|
From: Sauro A. <int...@in...> - 2007-09-28 16:58:06
|
Using the latest Quesa we found that it is very slow than the old 1.8
version and previous versions, especially with polygons and general
polygons.
The problems is evident with our programs, where using the latest
version of Quesa, navigating is very slow using any renderer.
We tested different versions of Quesa on different computers using
Geom Test with the following results:
Test MultiBox
Quesa Logo MultiBox Quesa Logo
Interactive
Interactive OpenGl OpenGl
FPS
FPS FPS FPS
Last Quesa on iMac Intel Core duo 1.83 GHz 12
1030 11 2500
Last Quesa on G5 Dual 1.8 GHz 9
630 11 1500
Quesa 1.8 on iMac Intel Core duo 1.83 GHz 36
970 ----- -----
Quesa 1.8 on G5 Dual 1.8 GHz 26
720 ----- -----
Old Geom Test 2003 on iMac and Rosetta 18
500 ----- -----
Old Geom Test 2003 on G5 Dual 1.8 GHz 28
730 ----- -----
Practically the latest version is faster using OpenGl and Quesa Logo,
but it is very slow with the MultiBox model, independently of the
kind of renderer, where last Quesa for Intel is slower than Quesa 1.8
emulated with Rosetta!
Also with our programs we have similar results, with the latest
version that is 3 times slower than Quesa 1.8 and previous versions
(12 FPS versus 36 FPS).
Any suggestions?
Sauro Agostini
|
|
From: James W. <ja...@fr...> - 2007-09-28 17:35:44
|
Sauro Agostini wrote: > Using the latest Quesa we found that it is very slow than the old 1.8 > version and previous versions, especially with polygons and general > polygons. > > The problems is evident with our programs, where using the latest > version of Quesa, navigating is very slow using any renderer. > > We tested different versions of Quesa on different computers using > Geom Test with the following results: > > > Test MultiBox > Quesa Logo MultiBox Quesa Logo > Interactive > Interactive OpenGl OpenGl > FPS > FPS FPS FPS > > Last Quesa on iMac Intel Core duo 1.83 GHz 12 > 1030 11 2500 > Last Quesa on G5 Dual 1.8 GHz 9 > 630 11 1500 > > Quesa 1.8 on iMac Intel Core duo 1.83 GHz 36 > 970 ----- ----- > Quesa 1.8 on G5 Dual 1.8 GHz 26 > 720 ----- ----- > > Old Geom Test 2003 on iMac and Rosetta 18 > 500 ----- ----- > Old Geom Test 2003 on G5 Dual 1.8 GHz 28 > 730 ----- ----- > > > Practically the latest version is faster using OpenGl and Quesa Logo, > but it is very slow with the MultiBox model, independently of the > kind of renderer, where last Quesa for Intel is slower than Quesa 1.8 > emulated with Rosetta! > > Also with our programs we have similar results, with the latest > version that is 3 times slower than Quesa 1.8 and previous versions > (12 FPS versus 36 FPS). > > Any suggestions? Are you sure you're comparing builds made with the same compiler and optimization options? I am a little skeptical, because outside the OpenGL renderer, I think most recent changes have been intended to improve performance. Besides my previous suggestion to try Shark, you could use cvs to try to narrow down the problem. That is, check out the sources as of various dates between 1.8 and now, to try to discover when a slowdown happened. -- James W. Walker, Innoventive Software LLC <http://www.frameforge3d.com/> |
|
From: Sauro A. <sag...@in...> - 2007-09-28 18:24:27
|
> > > >> Practically the latest version is faster using OpenGl and Quesa Logo, >> but it is very slow with the MultiBox model, independently of the >> kind of renderer, where last Quesa for Intel is slower than Quesa 1.8 >> emulated with Rosetta! >> >> Also with our programs we have similar results, with the latest >> version that is 3 times slower than Quesa 1.8 and previous versions >> (12 FPS versus 36 FPS). >> >> Any suggestions? > >Are you sure you're comparing builds made with the same compiler and >optimization options? I am a little skeptical, because outside the >OpenGL renderer, I think most recent changes have been intended to >improve performance. I compiled the release version just with the options of the downloaded project. Looking to the Quesa Logo FPS there is a big improvement, from 1500 to 2500, while multibox goes down from 36 to 12, so it doesn't seem depend on some optimization option, the difference is too big. > >Besides my previous suggestion to try Shark, you could use cvs to try to >narrow down the problem. That is, check out the sources as of various >dates between 1.8 and now, to try to discover when a slowdown happened. >-- I'll try to test with Shark. Sauro |
|
From: James W. W. <os...@jw...> - 2007-10-02 04:10:32
|
Progress report: I've identified a reason that the multi-box test got slower. It seems to be a result of calling glBindFrameBufferEXT on each call to GLDrawContext_SetCurrent. I have a fix, but I want to do a bit more testing before committing. |
|
From: James W. <ja...@fr...> - 2007-10-02 21:36:43
|
I've committed a fix for the multi-box slowdown. It may still be a few percent slower than 1.8, and I'm not sure why, maybe virtual method dispatch. Bear in mind that the multi-box test is not a very realistic benchmark. It has 6000 TriMeshes, each of which has its own attribute set but only 2 triangles. This is an exceptionally high amount of overhead for the amount of geometry. -- James W. Walker, Innoventive Software LLC <http://www.frameforge3d.com/> |
|
From: Sauro A. <sag...@in...> - 2007-10-03 11:14:03
|
I used multi-box because the results are similar to the models in our programs. We have mainly architectural models, with thousands of polygons and general polygons. I posted a 3DMF file of a small house at the following address http://www.interstudio.net/quesa/Farmhouse1.3DMF.zip Loading this model in Geom Text I have the following results (Dual 1.8 GHzG5): Previous 1.8 Quesa and Interactive renderer 2.9 - 3.2 FPS Last Quesa with Interactive or OpenGl renderer 8.1 - 8.4 FPS Other results with other models of Geom Test Test Quesa 1.8 Last Quesa Interactive OpenGl Box 2000 1500 MultiBox 25 8 Cone 60 280 General Polygon 2220 1900 Polygon 2200 1900 Trymesh 2200 2000 TriGrid 1500 1400 Torus 1400 1900 Quesa Logo 630 1200 Cone, Quesa Logo and Torus are better with the new version. Box and MultiBox are better with the old version. The others practically are the same. Sauro Agostini -- ----------------------------------------------------------------------------- _ ___ |_| __| Interstudio S.r.l. Tel + 39 0573 99291 Fax + 39 0573 992930 | |__ | Piazza Monteoliveto 6a http://www.interstudio.net |_____| I-51100 Pistoia Italy mailto:int...@in... ----------------------------------------------------------------------------- |