Bug 28068

Summary: RV100 performance regressions (since 6.8.3 from 2005) and 1.7.6/6.13 from 2010
Product: xorg Reporter: Darryl Miles <darryl>
Component: Driver/RadeonAssignee: xf86-video-ati maintainers <xorg-driver-ati>
Status: RESOLVED INVALID QA Contact: Xorg Project Team <xorg-team>
Severity: normal    
Priority: medium    
Version: 7.5 (2009.10)   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
Most interesting regressions.
none
1.7.6.RV100.Xorg.0.log
none
1.7.6.x11perf.log.20100506.RV100.FC12
none
6.8.2.RV100.Xorg.0.log
none
6.8.2.x11perf.log.20100506.RV100.FC3
none
6.8.2.xorg.conf
none
lspci.txt none

Description Darryl Miles 2010-05-11 11:23:45 UTC
I upgraded from FedoraCore4 (which has provided me good service for many years) to Fedora12.  In doing so there were some obvious performance and rendering differences that I could consider to be regressions.

In this bug report is to concentrate on describing the performance differences, after consultation with the good folk on irc:freenode/#radeon I ran some x11perf tests to highlight the issues and provide real numbers of my experience.

Overall many tests have gotten a few %  (between 5 and 15%) over the past 5 years, however some of the worst regressions have gotten 20 times slower.  Particulating relating to.


To explain the testing setup, it was exactly the same hardware, the X11 server was started manually and free standing while the system was in a low runlevel (in some cases 1/single-user).  Same LCD monitor, same resolution, same 64bit variety of OS/Xorg/drivers.  The tests were then run and the output logged.

The 2005 versions are Xorg 6.8.3, Kernel 2.6.9-22.ELsmp x86_64, using UMS.

The 2010 version is Xorg 1.7.6 with xorg-x11-drv-ati-6.13.0-0.21.20100219gite68d3a389.fc12.x86_64, Kernel 2.6.32.11-99.fc12.x86_64 using KMS.


The tests which have regressed and are the most critical (and the focus of this bug report) are:

WORSE -95% CRITICAL REGRESSION
6.8.2:     900 reps @   5.7020 msec (   175.0/sec): Scroll 500x500 pixels
1.7.6:      50 reps @ 108.3004 msec (     9.2/sec): Scroll 500x500 pixels

WORSE -42% CRITICAL REGRESSION
6.8.2:    1200 reps @   5.3795 msec (   186.0/sec): Copy 500x500 from pixmap to window 
1.7.6:     800 reps @   9.2924 msec (   108.0/sec): Copy 500x500 from pixmap to window

WORSE -95% CRITICAL REGRESSION
6.8.2:    1200 reps @   5.4315 msec (   184.0/sec): Copy 500x500 from window to pixmap 
1.7.6:      80 reps @ 109.8940 msec (     9.1/sec): Copy 500x500 from window to pixmap
Comment 1 Darryl Miles 2010-05-11 11:50:29 UTC
Created attachment 35571 [details]
Most interesting regressions.
Comment 2 Darryl Miles 2010-05-11 11:51:09 UTC
Created attachment 35572 [details]
1.7.6.RV100.Xorg.0.log
Comment 3 Darryl Miles 2010-05-11 11:52:35 UTC
Created attachment 35573 [details]
1.7.6.x11perf.log.20100506.RV100.FC12

x11perf of Selected tests

followed by "x11perf -all"
Comment 4 Darryl Miles 2010-05-11 11:53:58 UTC
Created attachment 35574 [details]
6.8.2.RV100.Xorg.0.log
Comment 5 Darryl Miles 2010-05-11 11:54:58 UTC
Created attachment 35575 [details]
6.8.2.x11perf.log.20100506.RV100.FC3

x11perf of Selected tests

followed by "x11perf -all"
Comment 6 Darryl Miles 2010-05-11 11:55:33 UTC
Created attachment 35576 [details]
6.8.2.xorg.conf
Comment 7 Darryl Miles 2010-05-11 11:58:47 UTC
Created attachment 35577 [details]
lspci.txt

Dell PE1800 (Power Edge, enterprise server, integrated RV100 graphics chipset).

Dual 3.2GHz Intel ia32/EM64T CPUs
Comment 8 Michel Dänzer 2010-05-12 04:53:27 UTC
Does

    Option "EXAPixmaps"

help on Fedora 12?
Comment 9 Adam Jackson 2018-06-12 19:06:09 UTC
Mass closure: This bug has been untouched for more than six years, and is not
obviously still valid. Please reopen this bug or file a new report if you continue to experience issues with current releases.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.