NVIDIA has published a presentation about new techniques that allow to decrease CPU L2 cache misses due to vertex array pointers dereferences and object bindings.
The GL_NV_vertex_buffer_unified_memory and GL_NV_shader_buffer_load extensions allows to use GPU addresses directly to specify vertex attrib, element array or buffer object. Measurements have shown that these techniques can result in more than 7x speedup!
4 thoughts on “Improve OpenGL Performance By Using GPU Addresses”
sounds both nice and terrible because that means again different code paths among different vendors 🙁
Yes its sucks for different code paths but shows the true potential of OpenGL when someone uses it correctly… I wish ATI would get their crap together but for now I only buy Nvidia GPUs due to their support of OpenGL is unmatched…
It’s called re-inventing the wheel
well to be fair with AMD/ATI, NVIDIA’s GLSL lack of conformance is unmatched as well… which in the end causes havoc because coders blame ATI when it’s NVIDIA’s GLSL compiler’s fault which accepts a mixture of GLSL and Cg
Comments are closed.