cudaGLMapBufferObject on 2.3 Is still Slow?


im modifing the simpleGL example that comes in CUDA SDK,

i’ve read other forums that “cudaGLMapBufferObject” and “cudaGLUnMapBufferObject” are really big bottlenecks, but those posts where from months ago, so my question is if there has been any improvement over it.

and another question: is it necessary to do cudaGLMapBufferObject every time i call the kernel?? why update the pointer position every time if it should be pointing to the same Vertex Buffer Object (in my case) just with modified coordinates?



Yes, they’re still 30-200us depending… which means if need to call them hundreds/thousands of times a second… well you’re not going to get ideal performance (better off rendering inside of cuda, not opengl - if the rendering task is simple).

my kernel is a function that deforms 3d meshes when the user chooses to, so i should call the Mapping functions only when the kernel is called, right? that way i dont loose too much rendering performance.