Intel 945GM performance +1000fps boost -WTF!?

stokker · 2009-05-04 22:39:22

So, I have the Intel 945GM Chipset on my laptop, which used to have 2x 512Mb of RAM in dual channel mode.
Yesterday I bought myself a new 2Gb DDR2-800 stick of RAM, and used it instead of the old 512 sticks.

At first I was kind of disappointed when I noticed that I had lost the dual channel support ( actually I didn't even consider that when i bought the memory).
But when I launched glxgears something strange happen. I was used to have ~200fps due to the crappy state of the xf-video-intel
driver. But now, all of a sudden. glxgears runs with a framerate of >1250fps (!)

Soo, i was curious and made some tests :

glxgears, using dual channel memory, 2x 512Mb PC2-4200

 
1008 frames in 5.0 seconds = 201.581 FPS
1031 frames in 5.0 seconds = 206.148 FPS
1020 frames in 5.0 seconds = 203.938 FPS
1023 frames in 5.0 seconds = 204.484 FPS
1023 frames in 5.0 seconds = 204.467 FPS
1023 frames in 5.0 seconds = 204.475 FPS
1023 frames in 5.0 seconds = 204.583 FPS
cpu usage glxgears : 12%

glxgears, using singel channel memory 1x 2Gb PC2-6400

 
6148 frames in 5.0 seconds = 1229.585 FPS
6300 frames in 5.0 seconds = 1259.909 FPS
6285 frames in 5.0 seconds = 1256.988 FPS
6287 frames in 5.0 seconds = 1257.321 FPS
6291 frames in 5.0 seconds = 1258.088 FPS
6242 frames in 5.0 seconds = 1248.318 FPS
cpu usage glxgears 75%

My xorg.conf:

          
        Option     "AccelMethod"      "EXA"
        Option     "XAANoOffscreenPixmaps" "true"
        Option      "DRI"                  "true"
        Option      "AddARGBGLXVisuals"      "true"
        Option      "RenderAccel"       "true"
        
        Identifier  "Card0"
        Driver      "intel"
        VendorName  "Intel Corporation"
        BoardName   "Mobile 945GM/GMS, 943/940GML Express Integrated
        Graphics Controller"
        BusID       "PCI:0:2:0"

WTF !?
Shouldn't 2x PC2-4200 in dual ch. --> = 1066 Mhz (?)
be faster than 1x PC2-6400 single ch. = 800Mhz

Ok, the intel 945GM uses the ram memory as video memory.
But why should this effect my 3D performace this much, and shouldn't it be the other way around ?

Can someone enlightened explain why?

.

agapito · 2009-05-04 22:55:55

It´s a confirmed bug on intel upstream i think. With Dual Channel you lose performance.

moljac024 · 2009-05-04 22:56:31

stokker wrote:

Can someone enlightened explain why?

For starters, glxgears is not a benchmark....

buttons · 2009-05-04 23:22:26

moljac024 wrote:

stokker wrote:
Can someone enlightened explain why?
For starters, glxgears is not a benchmark....

Don't start this. For this case, which is controlled for other factors, it is fine. A ten-fold difference cannot be ignored, nor does it happen by chance.

Defend your statements or do not make them.

imrehg · 2009-05-05 05:31:31

Did you run the second glxgears minimized, by any chance?
That would increase the FPS numbers in itself. The quoted CPU percentages hint that direction as well:
when you have open window, the drawing is limited by the memory speed - CPU utilization will be low.
when you minimize the window, it does not actually have to draw anything, so it will be limited by the CPU - high utilization results...

this is just a guess, though.

My EeePC is running ~99fps open window (cpu ~16%), ~885fps minimized window (cpu ~97%), with the xf86-video-intel driver, 2Gb DDR2-800 RAM...

stokker · 2009-05-05 08:37:21

imrehg wrote:

Did you run the second glxgears minimized, by any chance?.

no I didn't, same "test" both times.. (not minimized). I switched memories a few times just to be sure.
Strange results ... even though it's a known bug -is this the performance we can expect when the driver is "ready". ?
but the increased cpu load is of course a bad side effect..

imrehg · 2009-05-05 08:46:57

then it's quite puzzling...

when i tried it again, just by putting any other window in front of the glxgears window I could see the same effect as minimizing it.

Have you tried any other benchmarking software, that could verify these performance differences?

stokker · 2009-05-05 09:07:20

Have you tried any other benchmarking software, that could verify these performance differences?

Yea could do that. Any nice benchmarking application for this purpose ?
- moljac024?

Damnshock · 2009-05-05 10:04:00

agapito wrote:

It´s a confirmed bug on intel upstream i think. With Dual Channel you lose performance.

References?

PS: does this mean that if I take one of my modules it will improve my graphics performance?

moljac024 · 2009-05-05 10:18:53

stokker wrote:

Have you tried any other benchmarking software, that could verify these performance differences?
Yea could do that. Any nice benchmarking application for this purpose ?
- moljac024?

http://www.phoronix-test-suite.com/

lanxu · 2009-05-05 12:54:24

If you don't want to hassle with Phoronix test suite, you can always run quake3 benchmark.

- Install OpenArena from AUR (http://aur.archlinux.org/packages.php?ID=7748)
- Get http://jumi.lut.fi/~julankin/files/3D_b … ena.tar.gz
- Extract and run sh benchmark.sh

example output:

840 frames 13.0 seconds 64.8 fps 6.0/15.4/34.0/4.0 ms

stefanwilkens · 2009-05-05 13:13:37

any chance you had something covering part of the glxgears screen on the second run?

stokker · 2009-05-05 15:18:52

stefanwilkens wrote:

any chance you had something covering part of the glxgears screen on the second run?

nope, thats the thing... it wasn't, I'm surprised myself.
But I will do some more testing later tonight (if I have the time) just to be on the safe side.

schivmeister · 2009-05-06 12:43:06

Definitely worth a report to (1) the xorg team, and (2) the intel oss team, unless as mentioned it's a confirmed bug indeed.

Damn it this is seriously bogus. We need more of you to gear up and develop the driver instead of having a life!

stokker · 2009-05-06 22:23:51

Here's for no life

There is a HUGE performance drop on dual channel...
This time I was using globs (GL Open Benchmark Suite) from aur.

1 x 512MB DDR2-533 single ch.

2 x 512MB DDR2-533 dual ch.

1 x 1GB DDR2-667 single ch.

1 x 2GB DDR2-800 single ch.

also ran the glxgears again, just for fun.

single ch. 1x 512Mb PC2-4200 DDR2-533
4841 frames in 5.0 seconds = 968.152 FPS
5044 frames in 5.0 seconds = 1008.674 FPS
5033 frames in 5.0 seconds = 1006.545 FPS
5031 frames in 5.0 seconds = 1006.023 FPS
5021 frames in 5.0 seconds = 1004.153 FPS
5046 frames in 5.0 seconds = 1009.144 FPS
glxgears : 49% cpu

dual ch. 2x 512Mb PC2-4200 DDR2-533
1009 frames in 5.0 seconds = 201.627 FPS
1034 frames in 5.0 seconds = 206.753 FPS
1025 frames in 5.0 seconds = 204.907 FPS
1025 frames in 5.0 seconds = 204.970 FPS
1025 frames in 5.0 seconds = 204.999 FPS
1025 frames in 5.0 seconds = 204.936 FPS
glxgears 12% cpu

single ch. 1x 1Gb PC2-5300 DDR2-667
6022 frames in 5.0 seconds = 1204.326 FPS
6106 frames in 5.0 seconds = 1221.068 FPS
6131 frames in 5.0 seconds = 1226.123 FPS
6098 frames in 5.0 seconds = 1219.478 FPS
6099 frames in 5.0 seconds = 1219.615 FPS
6047 frames in 5.0 seconds = 1209.260 FPS
glxgears 64% cpu

singel ch.  1x 2Gb PC2-6400 DDR2-800
6140 frames in 5.0 seconds = 1227.808 FPS
6265 frames in 5.0 seconds = 1252.823 FPS
6250 frames in 5.0 seconds = 1249.911 FPS
6260 frames in 5.0 seconds = 1251.900 FPS
6259 frames in 5.0 seconds = 1251.631 FPS
6260 frames in 5.0 seconds = 1251.887 FPS
glxgears 73% cpu

- Whats up with the low cpu usage on dual ch ?

Btw,
the hardware is a Fujitsu-Siemens Amilo Si1520 laptop

and the software :
kernel26 2.6.29.2-1
libdrm 2.4.9-1
libgl 7.4.1-1
intel-dri 7.4.1-1
xorg-server 1.6.1-1
xf86-video-intel 2.6.3-4
etc. etc

conclusion : don't use dual ch. ram with the 945GM

tuxfusion · 2009-05-10 13:26:29

I'm still on moljac024 statements "glxgears is not a benchmark" , and buttons why defend a fact ?
http://keithp.com/blogs/Sharpening_the_ … ver_Focus/
The whole Intel Driver thingy (EXA, XXA, UXA) seems to be extremely dependend on RAM. ( see above url)
I'm with EXA again on my eee901 with the "greedy mode" for the record, but you might try to use UXA here as EXA was stated to have Memory thrashing ( url above ).
But what really would help if you would just , for instance, install geekbench from AUR.
It has memory bandwidth and performance testing and is really simple to install & run compared to PTS(phoronix).

Regards

schivmeister · 2009-05-11 19:52:47

I disabled compositing. Got back the second-best performance at 1000FPS (glxgeras). Eyecandy is now in the form of KDE + Enlightenment.

Now to try and see what combination of driver and options will get me the optimal 1500.

glxgears is a benchmark in this case because it shows you whether performance has increased or decreased - and that's all that matters. Whether it's a real-world reflection or not, I don't care. Fact is when glxgears' performance increases, so does my 3D games'.

Arch Linux

#1 2009-05-04 22:39:22

Intel 945GM performance +1000fps boost -WTF!?

#2 2009-05-04 22:55:55

Re: Intel 945GM performance +1000fps boost -WTF!?

#3 2009-05-04 22:56:31

Re: Intel 945GM performance +1000fps boost -WTF!?

#4 2009-05-04 23:22:26

Re: Intel 945GM performance +1000fps boost -WTF!?

#5 2009-05-05 05:31:31

Re: Intel 945GM performance +1000fps boost -WTF!?

#6 2009-05-05 08:37:21

Re: Intel 945GM performance +1000fps boost -WTF!?

#7 2009-05-05 08:46:57

Re: Intel 945GM performance +1000fps boost -WTF!?

#8 2009-05-05 09:07:20

Re: Intel 945GM performance +1000fps boost -WTF!?

#9 2009-05-05 10:04:00

Re: Intel 945GM performance +1000fps boost -WTF!?

#10 2009-05-05 10:18:53

Re: Intel 945GM performance +1000fps boost -WTF!?

#11 2009-05-05 12:54:24

Re: Intel 945GM performance +1000fps boost -WTF!?

#12 2009-05-05 13:13:37

Re: Intel 945GM performance +1000fps boost -WTF!?

#13 2009-05-05 15:18:52

Re: Intel 945GM performance +1000fps boost -WTF!?

#14 2009-05-06 12:43:06

Re: Intel 945GM performance +1000fps boost -WTF!?

#15 2009-05-06 22:23:51

Re: Intel 945GM performance +1000fps boost -WTF!?

#16 2009-05-10 13:26:29

Re: Intel 945GM performance +1000fps boost -WTF!?

#17 2009-05-11 19:52:47

Re: Intel 945GM performance +1000fps boost -WTF!?

Board footer