Best OpenGL Optimize method?
Author
Ok now that I''m getting into OpenGL bigtime- I have a slew of questions, this one is about optimization.
Assuming I have a model that is static, what method is best for optimization of drawing that model for speed? Should I use vertex arrays? DrawElement? or Display lists?
I have heard display lists offer no speed increase (another post), so what is their point then? reduce code overhead?
Ranger
display lists are to reduce code overhead, yes. if you have multiple instances of the same object you can just call the same display list to draw them all.
vertex arrays with glDrawElements are faster than glBegin/glVertex*/glEnd i think mostly b/c of function overhead.
what i''m trying to learn currently is how to cache an objects displaylist/geometry on the vid card. but my sense is that that is an implementation decision left to the card itself. i''m not sure.
-me
vertex arrays with glDrawElements are faster than glBegin/glVertex*/glEnd i think mostly b/c of function overhead.
what i''m trying to learn currently is how to cache an objects displaylist/geometry on the vid card. but my sense is that that is an implementation decision left to the card itself. i''m not sure.
-me
quote:Original post by PalidineThat''s what NVIDIA''s vertex array range extension does.
what i''m trying to learn currently is how to cache an objects displaylist/geometry on the vid card. but my sense is that that is an implementation decision left to the card itself. i''m not sure.
interesting.
that''s not the same as glDrawRangeElements() is it?
my understanding currently is that the above function is just a restriced version of glDrawElements()?
note to self: STFW
-me
that''s not the same as glDrawRangeElements() is it?
my understanding currently is that the above function is just a restriced version of glDrawElements()?
note to self: STFW
-me
Newer cards have vertex arrays, and if it''s a static object, you can use LOCKED vertex arrays.. these are ussually the fastest. I don''t know who told you that Display Lists offer no speed increase, but they do/can. OpenGL (depending on the implementation) does attempt to optomize your display list (minimize state changes, re-order the drawing functions, and it can store your vertices internally as static vertex arrays, etc, etc).
Anyways, to answer your question, I''d have to go with vertex arrays/locked vertex arrays (per card implementation would be best, or support certain cards, and fall back to normal if unkown card)... although, the smaller your vertex arrays are (low poly models), the less of a performance gain you''ll get.
Billy - BillyB@mrsnj.com
(Hey, I''''m no longer anonymous!)
Anyways, to answer your question, I''d have to go with vertex arrays/locked vertex arrays (per card implementation would be best, or support certain cards, and fall back to normal if unkown card)... although, the smaller your vertex arrays are (low poly models), the less of a performance gain you''ll get.
Billy - BillyB@mrsnj.com
(Hey, I''''m no longer anonymous!)
when you are talking about card vertex arrays you mean just standard openGL vertex arrays, right?
as in the card just recognizes when you are using them and caches the vertex info?
-me
as in the card just recognizes when you are using them and caches the vertex info?
-me
Wow, Billy, you registered. 
In regards to locked vertex arrays, generally, you don't see much of an improvement in speed unless you're using some form of multipass rendering. At least, that's been the case in my own code, and I've heard the same from others.
Palidine: the vertex array range extension isn't the same thing as glDrawRangeElements. The former allows you to allocate the memory you use for vertex arrays in video memory; this is the fastest method available on NVIDIA hardware. The latter isn't exactly a restricted version of glDrawElements; rather, it allows you to specify the minimum and maximum indicies used in your vertex array, which allows the video driver to more efficiently cache vertex data.
In regards to locked vertex arrays, generally, you don't see much of an improvement in speed unless you're using some form of multipass rendering. At least, that's been the case in my own code, and I've heard the same from others.
Palidine: the vertex array range extension isn't the same thing as glDrawRangeElements. The former allows you to allocate the memory you use for vertex arrays in video memory; this is the fastest method available on NVIDIA hardware. The latter isn't exactly a restricted version of glDrawElements; rather, it allows you to specify the minimum and maximum indicies used in your vertex array, which allows the video driver to more efficiently cache vertex data.
I would stay away from vertex array locking (CVAs). The extension is very ''experimental'', in the sense that it was never finalized. It''s not even a real standard extension, it was more or less a quick''n''dirty function they developed to make Carmack happy. In fact, early nVidia driver versions did only support one single array format in this mode: the one Quake used...
The only thing that CVAs brought me were problems. Their implementation is so extremely driver/manufacturer dependend, you can''t rely on their behaviour in any way.
If you want to be compatible: use standard vertex arrays. They offer good speed, and most modern drivers do a really good job accelerating data transfer through AGP streaming etc. If you want to go for maximum performance, and multi-manufacturer compatibility isn''t really an issue, then go with VAR/VAO. It will allow the highest possible transfer rates that your hardware can handle.
About Draw(Range)Elements: AFAIK, nVidia drivers do not make a difference between both, DrawRangeElements directly calls DrawElements. The speed difference is reportedly 0, and I can confirm that, at least in VAR mode. But there might be a difference with other manufacturers, so it''s probably safer to always use DrawRangeElements.
/ Yann
The only thing that CVAs brought me were problems. Their implementation is so extremely driver/manufacturer dependend, you can''t rely on their behaviour in any way.
If you want to be compatible: use standard vertex arrays. They offer good speed, and most modern drivers do a really good job accelerating data transfer through AGP streaming etc. If you want to go for maximum performance, and multi-manufacturer compatibility isn''t really an issue, then go with VAR/VAO. It will allow the highest possible transfer rates that your hardware can handle.
About Draw(Range)Elements: AFAIK, nVidia drivers do not make a difference between both, DrawRangeElements directly calls DrawElements. The speed difference is reportedly 0, and I can confirm that, at least in VAR mode. But there might be a difference with other manufacturers, so it''s probably safer to always use DrawRangeElements.
/ Yann
i agree with Yann here, ive never liked cva''s + personally can find no difference speedwise with drawrangeelements vs normal drawelements
http://uk.geocities.com/sloppyturds/gotterdammerung.html
http://uk.geocities.com/sloppyturds/gotterdammerung.html
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement