hello,
i've got performance question regarding Gem an openGL.
i made a test using an OS on a custom liveUSB, so that i can have the same software environment on 3 different computers: ubuntu 10.04, NVIDIA driver v270, pd vanilla 42.5, and Gem svn.
this computer are:
very old laptop: intel core 2 / GeforceGo 7700M
old laptop: intel core 2 duo / Geforce 9650M gt
new laptop: intel core i7 / Geforce GT 425M
i made a test with glsl performance, and everything is normal : a newer computer is many time faster than the older.
BUT, using a very simple patch (like attached), drawing only many time the same square, i can't see any really significant performance difference.
how could that be? what can be the botleneck? (with "lighting 0", i can draw about 400 square before the fps goes down)
but i've got a bigger problem : with "lighting 1", the 2 old laptop can draw about 10% less square (than with lighting 0). but the new one can draw only 3 times less square. So the new laptop is lot's slower than the old one when light is on.
does anyone already experience something like this? any idea why, and how to solve this?
thanks Cyrille
Don't know if that can help, but i get 20 fps when there is less than 1185 squares without light. And it is 1090 fps with light. Configuration : pd 0.42.6, GEM 0.93.SVN rev3578M, NVidia 9700 GTS (2 years old). ++
Jack
Le lundi 24 janvier 2011 à 19:27 +0100, cyrille henry a écrit :
hello,
i've got performance question regarding Gem an openGL.
i made a test using an OS on a custom liveUSB, so that i can have the same software environment on 3 different computers: ubuntu 10.04, NVIDIA driver v270, pd vanilla 42.5, and Gem svn.
this computer are:
very old laptop: intel core 2 / GeforceGo 7700M
old laptop: intel core 2 duo / Geforce 9650M gt
new laptop: intel core i7 / Geforce GT 425M
i made a test with glsl performance, and everything is normal : a newer computer is many time faster than the older.
BUT, using a very simple patch (like attached), drawing only many time the same square, i can't see any really significant performance difference.
how could that be? what can be the botleneck? (with "lighting 0", i can draw about 400 square before the fps goes down)
but i've got a bigger problem : with "lighting 1", the 2 old laptop can draw about 10% less square (than with lighting 0). but the new one can draw only 3 times less square. So the new laptop is lot's slower than the old one when light is on.
does anyone already experience something like this? any idea why, and how to solve this?
thanks Cyrille _______________________________________________ GEM-dev mailing list GEM-dev@iem.at http://lists.puredata.info/listinfo/gem-dev
hello,
i've got answer to this questions.
First : The bottleneck of the test patch i send is the fillrate.
from : http://en.wikipedia.org/wiki/Fillrate "the importance of the fillrate as a measurement of performance has declined as the bottleneck in graphics applications has shifted." So this 3 card have about the same fillrate, even if they are 5 years difference between them. (and jack GPU have a higher fillrate).
about the 2nd question , i found the performance killer. it's the GL_LIGHT_MODEL_TWO_SIDE option. when disable, performance are as expected.
since it is recommended for performance reason to disable this option, it's probable that nvidia has no plan to fix this. So, i think we have to do something in Gem. Changing the default setting is probably not recomended for compatibility reason, even if most of the time diference will be small.
But we have to think of an easy way to disable this.
any idea? a message to gemwin?
Cyrille
Le 24/01/2011 19:27, cyrille henry a écrit :
hello,
i've got performance question regarding Gem an openGL.
i made a test using an OS on a custom liveUSB, so that i can have the same software environment on 3 different computers: ubuntu 10.04, NVIDIA driver v270, pd vanilla 42.5, and Gem svn.
this computer are:
very old laptop: intel core 2 / GeforceGo 7700M
old laptop: intel core 2 duo / Geforce 9650M gt
new laptop: intel core i7 / Geforce GT 425M
i made a test with glsl performance, and everything is normal : a newer computer is many time faster than the older.
BUT, using a very simple patch (like attached), drawing only many time the same square, i can't see any really significant performance difference.
how could that be? what can be the botleneck? (with "lighting 0", i can draw about 400 square before the fps goes down)
but i've got a bigger problem : with "lighting 1", the 2 old laptop can draw about 10% less square (than with lighting 0). but the new one can draw only 3 times less square. So the new laptop is lot's slower than the old one when light is on.
does anyone already experience something like this? any idea why, and how to solve this?
thanks Cyrille
GEM-dev mailing list GEM-dev@iem.at http://lists.puredata.info/listinfo/gem-dev
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-01-25 16:32, cyrille henry wrote:
But we have to think of an easy way to disable this.
any idea? a message to gemwin?
make [gemwin] an abstraction, where all the openGL stuff is exposed on the patcher level. allow users to replace the shipped [gemwin] abstraction to remove code they don't think they will ever need.
fgasdr IOhannes
Le 25/01/2011 16:58, IOhannes m zmoelnig a écrit :
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-01-25 16:32, cyrille henry wrote:
But we have to think of an easy way to disable this.
any idea? a message to gemwin?
make [gemwin] an abstraction, where all the openGL stuff is exposed on the patcher level.
this is part of the multiple windows plan. is his going to append in a near future?
cyrille
allow users to replace the shipped [gemwin] abstraction to remove code they don't think they will ever need.
fgasdr IOhannes -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.10 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk0+8woACgkQkX2Xpv6ydvSqhACeJLEJW6iPCzJLB6yUqlZBF9rp Fu0AniJpOIhjlUYOunK1jOrZWJo/sG9T =6RS6 -----END PGP SIGNATURE-----
GEM-dev mailing list GEM-dev@iem.at http://lists.puredata.info/listinfo/gem-dev
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-01-25 16:32, cyrille henry wrote:
hello,
about the 2nd question , i found the performance killer. it's the GL_LIGHT_MODEL_TWO_SIDE option. when disable, performance are as expected.
actually you should be able to simply do something like [GEMglLightModeli GL_LIGHT_MODEL_TWO_SIDE GL_FALSE] at the beginning of your render-graph, to explicitely turn the performance killer off. it's a workaround though.
fgasdmr IOhannes
Le 24/02/2011 09:20, IOhannes m zmoelnig a écrit :
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-01-25 16:32, cyrille henry wrote:
hello,
about the 2nd question , i found the performance killer. it's the GL_LIGHT_MODEL_TWO_SIDE option. when disable, performance are as expected.
actually you should be able to simply do something like [GEMglLightModeli GL_LIGHT_MODEL_TWO_SIDE GL_FALSE] at the beginning of your render-graph, to explicitely turn the performance killer off. it's a workaround though.
that's what i am doing. it's working ok, except that I have to change some light in a few patch... (and to replace all gemwin with an abstraction)
thanks c
fgasdmr IOhannes -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.10 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk1mFN0ACgkQkX2Xpv6ydvRLgACcDPemqT1vuoBWO4NegVTx3ASN jOwAnjKm/hOOqNfdDoJncudRUvvN4fN5 =LuHI -----END PGP SIGNATURE-----
GEM-dev mailing list GEM-dev@iem.at http://lists.puredata.info/listinfo/gem-dev
-
cyrille henry -
IOhannes m zmoelnig -
Jack