glTexImage2D PBO Uploads take a lot of CPU time?

OpenGL OpenGL: Advanced Coding
1

I’m doing some GPGPU computing using Opengl and a Nvidia Quadro 2000M card.

While profiling I noticed that my glTexImage2D calls for texture uploads from PBOs are taking 20% of the cpu time, as much as the time spent copying over all the data to the PBOs.

glTexImage2D(GL_TEXTURE_2D, 0, 1, width_, height_, 0, GL_RED, GL_UNSIGNED_BYTE, NULL);

Where

glPixelStorei(GL_UNPACK_ALIGNMENT, 8);
glPixelStorei(GL_PACK_ALIGNMENT, 8);

I don’t understand why this is taking so much cpu time? Shouldn’t it just start the asynchronous DMA transfer to the GPU?

2

Have you tried with other alignments than eight bytes?

3

If you know texture size before you upload, then better create texture with glTexImage2D, then use PBO + glTexSubImage2d combo to update texture data. Another issue can be GL_RED. I think it is not accelerated (suitable for GPU) format, and driver must perform on-the-fly conversion to something useful for GPU. Check this http://developer.nvidia.com/content/nvidia-opengl-texture-formats