We’ve had the hunch for a couple of weeks now that the Blender Foundation would have an NVIDIA RTX-related announcement at SIGGRAPH, and lo and behold, it’s just happened. For those equipped with an RTX card and a code compiler, you could jump into testing today.
In a new blog post, Blender’s Patrick Mours says that RTX acceleration is first coming to Cycles, and it will include both ray tracing acceleration via the built-in RT cores, as well as AI denoising down-the-road with the help of the Tensors and their strong interference performance.
NVIDIA’s OptiX engine is being used for the ray tracing acceleration, as are most commercial applications adding RTX support in. Inside the Cycles rendering devices panel, a new OptiX option will appear next to CUDA and OpenCL, and when used, the ray tracing acceleration can be tapped into. While it might be possible to do heterogeneous rendering with RTX, it doesn’t appear to exist right now – not that it particularly matters. In our experience, the GPU is much more important than the CPU in most renders, and with the promised performance, we might be able to forget the CPU even exists. Check this out:
In well-known benchmark tests, “RTX On” makes an incredible improvement to full render times. Now, with RTX being involved, and special acceleration of any kind, we obviously have to concern ourselves with whether or not the final render actually looks identical, but ultimately, so much of the Cycles codepath is shared between all device types, so chances are that will in fact be the case. And… even if it’s slightly different, it may not matter more than the time saved.
We’ll definitely be taking Blender’s RTX support for a test drive as soon as we can. We’re still wrapping up some testing with the release candidates, and hope to deliver performance not long after the 2.80 final drops. When RTX support is built-in and is easy to use, without compiling code or setting up special environments first, we’ll punish our GPUs more.