vilda(Blenderstorm admin) wrote on the 24 May 08 at 10:34
I'm not sure about this because of 1 reason:
Recoding so many features into gpu acceleration is a big task, and It isn't sure it pays back. The future processors might pull graphics computing back to processors, and it isn't sure in how many years that happens(maybe soon).
It's maybe hard because compatibility problems. Different cards have different supported features and different APIs (NVidia / ATI). OpenGL and DirectX have managed it well.
But it worth. I belive most of you here already own a good graphic card.
I have a decent card (GeForce 7100), so I give a +1 vote.
Future computers will never match the amount of cores that GPU's have.
We are not asking to remove the CPU renderer, keep it in.
Just add code to check if a supported GPU is in the system and if it finds one or more, use it.
Current GPU's have limited floating point accuracy compared to CPU's and may thus produce less accurate results. Will this inaccuracy be visually noticeable? Not too important if all you want is a FAST render of an animation so that you can check/continue working with minimal interruption. You can always use the CPU for the final render if it turns out to produce better results.
Ray-tracing is perfect for parallel processing in the GPU because each ray can be handled by a unique thread.
I am currently getting to know CUDA and see great potential.