3D graphics benchmarks for FOSS drivers?

76 views
Skip to first unread message

R. Steven Rainwater

unread,
Apr 21, 2006, 12:48:59 PM4/21/06
to
I'm trying to select a new video card for my Linux box. I need 3D
hardware acceleration support in the driver. I am only interested in
cards with Free / Open Source Software drivers so cards like Nvidia are
current not in the running. I've had several people recommend the ATI
Radeon 9250 as the fastest, most stable 3D option. I'm curious if that's
the consensus here as well?

What I haven't been able to find is a comprehensive site that offers a
list of graphics hardware and benchmarks for free software 3D drivers.
Most sites only have benchmarks for proprietary stuff. Anyone know of a
website with anything like that?

John-Paul Stewart

unread,
Apr 21, 2006, 1:15:45 PM4/21/06
to
R. Steven Rainwater wrote:
>
> What I haven't been able to find is a comprehensive site that offers a
> list of graphics hardware and benchmarks for free software 3D drivers.
> Most sites only have benchmarks for proprietary stuff. Anyone know of a
> website with anything like that?

Go to groups.google.com and use their advanced search feature to find
message-ID <pan.2005.12.27....@deadspam.com>. You'll find
some benchmark info from glxgears on a Radeon 9200 with open source
drivers and some figures from my own FireGL card with both the
open-source and proprietary drivers. (Some cards newer than the 9250
are somewhat supported by the newest open source drivers.)

Googling for 'glxgears' results will inevitably turn up other figures.
It's not the best of benchmarks, but it's better than nothing.

Henrik Carlqvist

unread,
Apr 21, 2006, 2:33:07 PM4/21/06
to
John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
>R. Steven Rainwater wrote:
>> What I haven't been able to find is a comprehensive site that offers a
>> list of graphics hardware and benchmarks for free software 3D drivers.
>> Most sites only have benchmarks for proprietary stuff. Anyone know of a
>> website with anything like that?

Sorry, I have also searched without finding any good sites.



> Googling for 'glxgears' results will inevitably turn up other figures.
> It's not the best of benchmarks, but it's better than nothing.

I did a quick shell hack:
cat /proc/pci | grep VGA | colrm 1 4 ; cat /proc/cpuinfo | \
egrep "model name|MHz" ; xdpyinfo | grep version: ; glxinfo | \
grep -A2 "direct rendering" ; glxgears & sleep 25 ; killall glxgears

The above hack gave the following result on my Radeon 9200:

VGA compatible controller: PCI device 1002:5961 (ATI Technologies Inc)
(rev 1).
model name : Intel(R) Pentium(R) 4 CPU 2.60GHz
cpu MHz : 2598.767
model name : Intel(R) Pentium(R) 4 CPU 2.60GHz
cpu MHz : 2598.767
X.Org version: 6.8.0
direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2
13187 frames in 5.0 seconds = 2637.400 FPS
14505 frames in 5.0 seconds = 2901.000 FPS
14588 frames in 5.0 seconds = 2917.600 FPS
14594 frames in 5.0 seconds = 2918.800 FPS
13883 frames in 5.0 seconds = 2776.600 FPS

I have tuned my xorg.conf with the following options in the Device
section:

Option "AGPMode" "4"
Option "EnablePageFlip" "on"

If some more people try the same script and post the result we might be
able to get a picture of the performance difference between different
cards and drivers.

regards Henrik
--
The address in the header is only to prevent spam. My real address is:
hc8(at)uthyres.com Examples of addresses which go to spammers:
ro...@variousus.net root@localhost

Chris Cox

unread,
Apr 21, 2006, 6:20:39 PM4/21/06
to
R. Steven Rainwater wrote:
> I'm trying to select a new video card for my Linux box. I need 3D
> hardware acceleration support in the driver. I am only interested in
> cards with Free / Open Source Software drivers so cards like Nvidia are
> current not in the running. I've had several people recommend the ATI
> Radeon 9250 as the fastest, most stable 3D option. I'm curious if that's
> the consensus here as well?

Probably. It's VERY slow though. Depending.. an X300 can be tweaked
to work with the free driver as well (speaking from experience).


>
> What I haven't been able to find is a comprehensive site that offers a
> list of graphics hardware and benchmarks for free software 3D drivers.
> Most sites only have benchmarks for proprietary stuff. Anyone know of a
> website with anything like that?

Not sure... the performance is really bad.

Johannes Truschnigg

unread,
Apr 22, 2006, 7:54:43 AM4/22/06
to
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Fri, 21 Apr 2006 11:48:59 -0500
"R. Steven Rainwater" <st...@ncc.com> wrote:

> What I haven't been able to find is a comprehensive site that offers
> a list of graphics hardware and benchmarks for free software 3D
> drivers. Most sites only have benchmarks for proprietary stuff.
> Anyone know of a website with anything like that?

Now. I can provide you with a few numbers, however: The fastst ATi-card
based on chips driven with hardware-accelerated GLX by the
"radeon"-driver are Radeon 8500 and Radeon 9100 series. 9250 is much
slower than those, due to reduced clocks and count of pixel pipelines.
I yield about 180fps with my radeon 8500 ond a Athlon 64 X2 3800+ with
2GB of RAM in Quake 3 with an untweaked "fastest"-configuration
(resolution 640x480 instead of the normal 512x384 that the
"fastets"-preset provides). (Gentoo, aggressively compiled/tweaked for
my arch, 32bit)

Btw, I'd be happy to see some figures of Quake 3 running on Intel
Graphics Accelerator 950 based hardware. So if anyone can provide
those, please post.

- --
with best regards:
- - Johannes Truschnigg ( johannes....@gmx.at )

www: http://gnulords.org/~colo
phone: +43 650 2 133337
jabber: johannes....@gmail.com

Please do not bother me with HTML-eMail or attachments. Thank you.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)

iD8DBQFEShmDnnUApj8OcoIRAiyTAJsHE9jzHaxeHkx89IRZ5klvJLLcswCfVG+w
qa2w6d+k1QRvvOIGxGB0s80=
=Doj1
-----END PGP SIGNATURE-----

Michael Mauch

unread,
Apr 22, 2006, 11:55:00 AM4/22/06
to
Henrik Carlqvist wrote:

> The above hack gave the following result on my Radeon 9200:

> VGA compatible controller: PCI device 1002:5961 (ATI Technologies Inc)
> (rev 1).

> 13883 frames in 5.0 seconds = 2776.600 FPS

Interesting. I have a ATI Radeon 9250. I changed your script to use
"lspci" instead of "cat /proc/pci", because I don't have /proc/pci.

# lspci | grep VGA | colrm 1 4 ; cat /proc/cpuinfo | \


egrep "model name|MHz" ; xdpyinfo | grep version: ; glxinfo | \
grep -A2 "direct rendering" ; glxgears & sleep 25 ; killall glxgears

0.0 VGA compatible controller: ATI Technologies Inc RV280 [Radeon 9200 PRO] (rev 01)
model name : AMD Athlon(tm) 64 Processor 3200+
cpu MHz : 2003.030
X.Org version: 7.0.0


direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

[1] 8562
4783 frames in 5.0 seconds = 956.468 FPS
4769 frames in 5.0 seconds = 953.753 FPS
4772 frames in 5.0 seconds = 954.334 FPS
4658 frames in 5.0 seconds = 931.433 FPS


These numbers where about the same with Xorg 6.8.2. So why are my
numbers so much lower than yours? You are not running that
closed-source fglrx driver, are you?

How big is your glxgears window (here it's 300x300 by default, shown by
xwininfo)?

What's your screen size and color depth?

# xdpyinfo | egrep "dimensions|depth of"
dimensions: 1280x1024 pixels (339x271 millimeters)
depth of root window: 24 planes

> I have tuned my xorg.conf with the following options in the Device
> section:

> Option "AGPMode" "4"

I'm using "AGPMode" "8" (since Xorg 7).

> Option "EnablePageFlip" "on"

I don't have this, will try in a minute.


Regards...
Michael

Michael Mauch

unread,
Apr 22, 2006, 12:44:15 PM4/22/06
to
I wrote:
> Henrik Carlqvist wrote:

> > The above hack gave the following result on my Radeon 9200:

> > VGA compatible controller: PCI device 1002:5961 (ATI Technologies Inc)
> > (rev 1).

> > 13883 frames in 5.0 seconds = 2776.600 FPS


> > Option "EnablePageFlip" "on"

> I don't have this, will try in a minute.

With "EnablePageFlip" "on" I get about 1300 fps - better, but still only
half as fast as Henrik. I tried fvwm instead of KDE/Gnome, no changes.

Regards...
Michael

John-Paul Stewart

unread,
Apr 22, 2006, 3:19:55 PM4/22/06
to

OK. Here's my info (note that I have two hyperthreaded CPUs, not four
physical CPUs):

VGA compatible controller: ATI Technologies Inc Radeon R300 NG [FireGL
X1] (rev 80)
model name : Intel(R) Xeon(TM) CPU 3.06GHz
cpu MHz : 3067.209
model name : Intel(R) Xeon(TM) CPU 3.06GHz
cpu MHz : 3067.209
model name : Intel(R) Xeon(TM) CPU 3.06GHz
cpu MHz : 3067.209
model name : Intel(R) Xeon(TM) CPU 3.06GHz
cpu MHz : 3067.209
X.Org version: 6.9.0


direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

20352 frames in 5.0 seconds = 4070.318 FPS
20986 frames in 5.0 seconds = 4197.098 FPS
20994 frames in 5.0 seconds = 4198.720 FPS
20847 frames in 5.0 seconds = 4169.368 FPS

That's with the 8.23 version of ATI's drivers. With the newest 8.24
version:

21138 frames in 5.0 seconds = 4227.480 FPS
21188 frames in 5.0 seconds = 4237.599 FPS
21162 frames in 5.0 seconds = 4232.217 FPS
21156 frames in 5.0 seconds = 4231.124 FPS

When I did try the opensource drivers (back in December) I only got
about 2650 FPS with this card (along with display corruption). I
haven't re-tried the opensource drivers since then.

Michael Heiming

unread,
Apr 22, 2006, 5:09:03 PM4/22/06
to
In comp.os.linux.hardware John-Paul Stewart <jpst...@binaryfoundry.ca>:

> Henrik Carlqvist wrote:
>> John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
>>
>>>R. Steven Rainwater wrote:

[ benchmarking graphics adapter ]

Wouldn't call the outcome "tuning"?

Mh, looks like a configuration problem or you have been ripped
off completely with your $$ machines? This is just a single CPU
box:

model name : AMD Athlon(tm) 64 Processor 3000+
cpu MHz : 2000.000
X.Org version: 7.0.0
direct rendering: Yes
server glx vendor string: NVIDIA Corporation
server glx version string: 1.4
54457 frames in 5.0 seconds = 10891.298 FPS
54431 frames in 5.0 seconds = 10886.148 FPS
54434 frames in 5.0 seconds = 10886.640 FPS
54432 frames in 5.0 seconds = 10886.289 FPS

Running FC5, with nothing I'd call even near highend:

cat /proc/driver/nvidia/cards/0
Model: GeForce 6600 GT

Still it outperforms your systems in order of magnitudes. Strange
isn't it?

One guess would be you haven't enabled agpgart or whatever you
use with ATI enabled properly?

Another, cheapo AMD + evil nvidia driver outperform overpriced
Intel + evil ATI driver easily. Sadly you might have only paid
for all the marketing hype, but disappointing performance is all
you got.

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvp...@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 466: last entry

Vladimir Florinski

unread,
Apr 22, 2006, 7:17:17 PM4/22/06
to
On Sat, 22 Apr 2006 23:09:03 +0200, Michael Heiming wrote:

>
> Mh, looks like a configuration problem or you have been ripped
> off completely with your $$ machines? This is just a single CPU
> box:
>
> model name : AMD Athlon(tm) 64 Processor 3000+
> cpu MHz : 2000.000
> X.Org version: 7.0.0
> direct rendering: Yes
> server glx vendor string: NVIDIA Corporation
> server glx version string: 1.4
> 54457 frames in 5.0 seconds = 10891.298 FPS
> 54431 frames in 5.0 seconds = 10886.148 FPS
> 54434 frames in 5.0 seconds = 10886.640 FPS
> 54432 frames in 5.0 seconds = 10886.289 FPS
>
> Running FC5, with nothing I'd call even near highend:
>
> cat /proc/driver/nvidia/cards/0
> Model: GeForce 6600 GT
>
> Still it outperforms your systems in order of magnitudes. Strange
> isn't it?
>

glxgears is a bogomips equivalent of OpenGL. It is only a measure of GPU
frequency. The 6600GT runs at a very high clock speed so it registers a
correspondingly high glxgears score.

> One guess would be you haven't enabled agpgart or whatever you
> use with ATI enabled properly?
>
> Another, cheapo AMD + evil nvidia driver outperform overpriced
> Intel + evil ATI driver easily. Sadly you might have only paid
> for all the marketing hype, but disappointing performance is all
> you got.

My position on this is that only NVidia cards have any support worth
mentioning in Linux. Open source drivers (especially for chips where
manufactures release specifications, such as Intel) are so poor that
I cannot honestly call these "supported". I wish manufactures follow
NVidia's lead and stop releasing specs, but rather focus on developing
quality closed-source drivers.

--
Vladimir

John-Paul Stewart

unread,
Apr 22, 2006, 7:10:31 PM4/22/06
to
Michael Heiming wrote:
>
>
> Mh, looks like a configuration problem or you have been ripped
> off completely with your $$ machines?

I wouldn't jump to that conclusion based on glxgears "benchmarks" (using
the term loosely). It's comparable to using BogoMIPS to judge CPU speed.

Somebody else in the thread used Quake FPS scores as a benchmark.
That's better than glxgears, especially if you're using your card for
serious gaming. Me, I prefer to look at the SPEC viewperf scores, since
that's a demanding workload with real applications (Unigraphics, CATIA,
Pro/ENGINEER from the engineering graphics side of things, and
Lightscape and Maya from the 3-D animation side of the equation, among
others). My FireGL X1 had some impressive scores on those benchmarks at
the time I bought (and that was more-or-less the type of workload I was
buying it for...not games).

Also keep in mind that these aren't the newest designs. My FireGL X1
card has been in my system for just under 3 years...and it was product
that had been on the market for several months at that time. It's
several graphics card generations and 3.5 years old now.

Michael Heiming

unread,
Apr 22, 2006, 8:31:59 PM4/22/06
to
In comp.os.linux.hardware Vladimir Florinski <vflo...@ucr.edu>:

> On Sat, 22 Apr 2006 23:09:03 +0200, Michael Heiming wrote:
[..]

>> Another, cheapo AMD + evil nvidia driver outperform overpriced
>> Intel + evil ATI driver easily. Sadly you might have only paid
>> for all the marketing hype, but disappointing performance is all
>> you got.

> My position on this is that only NVidia cards have any support worth
> mentioning in Linux. Open source drivers (especially for chips where
> manufactures release specifications, such as Intel) are so poor that
> I cannot honestly call these "supported". I wish manufactures follow
> NVidia's lead and stop releasing specs, but rather focus on developing
> quality closed-source drivers.

I'd rather prefer GPL drivers but had to get a card with decent
3D performance under Linux, while still halfway affordable. Since
I don't do any serious gaming, asked a friend who is deep into it
and he strongly recommended Nvidia under Linux.

Sure glxgears just gives an estimation, like Bogomips and hdparm
what the device might be capable of. But it shows at least a
tendency like others.

There are still some minor problems, like agpart sometimes not
enabled probably during boot and you can't do much about it, then
reboot. You can use NVAGP, but only if agpart isn't compiled
directly into your kernel. Due to the frequent kernel upgrades
through yum on FC5 this would need some time/work to recompile
the kernel every two weeks or automate this. Haven't heard about
nvagp giving some advantage about agpart, though I haven't
tested.

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvp...@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'

#bofh excuse 183: filesystem not big enough for Jumbo Kernel
Patch

Michael Heiming

unread,
Apr 22, 2006, 8:35:30 PM4/22/06
to
In comp.os.linux.hardware John-Paul Stewart <jpst...@binaryfoundry.ca>:
> Michael Heiming wrote:
>>
>>
>> Mh, looks like a configuration problem or you have been ripped
>> off completely with your $$ machines?

> I wouldn't jump to that conclusion based on glxgears "benchmarks" (using
> the term loosely). It's comparable to using BogoMIPS to judge CPU speed.

Sure, should have added ;-) after those lines. ;-)

[stuff]

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo zvp...@urvzvat.qr | perl -pe 'y/a-z/n-za-m/'

#bofh excuse 230: Lusers learning curve appears to be fractal

Henrik Carlqvist

unread,
Apr 22, 2006, 8:42:06 PM4/22/06
to
michae...@gmx.de (Michael Mauch) wrote:
> Interesting. I have a ATI Radeon 9250. I changed your script to use
> "lspci" instead of "cat /proc/pci", because I don't have /proc/pci.

Good suggestion. However, I had to add the path to /sbin/lspci as I don't
have /sbin in my path as a normal user.



> # lspci | grep VGA | colrm 1 4 ; cat /proc/cpuinfo | \ egrep "model
> name|MHz" ; xdpyinfo | grep version: ; glxinfo | \ grep -A2 "direct
> rendering" ; glxgears & sleep 25 ; killall glxgears

> These numbers where about the same with Xorg 6.8.2. So why are my


> numbers so much lower than yours? You are not running that
> closed-source fglrx driver, are you?

No, I am running the opensource drivers from X.org 6.8.0. I was hoping
that this would have been shown by the glx verndor string and glx version
string. However, another post from a user using fglrx also had SGI as
vendor string and glx version 1.2.

As you already have seen, some of the performance difference could be
explaned by the EnablePageFlip option. Most of it is probably also
explained by the fact that I am using 16 bit color depth. Some of the
difference might also be explained that the 9250 despite its higher number
is a little bit slower than 9200. A 9250 is identical to a 9200 except
that the core speed is 240 MHz for 9250 and 250 MHz for 9200. The 9250SE
is exaclty like 9250 except that it only has 64 bit bus to the memory
instead of 128 bit. Unfortunately not all 9250SE cards sold are clearly
marked as SE. I have seen 9250 cards from different vendors with very
different performance even though they had exactly the same PCI IDs. I
don't remember the numbers exactly now, but I think that one card had
about 2800 fps in glxgears and the other card had about 2300 fps in
machines that otherwise were configured in the same way. I don't know if
the card that gave the lower fps really is an SE card.

> How big is your glxgears window (here it's 300x300 by default, shown by
> xwininfo)?

My window also has the default size of 300x300.



> What's your screen size and color depth?
>
> # xdpyinfo | egrep "dimensions|depth of"
> dimensions: 1280x1024 pixels (339x271 millimeters) depth of root
> window: 24 planes

That was a very good addition to the script:

/sbin/lspci | grep VGA | colrm 1 8 ; cat /proc/cpuinfo | \
egrep "model name|MHz" ; xdpyinfo | \
egrep "version:|dimensions|depth of" ; glxinfo | \


grep -A2 "direct rendering" ; glxgears & sleep 25 ; killall glxgears

VGA compatible controller: ATI Technologies Inc: Unknown device 5961 (rev 01)


model name : Intel(R) Pentium(R) 4 CPU 2.60GHz
cpu MHz : 2598.767
model name : Intel(R) Pentium(R) 4 CPU 2.60GHz
cpu MHz : 2598.767
X.Org version: 6.8.0

dimensions: 1280x1024 pixels (361x271 millimeters)
depth of root window: 16 planes


direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

11380 frames in 5.0 seconds = 2276.000 FPS
13721 frames in 5.0 seconds = 2744.200 FPS
14129 frames in 5.0 seconds = 2825.800 FPS
14561 frames in 5.0 seconds = 2912.200 FPS
14331 frames in 5.0 seconds = 2866.200 FPS

Henrik Carlqvist

unread,
Apr 22, 2006, 8:44:57 PM4/22/06
to
John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
> server glx vendor string: SGI
> server glx version string: 1.2

> That's with the 8.23 version of ATI's drivers. With the newest 8.24
> version:

I was hoping that at least the vendor string and maybe also the version
string would change with the binary drivers. At least they did change for
a user with the nVidia binary drivers. Do you know of any good command
line tool that can tell which driver you are using?

John-Paul Stewart

unread,
Apr 23, 2006, 1:35:56 PM4/23/06
to
Henrik Carlqvist wrote:
> John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
>
>>server glx vendor string: SGI
>>server glx version string: 1.2
>
>
>>That's with the 8.23 version of ATI's drivers. With the newest 8.24
>>version:
>
>
> I was hoping that at least the vendor string and maybe also the version
> string would change with the binary drivers. At least they did change for
> a user with the nVidia binary drivers. Do you know of any good command
> line tool that can tell which driver you are using?

The *OpenGL* vendor and version string *do* change with the binary
drivers.

The earlier script used 'glxinfo | grep -A2 "direct rendering"' to get
the *GLX* extension vendor/version string (which should always be the
same, for any verion of X.org regardless of driver since they're all
using the same GLX version) whereas the *OpenGL* vendor/version info is:

OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: Fire GL X1 Pentium 4 (SSE2) (FireGL) (GNU_ICD)
OpenGL version string: 2.0.5755 (8.24.8)

That clearly shows the 8.24.8 version of the ATI driver is in use. I'd
change your 'glxinfo | grep -A2 "direct rendering"' to 'glxinfo | grep
-A2 "OpenGL vendor"' instead...assuming that works for non-ATI drivers.

Henrik Carlqvist

unread,
Apr 24, 2006, 2:27:04 AM4/24/06
to
John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
> That clearly shows the 8.24.8 version of the ATI driver is in use. I'd
> change your 'glxinfo | grep -A2 "direct rendering"' to 'glxinfo | grep
> -A2 "OpenGL vendor"' instead...assuming that works for non-ATI drivers.

Thanks, that boils down to:

cat /proc/driver/nvidia/cards/0 || /sbin/lspci | grep VGA | colrm 1 8 ; \
cat /proc/cpuinfo | \
egrep "model name|MHz" ; xdpyinfo | egrep "version:|dimensions|depth of" ; \
glxinfo | egrep -A2 "direct rendering|OpenGL vendor" ; \


glxgears & sleep 25 ; killall glxgears

I'm keeping the egrep for "direct rendering" as those lines also was
changed by the nVidia drivers, maybe because they replace some more
libraries? Unfortunately "/sbin/lspci | grep VGA" was unable to identify
the GeForce 6600 GT another poster had so I also added the command
"cat /proc/driver/nvidia/cards/0"

Vladimir Florinski

unread,
Apr 24, 2006, 2:39:32 AM4/24/06
to
On Sat, 22 Apr 2006 13:54:43 +0200, Johannes Truschnigg wrote:

>
> Btw, I'd be happy to see some figures of Quake 3 running on Intel
> Graphics Accelerator 950 based hardware. So if anyone can provide
> those, please post.
>

Well, I have this video card in my laptop (Lenovo X60s). Tell me where to
download a Quake 3 demo, and I will run a benchmark for you.

--
Vladimir

R. Steven Rainwater

unread,
May 10, 2006, 1:46:42 PM5/10/06
to
Henrik Carlqvist wrote:
> John-Paul Stewart <jpst...@binaryfoundry.ca> wrote:
>>R. Steven Rainwater wrote:
>>>What I haven't been able to find is a comprehensive site that offers a
>>>list of graphics hardware and benchmarks for free software 3D drivers.
>>>Most sites only have benchmarks for proprietary stuff. Anyone know of a
>>>website with anything like that?
>
> Sorry, I have also searched without finding any good sites.
>
> I did a quick shell hack:
> cat /proc/pci | grep VGA | colrm 1 4 ; cat /proc/cpuinfo | \
> egrep "model name|MHz" ; xdpyinfo | grep version: ; glxinfo | \
> grep -A2 "direct rendering" ; glxgears & sleep 25 ; killall glxgears
>
> If some more people try the same script and post the result we might be
> able to get a picture of the performance difference between different
> cards and drivers.

I had some spare time, so I've set up a wiki to collect free software
3D/OpenGL driver performance results using your script. The site
includes the latest version of the script, incorporating the changes
suggest by others on this thread. At present there are only a handful of
results but I'm hoping over time we can gather more. I'm also collecting
information on how we can help create better free software 3D/OpenGL
support. I've added some links to hardware reverse engineering projects,
DRI development projects, and even a project developing an open hardware
graphics accelerator card.

http://www.free3d.org/

-Steve

John-Paul Stewart

unread,
May 10, 2006, 2:36:59 PM5/10/06
to
R. Steven Rainwater wrote:
>
> I had some spare time, so I've set up a wiki to collect free software
> 3D/OpenGL driver performance results using your script. The site
> includes the latest version of the script, incorporating the changes
> suggest by others on this thread. At present there are only a handful of
> results but I'm hoping over time we can gather more. I'm also collecting
> information on how we can help create better free software 3D/OpenGL
> support. I've added some links to hardware reverse engineering projects,
> DRI development projects, and even a project developing an open hardware
> graphics accelerator card.
>
> http://www.free3d.org/

The FireGL numbers on your site look suspiciously like the ones I posted
here a while ago. If that's the case, then resolution was 1280x1024 and
colour depth was 24 bits for those tests.

I think it's also worth noting that the opensource DRI drivers for that
particular chip were new and not fully developped at the time of the
test. Support for that card *will* improve with time.

You might also want to consider showing the closed-source driver numbers
for comparision. That way people can evaluate whether or not they're
worth the hassle.

R. Steven Rainwater

unread,
May 10, 2006, 5:54:04 PM5/10/06
to
John-Paul Stewart wrote:
> R. Steven Rainwater wrote:
>> I had some spare time, so I've set up a wiki to collect free
>> software 3D/OpenGL driver performance results using your script...

>>
>> http://www.free3d.org/
>
> The FireGL numbers on your site look suspiciously like the ones I posted
> here a while ago. If that's the case, then resolution was 1280x1024 and
> colour depth was 24 bits for those tests.

Yes, I think they from your earlier post. I tried to add all the results
posted in this thread. A couple more have been added since I set up the
site. I considered adding an email addresses to each result like the
bogomips list but decided it was probably not a good idea.

> I think it's also worth noting that the opensource DRI drivers for that
> particular chip were new and not fully developped at the time of the
> test. Support for that card *will* improve with time.

My hope is to see support for all the cards improve until the
performance of the free drivers is comparable to that of the non-free
drivers.

> You might also want to consider showing the closed-source driver numbers
> for comparision. That way people can evaluate whether or not they're
> worth the hassle.

My guess is that anyone willing to consider using the non-free drivers
would probably not want to bother with the slower free drivers right
now. But for those who prefer to use only free software for whatever
reason, I wanted to make it easy to find the fastest configuration. It
might be worth posting some common values for non-free drivers just for
comparison but I want to keep the focus of the page on the free drivers.

-Steve

Henrik Carlqvist

unread,
May 15, 2006, 1:56:33 PM5/15/06
to
"R. Steven Rainwater" <st...@ncc.com> wrote:
> I had some spare time, so I've set up a wiki to collect free software
> 3D/OpenGL driver performance results using your script. The site
> includes the latest version of the script, incorporating the changes
> suggest by others on this thread. At present there are only a handful of
> results but I'm hoping over time we can gather more.

Today I ran the script at some machines with different cards:

Matrox G550:
VGA compatible controller: Matrox Graphics, Inc. MGA G550 AGP (rev 1).
model name : Intel(R) Pentium(R) 4 CPU 2.80GHz
cpu MHz : 2813.567
model name : Intel(R) Pentium(R) 4 CPU 2.80GHz
cpu MHz : 2813.567
XFree86 version: 4.3.0
dimensions: 1280x1024 pixels (401x302 millimeters)


depth of root window: 16 planes

direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

--
OpenGL vendor string: VA Linux Systems Inc.
OpenGL renderer string: Mesa DRI G400 20020221 AGP 4x x86/MMX/SSE
OpenGL version string: 1.2 Mesa 4.0.4
3329 frames in 5.0 seconds = 665.800 FPS
3427 frames in 5.0 seconds = 685.400 FPS
3426 frames in 5.0 seconds = 685.200 FPS
3426 frames in 5.0 seconds = 685.200 FPS
3426 frames in 5.0 seconds = 685.200 FPS

Radeon 9250, DRI disabled with "export LIBGL_ALWAYS_INDIRECT=1"
VGA compatible controller: PCI device 1002:5960 (ATI Technologies Inc) (rev 1).
model name : AMD Opteron(tm) Processor 250
cpu MHz : 2405.499
model name : AMD Opteron(tm) Processor 250
cpu MHz : 2405.499
X.Org version: 6.8.0
dimensions: 1600x1200 pixels (411x311 millimeters)


depth of root window: 16 planes

direct rendering: No


server glx vendor string: SGI
server glx version string: 1.2

--
OpenGL vendor string: Mesa project: www.mesa3d.org
OpenGL renderer string: Mesa GLX Indirect
OpenGL version string: 1.2 (1.5 Mesa 6.1)
3453 frames in 5.0 seconds = 690.600 FPS
4080 frames in 5.0 seconds = 816.000 FPS
4080 frames in 5.0 seconds = 816.000 FPS
4080 frames in 5.0 seconds = 816.000 FPS
4080 frames in 5.0 seconds = 816.000 FPS

Radeon 9250, DRI enabled
VGA compatible controller: PCI device 1002:5960 (ATI Technologies Inc) (rev 1).
model name : AMD Opteron(tm) Processor 250
cpu MHz : 2405.499
model name : AMD Opteron(tm) Processor 250
cpu MHz : 2405.499
X.Org version: 6.8.0
dimensions: 1600x1200 pixels (411x311 millimeters)


depth of root window: 16 planes

direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

--
OpenGL vendor string: Tungsten Graphics, Inc.
OpenGL renderer string: Mesa DRI R200 20030328 AGP 4x x86/MMX+/3DNow!+/SSE2 TCL
OpenGL version string: 1.3 Mesa 6.1
12138 frames in 5.0 seconds = 2427.600 FPS
13045 frames in 5.0 seconds = 2609.000 FPS
12964 frames in 5.0 seconds = 2592.800 FPS
13045 frames in 5.0 seconds = 2609.000 FPS

Radeon 9250 in another machine
VGA compatible controller: PCI device 1002:5960 (ATI Technologies Inc) (rev 1).
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cpu MHz : 3600.260
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cpu MHz : 3600.260
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cpu MHz : 3600.260
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cpu MHz : 3600.260
X.Org version: 6.8.0
dimensions: 1680x1050 pixels (431x272 millimeters)


depth of root window: 16 planes

direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

--
OpenGL vendor string: Tungsten Graphics, Inc.
OpenGL renderer string: Mesa DRI R200 20030328 AGP 4x x86/MMX/SSE2 TCL
OpenGL version string: 1.3 Mesa 6.1
12962 frames in 5.0 seconds = 2592.400 FPS
13190 frames in 5.0 seconds = 2638.000 FPS
13141 frames in 5.0 seconds = 2628.200 FPS
13191 frames in 5.0 seconds = 2638.200 FPS

Radeon 9000Pro
VGA compatible controller: ATI Technologies Inc Radeon R250 If [Radeon 9000]
(rev 1).
model name : Intel(R) Pentium(R) 4 CPU 3.06GHz
cpu MHz : 3057.993
model name : Intel(R) Pentium(R) 4 CPU 3.06GHz
cpu MHz : 3057.993
X.Org version: 6.8.0
dimensions: 1680x1050 pixels (431x272 millimeters)


depth of root window: 16 planes

direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.2

--
OpenGL vendor string: Tungsten Graphics, Inc.
OpenGL renderer string: Mesa DRI R200 20030328 AGP 4x x86/MMX/SSE2 TCL
OpenGL version string: 1.3 Mesa 6.1
15402 frames in 5.0 seconds = 3080.400 FPS
17096 frames in 5.0 seconds = 3419.200 FPS
17025 frames in 5.0 seconds = 3405.000 FPS
17093 frames in 5.0 seconds = 3418.600 FPS
17025 frames in 5.0 seconds = 3405.000 FPS

All the radeon configurations was tuned with the options "AGPMode" "4" and
"EnablePageFlip" "on". The Matrox configuration was tuned with the option
"AGPMode" "4".

It is interesting to see that the Radeon 9250 is faster with DRI disabled
than Matrox G550 with DRI enabled. The fastest card so far is the Radeon
9000Pro, but that one is even more obsolete than Radeon 9200. I suppose
that a Radeon 8500 would be even faster.

I tried to edit your wiki to add these benchmarks, but the page was read
only for me. I think it would be a good idea to add a column "options"
between the bit depth and fps column. In that column things like "AGPMode"
"4" and other things that affects performance could be noted.

R. Steven Rainwater

unread,
May 15, 2006, 4:31:21 PM5/15/06
to
Henrik Carlqvist wrote:
> "R. Steven Rainwater" <st...@ncc.com> wrote:
>>
>> http://www.free3d.org/

>
> Today I ran the script at some machines with different cards:

Thanks, I've added these results to the benchmark list.

> It is interesting to see that the Radeon 9250 is faster with DRI disabled
> than Matrox G550 with DRI enabled.

I believe this is because the Matrox driver only supports a small number
of DRI features compared to the ATI drivers. I assume the unsupported
features still rely on Mesa software emulation. See the Freedesktop DRI
feature matrix:

http://dri.freedesktop.org/wiki/FeatureMatrix

> The fastest card so far is the Radeon 9000Pro, but that one is even
> more obsolete than Radeon 9200. I suppose that a Radeon 8500 would
> be even faster.

It may be old but I was able to find plenty of them for sale online (new
and used), very inexpensively. Based on the performance results so far,
I'm now recommending the 9000 Pro. I wish I'd know that a week ago - I
just bought a 9250! :-)

> I tried to edit your wiki to add these benchmarks, but the page was read
> only for me.

Edit access, requires an account login (to prevent wiki spamming). Click
the "Login" button at the bottom and then "register" and it will email
you a password.


> I think it would be a good idea to add a column "options" between
> the bit depth and fps column.

Done...

-Steve

Henrik Carlqvist

unread,
May 16, 2006, 1:35:25 AM5/16/06
to

I did that and after logging in I get buttons that say "edit". When I
click the edit button for the table I see the source for the table but I
am still unable to edit it. Above the source I see the following text:

-8<-----------------------------------------------------------
This page is read only. You can view the source, but not change it. Ask
your administrator if you think this is wrong.
-8<-----------------------------------------------------------

The Radeon 9250 with DRI disabled also had the EnablePageFlip option on.
My Radeon 9200 also had the options AGPMode 4 and EnablePageFlip on.

Henrik Carlqvist

unread,
May 16, 2006, 1:41:59 AM5/16/06
to
"R. Steven Rainwater" <st...@ncc.com> wrote:
> I'm now recommending the 9000 Pro. I wish I'd know that a week ago - I
> just bought a 9250! :-)

I would guess that the 9200 Pro would be at least as fast as the 9000 Pro
if used with the same options and with the same bit depth. X.org 7.0.0
also is able to set AGPMode to 8 which the 9200 series should be capable
of.

R. Steven Rainwater

unread,
May 16, 2006, 2:23:00 PM5/16/06
to
Henrik Carlqvist wrote:
> "R. Steven Rainwater" <st...@ncc.com> wrote:
>>Edit access, requires an account login (to prevent wiki spamming). Click
>>the "Login" button at the bottom and then "register" and it will email
>>you a password.
>
> I did that and after logging in I get buttons that say "edit". When I
> click the edit button for the table I see the source for the table but I
> am still unable to edit it. Above the source I see the following text:
>
> -8<-----------------------------------------------------------
> This page is read only. You can view the source, but not change it. Ask
> your administrator if you think this is wrong.
> -8<-----------------------------------------------------------

Sorry about that. I haven't used DokuWiki much and I think I may not
have had the ACLs configured incorrectly. It seems to be letting me set
up user accounts with edit permissions now. So I think it's fixed.

-Steve

Henrik Carlqvist

unread,
May 16, 2006, 5:08:25 PM5/16/06
to

I am still unable to edit, with the same message about "Ask your
administrator...". On the other hand I no longer have anything to add, all
the contents seems complete and correct to me now.

Reply all
Reply to author
Forward
0 new messages