See mental ray render a million mechs on the GPU
Nvidia has posted an interesting video of the new GPU-based progressive ambient occlusion system being developed for future releases of mental ray.
The renderer’s previous implementation of AO on the GPU used a sampling mode optimised for batch rendering: comparatively fast, but prone to odd artefacts when used for an interactive preview of the output.
The new progressive sampling scheme isn’t as fast to converge, but removes the artefacts.
Better antialiasing – and support for instancing
Nvidia has also improved anti-aliasing quality, and added support for instancing, making it possible to render scenes with much heavier geometry without having to switch back to the CPU.
The obligatory instancing demo video is pretty crazy, showing a 21-million-triangle mech model being instanced a million times and remaining interactive, albeit on a machine with a top-of-the range Quadro K6000 GPU.
So far, there’s no indication of which commercial build of mental ray the work will appear in.