I am just trying to prepare a simple framework (i.e. base-point) for further performance analysis of various shader code, GL settings, draw methods, etc.
Originally Posted by Alfonse Reinheart
Yes. Small triangles will be vertex-limited. But large triangles will be fragment-limited. One triangle has 3 vertices, but up to 1 million fragments, each of which takes separate attention from the gpu. Try making tiny triangles, and see if the speed shoots up.
Originally Posted by red1939
Your application will be limited by vertex processing if the overhead of vertex processing exceeds any other form of processing done by the GL or the application. Saying "small triangles will be vertex-limited" is kind of nonsense. And even if your triangles are small, the likelihood of becoming limited by vertex processing for a few hundred triangles is still very low unless your application is very, very trivial and your fragment shaders do absolutely nothing than export a constant color. You cannot be sure of anything unless you get hard numbers, especially when stuff seems to be trivial.
Originally Posted by Roaoul
Why up to 1 million? What if the triangle is large enough that it simply cover the whole screen after clipping? Does a full-HD fragment buffer only consist of 1 mio pixels? No.
Originally Posted by Raoul
Oh. Newbie question: What is the maximum number of pixels a triangle could put to a "full-HD fragment buffer"? -thanks
Originally Posted by thokra
A full-HD buffer usually has a resolution of 1920 x 1080 pixels. That's a little higher than a million.
Are you using 1000 draw calls?
Nope. 1000 x 3 triangles via one DrawArrays.