> Can someone clarify this for me? Are OpenGL/D3D buffers that get stuff memcpy'd into them by the CPU actually "slices of GPU memory," or are they more often reserved driver memory that eventually get DMA'd to the GPU?

The answer is, as with all things OpenGL, it depends. You might get back a pointer to GPU memory that you can directly write to, or you'll get back some chunk of system memory the driver has.

The ARB_buffer_storage extension improves matters as you can almost guarantee that you'll get GPU memory, and you can keep it mapped for the entire lifetime of your application (the old buffer APIs wouldn't let you keep it mapped during a draw call). The downside is that you're now responsible for synchronising access to that data.

But as for "is it quicker?", maybe. DMA transfers aren't free, they take time to setup. Usually they need to operate from a limited pool of source memory. If the driver has to take a local copy of your data to copy it (which it will do for every glBufferData/SubData call), then you might as well copy it yourself, GPUs aren't hurting for PCIe bandwidth these days. In addition, you can use a separate thread/CPU core to do the copy, since unlike every other OpenGL call, mapping memory and memcpy'ing doesn't require an OpenGL context.