Korny Klown2
Well, that's a decent start but still noisy and
1. the same scene takes twice as long on my machine 00:48
2. when I tweak it to get it noisefree I immediately run into rendertimes of 3minutes and more.
Your machine is pretty weak. I personally think that the noise is acceptable. I don't know many clients who want to pay for a Cornell Box render.
Some of that noise would be less visible in a production scene where you have textures, different environment and more artistic lighting. I've no problems getting acceptable renders using the methods discussed. It could be faster I agree, which is why I want to make a better scene to test it out. Cornell Box test is to test the contributions of GI. Your MR render doesn't look accurate.
If your clients are pushing for higher quality, you should be charging more to get them the results which includes the cost of rendering. Your Maya render is pretty ugly IMO. I don't know what your acceptable noise tolerance is but, can you post a render of your 3 minute frame? You said you would have to add AO to your MR render, and I agree. How long does that take for you to add a seperate AO render pass? The Mantra render doesn't need it in my opinion.
Korny Klown2
I want a cristal clear image in a decent amount of time and I want control over where I want quality to be focused and where I want to reduce quality.
Try turning off GI if you want/have to render with Mantra then. If machine time is not cheaper for you than artist time. The noise will then come from the Light Sampling quality. Go to town setting up hundreds of lights to mimic the effect of GI. Total control.
The methods discussed, I find have been interchangeable with Mental Ray Unified Sampling, Vray, Arnold, Maxwell, any GPU render. If that's still not to your liking then move on. I love the accuracy, ease and photographic nature of Maxwell, but F**CK does it take forever to render. So I rarely use it unless the situation calls for it. Tools for the job.
Korny Klown2
The problem is, I don't care about a modern renderer because my clients don't care. If I'd work at ILM or DD or so I would probably like Mantra because I get high quality and don't have to care about rendertime. But my clients more often than not wouldn't see the extra quality, all they see on the bill is that it took two days longer to render.
I don't work at ILM or DD, I work in a small shop where sometimes I have to do everything and I like Mantra. Like I said it could be faster but we're at a point where I feel we matched your poor non-AO MR render with a better more accurate (and possibly faster, until you add AO) Mantra render.
The next step is for you to add DOF, Motion Blur and Animation to your test. We can compare set up time and render times then. If single image print is your end goal. I have to agree Mental Ray works wonderfully. Vray better… But you said you do VFX generalist stuff. Maybe if we focus on getting one of your water sims or flame sims optimized that'd be a better case. Cornell Box would be more Arch/Product Viz territory.
Overall, you like Mental Ray it's what you're used to which is understandable. Seems like you have not adopted the unified sampling method Mentral Ray now offers though. If you want to stick with MR, do it.
Use Houdini as a pipeline tool. It's been tailored to send data back to Maya for finalizing. If speed is your concern on a budget, look into GPU rendering. There are coveats there but my gosh it's so worth it when you can use it.