As I’ve talked with many friends about aliasing/sampling issues, I’ve found that not everyone is on the same page (including my self), and unless you are an advanced user of Mental Ray, you’re not quite too sure where to begin your research. So I wanted to put together a short summary of sampling information that I know about for those of us that often get lost on the threads from the ARC forums. So sampling.
I don’t want to go into adapative sampling because there are tons of threads out there. Just googling it will give you so much information. This link here (http://www.impresszio.hu/szabolcs/MentalRay/MentalRaySampling.htm) is incredibly helpful.
I want go into Unified sampling (which have increasing amount of threads, but because it is still hidden in maya, I figured I’d help spread the word.) To gain access to the unified sampling methods in Maya you need a script (http://forums.cgsociety.org/showthread.php?t=971719). This script creates a gui for the features that are “hidden” way deep in the string options of Maya. Unified Sampling is a newer form of sampling an image. For complex scenes and scenes with motion blur, it is faster than regular Adaptive Aliasing. For simple scenes, it’s not quite there yet. But give it a release or two and it should catch up. The fine folks at Mental Images are doing a great job with refining their product and listening to the consumers, it’s just Autodesk’s lousy implementation that is keeping these features out of the hands of the technically challenged (like myself).
USampling is an Iray technology that is being migrated into Mental Ray. If you combined USampling and Progressive rendering, you have a pretty close representation of the sampling that iRay does. When you expose the controls of Unified Sampling you will have very similar controls that you have with the basic AA technique. You have a min and max samples, but you also have an error cutoff threshold and a quality knob. This sampling technique is linear and not exponential like in AA. So 2 samples really is 2 samples and not 16. The error cutoff is similar to the contrast in AA. But it tells the render below what contrast to stop sampling. So an error cutoff at .02 tells the renderer to not sample if something has a contrast of .02 or less. With each release, this method gets about 10-15% faster than the previous release.
So instead of setting up all of your lights and materials and setting samples of 16 or 32 on each material and light, you leave everything low, like really low. We are talking 1 to 2 samples. Then you let USampling do the rest of the work for you and you just control the quality knob. It’s so much easier to set up and to control. USampling does a lot more under the hood than AA, that’s why you won’t see much of a gain in simple scenes. But for scenes with tons of glossy reflections, I would definitely test this method out, especially if you are using it with the native ibl or the user_ibl shaders. A lot of these new shaders and lights and bsdf models take full control of USampling and is optimized for it better than older materials like the soon-to-be deprecated MIA material.
Apparently, things like Importons and Irradiance particles are built to be optimized for USampling. So in theory, or in short time USampling will help these indirect lighting solutions and you will shortly be able to brute force final gather or IP with good render times and good quality.
So keep your eyes out for further releases and information about this. MR has tons of amazing features that other renderers don’t have. Unfortunately we have to wait on someone that knows how to script to get it into a gui while Autodesk plays catch up.