86 replies to this topic

[Korrupted]

That makes more sense now.

[Zulu.Walker]

How about trying this out, also includes a nifty utility:

OpenGL Extensions Viewer

It would make things easier IMHO, and it's not as buggy/picky with GioFX's fullscreen problems. I can run the included benchmarks in all resolutions, also has AA/AF tests and GPU capabilities.

Maybe a better Better OpenGL Benchmarking app?

[consolation]

Zulu.Walker, on Apr 4 2007, 12:21 AM, said:

How about trying this out, also includes a nifty utility:

OpenGL Extensions Viewer

It would make things easier IMHO, and it's not as buggy/picky with GioFX's fullscreen problems. I can run the included benchmarks in all resolutions, also has AA/AF tests and GPU capabilities.

Maybe a better Better OpenGL Benchmarking app?

Multi-Render tests will crash on some hackies using OG Extension viewer, I think it's still affected by screen resolution (display mode selection) + lod/aa settings it would be even more of a pita to standardize. We could all agree to bench @ a lowest common denominator (1024x768) or a common resolution (1280x1024). First option will work on any display, but gives old/low end card a deceptive advantage, second may not be possible on all displays.  Ritalin's normalization method seems to provide results approximate to what you see in the real world.

if you are interested in how these compare to PPC macs

mid-range

high-end

-you will need to register for the SDs site, but it's well worth the hassle.

[consolation]

Actually - OpenMark seems reasonably resistant to resolution change, I benched at different resolutions:

1024x640=  655360pixels 16313
1440x900=  1296000pixels 16040
1280*1024=1310720pixels 16040

Doubling the pixels resulted in only a 1.7% drop; so the benchmarks at different resolutions are comparable.

Attached Files

•  OpenMark12801024.jpg   53.01K   2 downloads
•  OpenMark1440900.jpg   92.12K   1 downloads
•  OpenMark1024640.jpg   52.74K   0 downloads

[Zulu.Walker]

So I guess it doesn't really matter what resolution you bench it on, it just calculates pure GPU processing power. Nice.

I'm liking this bench more. Let's forget about me suggesting OpenGL Extensions Viewer And considering that the 7600GTs returned almost the same results (16040), makes it much more reliable than the other GPU benchmarks I know, even at different resolutions.

[consolation]

Don't get me wrong, I like OGLE Viewer - it allows you a far more detailed look at your card's abilities than OM; especially in separating the abilities of newer and older cards etc etc... It's just that, for a "quick'n'dirty" overview of raw speed, it's a bit of an over-kill, IMHO.

I'd be interested to see the OGLE Viewer results for 7900s with high LOD/AA vs 7600GTs at same settings.  I suspect the high core speed of the 7600GTs lets you draw a lot of relatively simple polygons but, once it has to process them more it should fall off relatively to the slower but more sophisticated high-end cards.  That, or the kext just doesn't take advantage of those cards as it should - I'm speculating here...

P.S. A page I found very useful for card specs is:
http://users.erols.com/chare/video.htm

[Korrupted]

Kind of offtopic, but here's a screen I took of my results with my 7900 GS in the OGLE Viewer.

[consolation]

DaxTsurugi, on Apr 5 2007, 08:45 AM, said:

Kind of offtopic, but here's a screen I took of my results with my 7900 GS in the OGLE Viewer.

Hmm... that's rather poor.

this is my 7600gt



Could you rerun your test with the following settings:

1280X1024X32 (60Hz) Display Mode
Tick multisample set to 4
Untick LOD, set slider to 0
Anistropy set to 16
Tick Use Fog

This the same GT with those settings

Attached Files

•  OGLEV7600gt.tiff   122.01K   2 downloads
•  OGLEV76gt_2.tiff   121.97K   0 downloads

[Korrupted]

[consolation]

That's a lot more like the results you would expect.

[Korrupted]

So the kexts aren't unbalanced then?

[consolation]

DaxTsurugi, on Apr 9 2007, 07:31 PM, said:

So the kexts aren't unbalanced then?

I think they probably use the same kexts, I thought that  maybe they are not taking full advantage of your card. Seem like they are; just that the 7600GT is really good at churning out simple polygons. But the (relative) small number of pixel/vertex shaders shows when you start processing the output more.

[Linux-Kelvin]

This is the Benchmarks of my Hackintosh:

Display Card: Leadtek/Winfast PX7900GS TDH

1280X1024X32 (60Hz) Display Mode
Tick multisample set to 4
Untick LOD, set slider to 0
Anistropy set to 16
Tick Use Fog

[SA22C]

Here's the results for my 7900 GTO 512 MB.

Test settings:
1280X1024X32 (60Hz) Display Mode
Tick multisample set to 4
Untick LOD, set slider to 0
Anistropy set to 16
Tick Use Fog

 OpenGL_Viewer.jpg   201.04K   48 downloads

Interesting that the results are basically the same as the 7900GS in this test, but that my card pulls away quite handily in the OpenMark test, posting a score that is approximately 50% faster.

[u1m2]

x1900xt crossfire edition, 512MB

 OpenMark_result.png   17.87K   11 downloads

[lord xeon]

using the settings established above for OpenGL Extensions Viewer:

Display Mode: 1280 x 1024 x 33bpp (60Hz)
Framebuffer: Standard

untick Multisample set slider to 4
untick LOD Bias set slider to 0
tick Antisotropy set slider to 16
tic Use Fog

my results are:



i have nVidia GeForce 7600 GT with 256 VRAM

[consolation]

lord xeon, on Apr 17 2007, 02:57 PM, said:

using the settings established above for OpenGL Extensions Viewer:

Display Mode: 1280 x 1024 x 33bpp (60Hz)
Framebuffer: Standard

untick Multisample set slider to 4
untick LOD Bias set slider to 0
tick Antisotropy set slider to 16
tic Use Fog

my results are:

i have nVidia GeForce 7600 GT with 256 VRAM

you are supposed to tick Multisample ;-)

[FavleX]

These are my results:

stock conditions

score 294 with Open GL  

fsb=166

score 373 with Open GL

My stupid question:changing the fsb at the boot influences the performance on the entire system as indicated , or the becnhmarks results are not real at all , 'cause the time is not set correctly ? ( I mean the clocktime is going slower)

Attached Files

•  Picture_2.jpg   29.72K   2 downloads
•  Picture1.jpg   28.59K   1 downloads

[lord xeon]

consolation, on Apr 17 2007, 08:11 AM, said:

you are supposed to tick Multisample ;-)

oops, thanks.
here are the real ones i guess:

[consolation]

favalessa, on Apr 18 2007, 03:54 AM, said:

or the becnhmarks results are not real at all , 'cause the time is not set correctly ? ( I mean the clocktime is going slower)

yep that's the one. The bios sets your fsb, not the fsb=xxx switch. That only tells OS X what fsb it's running on. Now, if the real time is faster than the reported time, the benchmark will think the card did more "work" in a shorter period = inflated results.