I posted a demo of the new Arauna2 renderer on ompf2.com: http://ompf2.com/viewtopic.php?f=5&t=1887
A youtube of the demo can be found here: http://www.youtube.com/watch?v=Znr1JJLI5uY
A higher quality version of the video can be downloaded as well (warning: large; 850Mb): https://mega.co.nz/#!pgJQUJ4Z!XaJIW0B_BMha39FIk8TsOk2CV2RTw5wRPEpDbrVeNBk
And finally, the demo, which requires a decent CUDA device and 64-bit windows: https://mega.co.nz/#!UphmTTxb!JAfhjCfDa3-zDo3Zy5gD9EArYxlkJKa10XPqZY8sZRs A hires patch for the demo can be found here: https://mega.co.nz/#!ZwBjWChQ!cgoSi1I-7gLKkBooUHFgmQWClwMocl-qqQ8YeiR2vx8
Arauna2 is an interactive path tracer, which produces images using a random process. As a result, images contain noise, which will fade away as the pixel values approach the 'expected value'. Arauna2 has been optimized to produce decent quality images very quickly, and converges to high quality images in seconds. It is intended for architectural walkthroughs, but something involving games would be even more awesome. As you can see in the demo, we are not there yet, sadly.
I believe Arauna2 is currently the fastest path tracer, but that is of course strongly dependent on requirements. Main limitation is the fixed shading path, which supports the full Phong model, specular, dielectrics, textures, normal mapping and emissive surfaces, as well as every sensible combination of those. Illumination is provided using light emitting surfaces, meshes, point lights, spotlights (including IES profile support) and directional lights.
This is an awesome demo! Thanks for sharing. Any plans to support OpenCL? Does convergence happen much quicker on higher-end devices (or using 4-way SLI devices for example)?
Yeah, OpenCL (or something else that runs on all GPUs) is planned, but will take a while. A CPU tracer is done already, but that's obviously 'rather slow'. Convergence depends on compute power, but path tracing does not scale linearly: to halve variance, you need to quadruple the amount of samples. Two Titans are not that much better than one, and since the tracer will produce images by rendering a full frame on each card, with different random seeds, multiple cards do not improve frame rate.