34 replies to this topic

[Alex DeWolf]

Hello all:
I have an Asus F8SN with a Nvidia 9500M with 512MB VRAM. I got 10.5.4 installed and working with QE/CI support using the latest NVinject. However the OPENGL support is really slow. How can I get topengl working properly?

Alex

[spanakorizo]

how do u know is slow?

[Alex DeWolf]

spanakorizo, on Sep 12 2008, 09:50 PM, said:

how do u know is slow?

I think the attached image shows pretty slow opengl tests. Also HalfLife2 on Crossover is really slow on this machine, on a Mac Book with the ATI X3100 it is much smoother and faster.

Alex

Attached Files

•  Picture_3.png   56.53K   137 downloads

[Slice]

Alex DeWolf, on Sep 13 2008, 09:11 AM, said:

I think the attached image shows pretty slow opengl tests. Also HalfLife2 on Crossover is really slow on this machine, on a Mac Book with the ATI X3100 it is much smoother and faster.

Alex

OpenGL Extension Viewer is a strange program. As well as XBench. Better try Open Mark and Cinebench.
Crossover uses windows drivers so I doubt it can run fast OpenGL.
What is ATI X3100? May be you mean Intel X3100?

It would be nice if you publish your config

Quote

sudo -s
kextstat >kextstatDeWolf.txt
ioreg -l -x -w 2048 >ioregDeWolf.txt

[aqua-mac]

Can you show your info from system profiler under the section Graphics card. I doubt very much that you do actually have QE or CI enabled.

[mitch_de]

Slice, on Sep 13 2008, 08:00 AM, said:

OpenGL Extension Viewer is a strange program. As well as XBench. Better try Open Mark and Cinebench.
Crossover uses windows drivers so I doubt it can run fast OpenGL.
What is ATI X3100? May be you mean Intel X3100?

It would be nice if you publish your config

No, OpenGL Extensions Viewer benches are very good and reliable - much better than Xbench, where GMA950 are faster than ATI 2600XT!!!! Xbench OpenGL values are for the trash (rest of valuesa are OK).

Perhaps you use Softwarerendering mode (CPU) , not hardware OpenGL mode(GPU)!
how check if sw rendering (CPU!!) or hw (GPU) rendering:
Look in TAB Extensions , there should NOT be an Apple Software Renderer .... Rendering, insted  Intel GMA X3100 Renderer (HW)
I posted my ATI HD3850 Info : Renderer: ATI..... (GPU) and my Bench Vales at 1600x1200/32 Bit/BenchMode
Also in the test this is shown upper left corner Renderer:

If it shows Intel GMA X3100 then all is right, i would set the framebuffer as standard).
X3100 isnt so fast in higer resolution - as yours.

Attached Files

•  Bild_131.jpg   32.17K   56 downloads
•  3850.jpg   89.48K   62 downloads

[Alex DeWolf]

Attached is the system profiler screen shot, ioreg and kext txt files. Thanks for everyone's help.

Alex DeWolf

Attached Files

•  Picture_1.png   77.87K   66 downloads
•  ioregDeWolf.txt   345.16K   13 downloads
•  kextstatDeWolf.txt   8.14K   6 downloads

[Cheops]

Try setting one of your corners in expose to put the display to sleep then re awake it and re-run the tests

Ade.

[Alex DeWolf]

mitch_de, on Sep 12 2008, 11:27 PM, said:

No, OpenGL Extensions Viewer benches are very good and reliable - much better than Xbench, where GMA950 are faster than ATI 2600XT!!!! Xbench OpenGL values are for the trash (rest of valuesa are OK).

Perhaps you use Softwarerendering mode (CPU) , not hardware OpenGL mode(GPU)!
how check if sw rendering (CPU!!) or hw (GPU) rendering:
Look in TAB Extensions , there should NOT be an Apple Software Renderer .... Rendering, insted  Intel GMA X3100 Renderer (HW)
I posted my ATI HD3850 Info : Renderer: ATI..... (GPU) and my Bench Vales at 1600x1200/32 Bit/BenchMode
Also in the test this is shown upper left corner Renderer:

If it shows Intel GMA X3100 then all is right, i would set the framebuffer as standard).
X3100 isnt so fast in higer resolution - as yours.

Well what I was saying is that it seems like rendering on my Asus laptop with the Nvidia 9500M and 512MB VRAM is slower than my Mac Book with the Intel (sorry not ATI) X3100 with shared memory.
Here is a screenshot of the extensions tab.

Thanks
Alex

Attached Files

•  Picture_2.png   52.35K   48 downloads

[Cheops]

See my post above!

Ade.

[Alex DeWolf]

Cheops, on Sep 13 2008, 07:31 AM, said:

Try setting one of your corners in expose to put the display to sleep then re awake it and re-run the tests

Ade.

I did this and no change.

Alex

Attached Files

•  Picture_3.png   54.52K   33 downloads

[Cheops]

Does anything else seem smoother?

Ade.

[Alex DeWolf]

Cheops, on Sep 13 2008, 07:39 AM, said:

Does anything else seem smoother?

Ade.

Nope.

Alex

[Headrush69]

mitch_de, on Sep 13 2008, 02:27 AM, said:

No, OpenGL Extensions Viewer benches are very good and reliable - much better than Xbench, where GMA950 are faster than ATI 2600XT!!!! Xbench OpenGL values are for the trash (rest of valuesa are OK).

They are both useless for any kind of real world benchmarking.

Using these the apps the results between my Geforce 7900GS and Geforce 8800GTX were hardly substantial. (and to average joe meaningless)
But any in game experience with 3D was incredibly dramatic. Using a benchmark tool like Santaduck's UT2004 tool showed the difference.

They are useful for things like noticing when hardware acceleration is or isn't working and maybe a rough level of performance, but other than that, not much.

[Alex DeWolf]

Cheops, on Sep 13 2008, 07:31 AM, said:

Try setting one of your corners in expose to put the display to sleep then re awake it and re-run the tests

Ade.

Well check this out. There is an almost 10 fold speed improvement in the opengl test (1st screenshot is before and 2nd is after) However, HalfLife2 is still choppy.

Alex

Attached Files

•  Picture_3.png   54.52K   30 downloads
•  Picture_4.png   81.76K   28 downloads

[Alex DeWolf]

bump

[mitch_de]

Headrush69, on Sep 13 2008, 05:13 PM, said:

They are both useless for any kind of real world benchmarking.

Using these the apps the results between my Geforce 7900GS and Geforce 8800GTX were hardly substantial. (and to average joe meaningless)
But any in game experience with 3D was incredibly dramatic. Using a benchmark tool like Santaduck's UT2004 tool showed the difference.

They are useful for things like noticing when hardware acceleration is or isn't working and maybe a rough level of performance, but other than that, not much.

Then try it at higer res or (above 2000x1400 - like gamemagazines testresolutions), better aktivate the FSAA (Multisampling) to 4 or 8.
You will see a BIG difference between both cards  !!!
FSAA ist very hard work for GPU and also VRAM-Speed. OpenGL driver differences and CPU-Limits have not more such a big effect aktivating FSAA*4 or FSAA*8.
ALSO:
real word game benches ahve some MINUS:
- much more CPU + RAM dependend - bad comparable between different CPU + RAM systems
- an GPU which wins GOLD on UNREAL may not get any upper place (means loose) in an other game bench (Crysis)
- its too much work to get an GPU value which is only comparable for systems with SAME MB RAM, SAME CPU TYPE, SAME CPU GHZ
conclusion:
real world benches are only great for testing GPUs in ONE system and it is a must to test at least 3 differnet real world games to get an good average bench value.
Game Magazines or HW Mags tests this way - one test pc , 3 games, 10 cards. To much work and not possible for us - complete different CPU+RAM systems, noone has chance to have 3+ realword games with same resolution, same game settings,...
Running OpenGLExtensionsViewer Benches (switching on FSAA at least *2, newer cards *4 or *8) helps here really for "homeuser GPU tests".

[Cheops]

So it is proven that on laptops there are problems getting the nvidia hardware activated. Even though it says hardware accelorated the fact is it lies and isn't true so beware laptop users with nvidia cards we need a fix for this ASAP.

Thanks

Ade

[mitch_de]

Cheops, on Sep 14 2008, 11:42 AM, said:

So it is proven that on laptops there are problems getting the nvidia hardware activated. Even though it says hardware accelorated the fact is it lies and isn't true so beware laptop users with nvidia cards we need a fix for this ASAP.

Thanks

Ade

To check this you can test with OpenGLExtensionvierer the results from "hw opengl" and Apple Software Rendering".
If the two OpenGLExtemsiomsViewer Benchmark values are near the same hw opengl isnt really/not 100% working - even shown in systemprofiler as CI/QE card.
If HW Bench values  are at least by 10* higher(faster) than SW bench values, hw opengl is working, maybe some driver glitches with your mobile gpu.
HOW: Look at my screenshoot , easy switch between those 2 modes (HW OPENGL / Apple Softwarerendering).

My ATI3850 is in Apple Software Rendering Mode > 200 Times (openGl 1.1 -2.0) SLOWER than HW Rendering, At 2.1 20 times faster. Running in my res: 1600x1200/ 32 Bit, C2D 2.4 GHZ. The smaller the resolution or faster the CPU, the less the difference between both modes.
So your Mobile GPUs should be at very least 20 times faster in HW OpenGL Mode in all is OK, or even 10 times faster with some OpenGL driver problems.

Attached Files

•  Bild_138.jpg   47.88K   35 downloads

[Alex DeWolf]

mitch_de, on Sep 14 2008, 03:19 AM, said:

To check this you can test with OpenGLExtensionvierer the results from "hw opengl" and Apple Software Rendering".
If the two OpenGLExtemsiomsViewer Benchmark values are near the same hw opengl isnt really/not 100% working - even shown in systemprofiler as CI/QE card.
If HW Bench values  are at least by 10* higher(faster) than SW bench values, hw opengl is working, maybe some driver glitches with your mobile gpu.
HOW: Look at my screenshoot , easy switch between those 2 modes (HW OPENGL / Apple Softwarerendering).

My ATI3850 is in Apple Software Rendering Mode > 200 Times (openGl 1.1 -2.0) SLOWER than HW Rendering, At 2.1 20 times faster. Running in my res: 1600x1200/ 32 Bit, C2D 2.4 GHZ. The smaller the resolution or faster the CPU, the less the difference between both modes.
So your Mobile GPUs should be at very least 20 times faster in HW OpenGL Mode in all is OK, or even 10 times faster with some OpenGL driver problems.

OK I ran this first using Apple software rendering then Nvidia's open GL engine. Ineed the Apple software rendering is like 100 times slower. What I am wondering about are the numbers for the last 2 tests. Is this why HalfLife2 is so slow?

Alex

Attached Files

•  Picture_5.png   58.42K   22 downloads
•  Picture_6.png   59.07K   24 downloads