Going to learn PTX and write a GPU compiler

Accelerated Computing CUDA CUDA Programming and Performance
21

not only :D

nvcc --ptx xvinia.cu -o kukuza.ptx

;)

So… We measured memory transfers between RAM and VRAM and was about 2.6GB/s, then we tried to do the same test with OpenGL by loading very large textures, but it seems OpenGL transfers data much faster than cudaMemcpy, is this really so or OGL uses some kind of compression or smth else? Does cudaMemcpy perform any CRC checking?