glxgears comprison

[Fintan]

Just for the hell of it I did the following comparison with glxgears between
Kubuntu Gutsy Tribe3 (updated today):
average: 195.543 FPS

Ubuntu Gutsy Tribe3 (updated today):
average: 1482.377 FPS

pclinuxos 2007:
average: 3162.939 FPS

these were done on an ati 9250 pro card and installation without installing anything extra or configuring xorg.conf.

I then copied the xorg.conf over from pclinuxos to the kubuntu gutsy intsall and got this:
average: 552.477 FPS

Which is a nice improvement but nowhere near ubuntu much less pclinuxos.

I realize that glxgears is not necessarily precise but the discrepancies are there.

Food for thought would be appreciated.

EDIT:
Oh I forgot the test I did with Mint Cassandra kde community edition beta vesion .13 (based on kubuntu feisty):
average: 2541.672 FPS

[dibl]

Interesting. Knowing zero about making an ATI card work (except "VESA"), all I can wonder is, are the xorg.conf options all the same, between the Kubuntu and Ubuntu installations? If yes, then it would seem that the two different X servers, gdm and kdm, are running your graphics at different speeds, for some reason. And of course whatever X server PCLinux is using is 3 times faster. :P

[Fintan]

Seems like it and I just inserted my forgotten Mint test which is based on Kubuntu Feisty.

So it gets even more interesting. I would love to know how to tweek this one.

As far as I can tell the ubuntu and kubuntu xorgs were the same. The mint and pclinuxos are the same as well but different from the *buntu. I can post them if needed.

[dibl]

I've never made much of a study of GLX, but obviously it has enabled some interesting capabilities with Beryl and Compiz. glxgears basically just informs you how well-suited your particular hardware and software setup is to run a glx application, AFAIK. But of course the X server system that runs your Linux GUI is also a factor in any screen presentation, so I would think that you may be observing some difference between the Gnome gdm xserver, and the KDE kdm xserver. Just a simple-minded guess, but I don't know what else would explain the difference on the same hardware platform.

[Fintan]

As far as gdm and kdm in *buntu so far so good (or bad) but what about the difference between kdm in pclinux (kde) and kubuntu gutsy and mint both kubuntu based? Both pclinux and mint are "faster" ??? and kubuntu gutsy is 3 times "faster" with the pclinux or mint xorg.conf ::)

[myNigga]

There is a "trick" that many users are not aware of. Open the console and leave glxgears running for about 30 seconds, so you get a stable output of fps. Then just give konsole the focus, making it the current window and thus, sending the glxgears window to the back. Let it run a bit and see the difference.
In my setup, when glxgears is the active window, something between 9850fps and 10100fps.
When konsole has the focus instead of glxgears, I get 14700-15200 fps.
With this kind of instability, glxgears can't be a reliable benchmarker, it's sole purpose is to show whether a system has acceleration or not, IMHO (and it is also a good tool for burning in the CPU).
When you "benchmark" again, make sure all of the distros have the same conditions running glxgears, that is, either all of them have focus on it or none has it.

[dibl]

Cool -- I never noticed that. Gotta go -- gotta run glxgears!

;D

[dibl]

Eh -- it didn't change on mine. This isn't the best my system can do, because (a) Beryl is running already, and (b) I'm capturing a videotape at the moment. But, in the following output, it starts with the focus on the glxgears window, then I changed it to the console for about 6 cycles, then I changed it back to glxgears for the rest of the run. Ad you can see, it didn't seem to make a big difference:


dibl@ufeisty64:~$ glxgears
54047 frames in 5.0 seconds = 10809.392 FPS
54351 frames in 5.0 seconds = 10870.133 FPS
53558 frames in 5.0 seconds = 10711.443 FPS
52810 frames in 5.0 seconds = 10561.844 FPS
57801 frames in 5.0 seconds = 11560.153 FPS
57749 frames in 5.0 seconds = 11549.756 FPS
52633 frames in 5.0 seconds = 10526.562 FPS
54533 frames in 5.0 seconds = 10906.469 FPS
54225 frames in 5.0 seconds = 10844.963 FPS
54513 frames in 5.0 seconds = 10902.497 FPS
54635 frames in 5.0 seconds = 10926.850 FPS
54591 frames in 5.0 seconds = 10918.123 FPS
54483 frames in 5.0 seconds = 10896.566 FPS
54519 frames in 5.0 seconds = 10903.761 FPS
54724 frames in 5.0 seconds = 10944.635 FPS
54546 frames in 5.0 seconds = 10909.182 FPS
54341 frames in 5.0 seconds = 10868.160 FPS
54511 frames in 5.0 seconds = 10902.064 FPS
54575 frames in 5.0 seconds = 10914.827 FPS
54630 frames in 5.0 seconds = 10925.917 FPS
54604 frames in 5.0 seconds = 10920.796 FPS
54515 frames in 5.0 seconds = 10902.969 FPS
54583 frames in 5.0 seconds = 10916.517 FPS

[myNigga]

Hey, Dibl! =)

Here is some output of my rig:



48180 frames in 5.0 seconds = 9635.923 FPS
49666 frames in 5.0 seconds = 9933.135 FPS
49698 frames in 5.0 seconds = 9939.434 FPS
49688 frames in 5.0 seconds = 9937.524 FPS
49689 frames in 5.0 seconds = 9937.766 FPS
49688 frames in 5.0 seconds = 9937.511 FPS
49704 frames in 5.0 seconds = 9940.698 FPS
49691 frames in 5.0 seconds = 9938.168 FPS
49691 frames in 5.0 seconds = 9938.141 FPS
75361 frames in 5.0 seconds = 15072.082 FPS
82819 frames in 5.0 seconds = 16563.707 FPS
82820 frames in 5.0 seconds = 16563.967 FPS
82822 frames in 5.0 seconds = 16564.396 FPS
82796 frames in 5.0 seconds = 16559.184 FPS
82844 frames in 5.0 seconds = 16568.652 FPS
82828 frames in 5.0 seconds = 16565.410 FPS
82847 frames in 5.0 seconds = 16569.305 FPS
82830 frames in 5.0 seconds = 16565.971 FPS
81675 frames in 5.0 seconds = 16334.888 FPS
54179 frames in 5.0 seconds = 10835.727 FPS
49700 frames in 5.0 seconds = 9939.990 FPS
49690 frames in 5.0 seconds = 9937.992 FPS
49695 frames in 5.0 seconds = 9938.809 FPS
49708 frames in 5.0 seconds = 9941.467 FPS
49698 frames in 5.0 seconds = 9939.557 FPS
49638 frames in 5.0 seconds = 9927.442 FPS
49694 frames in 5.0 seconds = 9938.688 FPS
49705 frames in 5.0 seconds = 9940.811 FPS
49636 frames in 5.0 seconds = 9927.069 FPS


You see the first runs when glxgears has focus, the intermediary output when konsole has focus, and then glxgears getting focus back.
This is without Beryl. I'm not saying that your reading is wrong, it's just something I noticed throughout the years running it in very different machines. First time I saw it was in 2001, using Mandrake with Mesa, an ATI Radeon VIVO 64MB card and kernel 2.2.
I could later observe it with a Radeon 8500 using ATI's driver, then with a MX440 and a 6600GT card. I found it curious that glxgears had this kind of throttling when losing focus.
Maybe you want to try without Beryl. ;D

[dibl]

That's crazy! ;D

Yeah, OK, you have piqued my curiosity. I've got about 10 windows to close (but not my video tape capture), and then I'm turning off Beryl, and then we'll see what changes.

[dibl]

Well, it's always interesting to see what happens to the video performance when you shut down Beryl, close Firefox, overclock your Nvidia card a little, and stuff like that! ;D ;D ;D

OK, in this run, the first 5 fps calculations were with the focus on the console window, the second 5 were with the focus on the glxgears window, and the last 6 were back on the console window. Clearly, on my rig, the reported fps rate is not "connected" to the windowing system. BTW, I'm running Ubuntu 64-bit at the moment, so possibly this is different for Gnome than KDE. I'll try it again in KDE, but not for a few hours while this video tape runs.



dibl@ufeisty64:~$ glxgears
96895 frames in 5.0 seconds = 19378.941 FPS
96932 frames in 5.0 seconds = 19386.293 FPS
96910 frames in 5.0 seconds = 19381.941 FPS
96812 frames in 5.0 seconds = 19362.260 FPS
97029 frames in 5.0 seconds = 19405.602 FPS
96566 frames in 5.0 seconds = 19313.084 FPS
96850 frames in 5.0 seconds = 19369.988 FPS
96861 frames in 5.0 seconds = 19372.057 FPS
96808 frames in 5.0 seconds = 19361.508 FPS
96837 frames in 5.0 seconds = 19367.311 FPS
96074 frames in 5.0 seconds = 19214.793 FPS
96932 frames in 5.0 seconds = 19386.311 FPS
96955 frames in 5.0 seconds = 19390.912 FPS
97009 frames in 5.0 seconds = 19401.670 FPS
96999 frames in 5.0 seconds = 19399.789 FPS

[myNigga]

Just for the sake of it, my rig is:
A64 X2 4800+, Geforce 6800 vanilla. Kubuntu 64bit. No overclock in CPU or vcard, and just using the regular setup of xorg.conf.

And here is my output with Beryl enabled:


30568 frames in 5.0 seconds = 6113.550 FPS
32027 frames in 5.0 seconds = 6403.357 FPS
32110 frames in 5.0 seconds = 6420.262 FPS
32044 frames in 5.0 seconds = 6407.094 FPS
32176 frames in 5.0 seconds = 6433.463 FPS
31848 frames in 5.0 seconds = 6369.594 FPS
32168 frames in 5.0 seconds = 6433.548 FPS
32076 frames in 5.0 seconds = 6415.125 FPS
32159 frames in 5.0 seconds = 6431.761 FPS


The first 4 reads are when glxgears had focus, the rest is when konsole had focus. Unchanged. This throttling seems to be KDE-related. BTW, I never did this test in Gnome because I always used KDE on my machines.

[dibl]

Are you running the Nvidia proprietary driver or "nv"?

My CPU is an Intel X6800 (2.9GHZ) overclocked to 3.3GHz, and the Nvidia card is a GF 7900GS. I overclocked the Nvidia card just for this little experiment -- I usually just let it run at its defaults.

Have you observed that difference in the glxgears speed on other hardware?

[dibl]

I just ran it in Kubuntu 32-bit -- no difference, really. The first 5 calculations are with the focus on the glxgears window, then I changed it to the Konsole window for 5 calculations, then back to glxgears. Doesn't seem to make much difference here. I'm running a 1600 x 1200 desktop, btw, on a Samsung SyncMaster 1100 CRT. I didn't overclock the Nvidia card, so this is what it does at default, in KDE:


dibl@feisty:~$ glxgears
79168 frames in 5.0 seconds = 15833.419 FPS
85407 frames in 5.0 seconds = 17081.380 FPS
85456 frames in 5.0 seconds = 17091.002 FPS
85442 frames in 5.0 seconds = 17088.325 FPS
85386 frames in 5.0 seconds = 17077.053 FPS
85301 frames in 5.0 seconds = 17060.136 FPS
85417 frames in 5.0 seconds = 17083.341 FPS
85427 frames in 5.0 seconds = 17085.284 FPS
85401 frames in 5.0 seconds = 17080.124 FPS
85431 frames in 5.0 seconds = 17086.064 FPS
85235 frames in 5.0 seconds = 17046.972 FPS
85303 frames in 5.0 seconds = 17060.583 FPS
85423 frames in 5.0 seconds = 17084.549 FPS
85410 frames in 5.0 seconds = 17081.888 FPS
85414 frames in 5.0 seconds = 17082.687 FPS

[3vi1]

I'm jealous.

If I run glxgears in Kubuntu (64-bit) w/Compiz Fusion, my Xserver reboots.

[myNigga]

Yes, Dibl.
Here I have nvidia-glx-new (97.xx), but this also happened with 96.xx and some previous versions.
I hope my machines weren't cursed all these years. LOL
No overclock in CPU or GPU, stock speeds only.
I remember this behaviour in these distros: Mandrake, Conectiva (several versions) and now Kubuntu.
Don't blame me, I have nothing to do with that. =P

[Fintan]

If I run glxgears in Kubuntu (64-bit) w/Compiz Fusion, my Xserver reboots.

That was good ;D

BTY I noticed that little discrepancy when switching focus on windows.

My point was more the difference between distros on the same hardware.

Especially between *buntu's and derivates (mint).

[myNigga]

I see you have run the tests without acceleration, yes?
The discrepancy could be something with the particular MESA implementation, since it drives software OpenGL. I doubt distro-makers would mess with that, but the only other thing would be the video driver, which is even less likely to be modified by distro-makers.
I wouldn't think of a Xorg issue, this is just a graphical environment and doesn't really "drive" the mentioned hardware, it is all the module's task.

[goforthis]

glxgears uses the processor not the video hardware I just found that out
from another long time user
The things I keep learning ::)

[myNigga]

It makes sense when looking at Ksensors: CPU goes from 1000MHz to 2400MHz (full speed) when launching the app. I once saw that Warcraft3 and some other games had the CPU at full speed for about 5 to 10% of the time the game was being played.
But what about the performance increase when using an opengl capable video driver? And does it mean a 3GHz CPU with a Geforce7600 will run it faster than a 2GHz CPU with a Geforce8800, if the app is "CPU bound"?