5 replies to this topic

[Mark]

I recall looking at improving geometry throughput some years ago, when (as far as I remember) the best solution was to write seperate code for ATI and Nvidia cards, using VAR and Fences or VBO. But I've got this niggling feeling there is now an ARB extension that combines and supercedes both. Could anyone confirm this or point me in the right direction for tackling this problem?

Any pointers much appreciated. Thanks.

[Ed Mack]

They are to be in the next openGL spec as a core feature. However, I'm unsure of the extension layout atm - sorry.

[SigKILL]

Yup, these have merged to the vertex_buffer_object extension (probaly prefixed by ARB_, but not sure: google is your friend.). I think it's been part of the standard since 1.5 or something...

-si

[Mark]

Brilliant. Yeah, it's been that long since I looked into OpenGL :)

Thanks for the quick responses.

[SigKILL]

Just to get my post count up. The vbo extension is very nice, for one thing binds have very low overhead (nothing really happends before glVertexPointer* or something), so you could have one vbo per set of normals, vertices etc.etc. And as everything else in OGL you have a low per-call overhead, so there is not real 'batching' issue with many small meshes like in D3D (I've never heard about an 'optimal' vbo size, I always hear about the 'optimal' size of D3D vertex buffers). There is also some docs describing usage of the extension over at the developer section at www.nvidia.com. And www.opengl.org is the usual reference for anything OGL.

-Si

[yapaa237]

There is also a tutor in devmaster.net :-)



Cheers.