Total time: 114.26

Quitting..

--- Additional comment from mcepl on 2009-06-05 08:35:57 EDT ---

Tom, couldn't we get /var/log/Xorg.0.log, /var/log/dmesg, and (if you have any) /etc/X11/xorg.conf? I am not sure how much helpful it actually is, but just to be sure we are not overlooking anything.

Thank you

--- Additional comment from selinux on 2009-06-05 09:17:56 EDT ---

Created an attachment (id=346647)
Xorg.0.log booting with "nomodeset nopat"

Sure, I'll attach as requested, starting with Xorg.0.log

Here are requested files for a boot with "nomodeset nopat"/1280x1024 obtained when I ran:

[tbl@tlondon ~]$ gtkperf -c 500 -a
GtkPerf 0.40 - Starting testing: Fri Jun  5 06:14:20 2009

GtkEntry - time:  0.84
GtkComboBox - time:  6.75
GtkComboBoxEntry - time:  5.20
GtkSpinButton - time:  3.89
GtkProgressBar - time:  1.73
GtkToggleButton - time:  1.75
GtkCheckButton - time:  0.80
GtkRadioButton - time:  1.97
GtkTextView - Add text - time:  9.07
GtkTextView - Scroll - time:  4.73
GtkDrawingArea - Lines - time: 27.24
GtkDrawingArea - Circles - time: 36.75
GtkDrawingArea - Text - time: 16.61
GtkDrawingArea - Pixbufs - time:  3.54
 --- 
Total time: 120.88

Quitting..
[tbl@tlondon ~]$

--- Additional comment from selinux on 2009-06-05 09:18:46 EDT ---

Created an attachment (id=346648)
output of "dmesg" booting with "nomodeset nopat"

Dmesg output

--- Additional comment from selinux on 2009-06-05 09:19:36 EDT ---

Sorry, forgot to mention: "no xorg.conf".

--- Additional comment from selinux on 2009-06-05 09:20:22 EDT ---

I'll boot later (with newer monitor) to get Xorg.0.log and "dmesg" booting with KMS.

--- Additional comment from selinux on 2009-06-05 09:39:09 EDT ---

BTW, "gtkperf" does not seem to run "smoothly" on my box: it appears to hang sometimes during:

GtkComboBox - time: 11.59
GtkComboBoxEntry - time:  8.86

Notice that the previous run reported in #3 above was:

GtkComboBox - time:  6.75
GtkComboBoxEntry - time:  5.20

I've noticed quite a bit of variance here, and curiously, it appears I can affect this by "just moving the mouse".

Sounds like this comment from fedora-test:

On an ECS 945 GCT-M/1333 the X server randomly stalls,
apparently frozen.  However, moving the mouse restores
normal operation, at least for a while.  This happens with
getperf.  It also seems to happen during graphical install.

Anything obvious for me to do to provide more here?

--- Additional comment from idht4n on 2009-06-05 10:04:43 EDT ---

Created an attachment (id=346656)
my Xorg.0.log with intel driver, nomodeset

--- Additional comment from idht4n on 2009-06-05 10:07:30 EDT ---

Created an attachment (id=346658)
my dmesg output, nomodeset

--- Additional comment from selinux on 2009-06-05 12:03:56 EDT ---

Created an attachment (id=346676)
Xorg.0.log booting with KMS

Here is Xorg.0.log from boot with KMS/1280x1024:

[tbl@tlondon ~]$ time gtkperf -c 500 -a
GtkPerf 0.40 - Starting testing: Fri Jun  5 08:56:36 2009

GtkEntry - time:  0.72
GtkComboBox - time:  6.59
GtkComboBoxEntry - time:  4.58
GtkSpinButton - time:  3.64
GtkProgressBar - time:  2.38
GtkToggleButton - time:  2.19
GtkCheckButton - time:  0.62
GtkRadioButton - time:  2.01
GtkTextView - Add text - time: 11.47
GtkTextView - Scroll - time:  6.43
GtkDrawingArea - Lines - time: 25.90
GtkDrawingArea - Circles - time: 42.13
GtkDrawingArea - Text - time: 14.65
GtkDrawingArea - Pixbufs - time:  2.44
 --- 
Total time: 125.76

Quitting..

real	2m6.096s
user	0m15.734s
sys	0m0.591s
[tbl@tlondon ~]$

--- Additional comment from selinux on 2009-06-05 12:05:13 EDT ---

Created an attachment (id=346677)
dmesg output from "gtkperf" run booting with KMS

--- Additional comment from janne on 2009-06-11 06:40:57 EDT ---

Created an attachment (id=347371)
X.org.log from 82865G F10 Compaq Evo D300

--- Additional comment from janne on 2009-06-11 06:44:02 EDT ---

I'm also experiencing very poor performance on my Compaq Evo D300 with 82865G on a F10 system. (X.org.log attached above, id=347371). Everything's painfully slow compared to my previous Mandriva installation, scrolling minimizing, moving etc.

Here's my gtkperf:

intel driver 1680x1050
GtkEntry - time:  0.66
GtkComboBox - time:  5.56
GtkComboBoxEntry - time:  4.03
GtkSpinButton - time:  1.66
GtkProgressBar - time:  1.02
GtkToggleButton - time:  1.45
GtkCheckButton - time:  0.57
GtkRadioButton - time:  1.30
GtkTextView - Add text - time:  5.99
GtkTextView - Scroll - time:  3.81
GtkDrawingArea - Lines - time:  2.47
GtkDrawingArea - Circles - time:  2.74
GtkDrawingArea - Text - time: 20.31
GtkDrawingArea - Pixbufs - time:  1.14
 --- 
Total time: 52.73

--- Additional comment from jason on 2009-06-14 17:05:59 EDT ---

On my 965GM laptop, the vesa driver is also faster, though not as dramatically.  Total times are nomodeset/vesa ~5.3, KMS/UXA ~11, nomodeset/EXA ~14.

Comment 1 Matěj Cepl 2009-11-05 17:14:46 UTC
Since this bugzilla report was filed, there have been several major updates in various components of the Xorg system, which may have resolved this issue. Users who have experienced this problem are encouraged to upgrade their system to the latest version of their packages (at least F12Beta, but even better if the very latest versions).

Please, if you experience this problem on the up-to-date system, let us now in the comment for this bug, or whether the upgraded system works for you.

If you won't be able to reply in one month, I will have to close this bug as INSUFFICIENT_DATA. Thank you.

[This is a bulk message for all open Fedora Rawhide Xorg-related bugs. I'm adding myself to the CC list for each bug, so I'll see any comments you make after this and do my best to make sure every issue gets proper attention.]

Comment 2 David 2009-11-05 21:27:13 UTC
I have been running rawhide and the performance problem has not been
resolved as of last week.  At that time, I finally had enough of
the poor performance of the intel driver and talked my boss into
updating my desktop.  I still have the old desktop kicking around, but
it's a hassle for me to test anything on it now.  Let me know if something
has been changed in the last week that might have improved the performance
and I'll test for you.  Otherwise, the intel driver in rawhide is just
as slow as ever.  About 10x slower running gtkperf than my current
desktop that has comparable specs but ATI graphics.  As far as I'm
concerned, you can close this bug if you want.  I'll never use another
Intel graphics chip with Linux.

Comment 3 Matěj Cepl 2009-11-08 02:08:17 UTC
Thank you for letting us know.

Comment 4 Jason Merrill 2009-11-08 06:27:26 UTC
Closing the bug seems premature to me.  David says that his machine is still slow, and gtkperf shows no improvement on my system with the latest X updates.  Just because one person doesn't care about the bug anymore doesn't mean that the rest of us don't care either.

Comment 5 Matěj Cepl 2009-11-08 10:09:57 UTC
(In reply to comment #4)
> Closing the bug seems premature to me.  David says that his machine is still
> slow, and gtkperf shows no improvement on my system with the latest X updates. 
> Just because one person doesn't care about the bug anymore doesn't mean that
> the rest of us don't care either.  

Rest of you filed their own bugs (e.g., you have bug 505485), where I can expect some cooperation from you as reporters.

Comment 6 David 2009-11-09 15:24:02 UTC
Are you saying I'm uncooperative?  ;)

If it would help, I could probably drop the desktop off here for a month:

http://www.redhat.com/about/contact/ww/americas/mountainview.html

Just let me know.


Note You need to log in before you can comment on or make changes to this bug.