This was just mentioned in the hardware section earlier, and I had previously thought about it too.
If you get a really fast machine for renderings, but it’s going to suck that the final renderings on the CPU will probably be a lot slower.
Are there plans (or is it intended) that the RT portion will be used to render final images? Or, is there someway that a GPU can be added to help process the finals?
When I saw your post heading I thought “What a stupid question, we all know the answer already. I thought jujubee was one of the main guys here?”
but after reading what is in your thread, it’s actually a VERY GOOD question to which I would also like to know the answer
I imagine at least the lightcache part of the calculation could be done by RT.
The lightcache takes quite some time with my scenes, about 10 min. for final renderings, and it can’t be distributed, so that would be cool.
From what I do understand, Vray RT is brute Froce rendering (DMC/QMC)
I do not know if RT GPU has the same approch, but I would guess that the only way to re-use that in production rendering would be to select a Brute Force GI method.
My initial reaction would be no. Right now there are many, many features not yet supported by RT. But if your final render could be completely done in RT, why would you not just use RT if it is faster for you?
I would think that things get a little difficult when sending information back and forth between the CPU/GPU systems and how efficient that transfer process was could be a significant barrier.
Definitely going to follow this thread, should be some interesting discussions.
The question I have about all of this CPU/GPU talk is what is the ultimate goal people are looking for? Faster render times? More flexibility for making adjustments quickly?
Just curious. 
I’m thinking about twice of that solutions… Fast render and confortable viewport tweaking… But i’ve also a question, about proxys… My scene work with a tone of them and multiscatter for exemple… And how vray gpu work with that? does it work with proxys and displacement? Could we hope to render 5300px X 5300 on a GPU ? The 1 or 2go vram in the last cg card are so little in comparison of my 18 or 24 go ram in a workstation to render all picture i want…
What is the GPU state in my pipeline if it couldn’t render everything i want?
That’s a really great question. As many of the posts have pointed out “but i use Xxx amount of proxy’s, plugins etc etc..” I think RT can only go so far as a production renderer. While the speed is great, I’m not sure the artistic limitations are worth the speed increase. Those simplistic form studies and very simple massing type renders will be a joy to do in RTGPU, but I just wonder how much time will be spent in the future worrying about things that RT GPU won’t be able to do rather than can do.
Perhaps that Hybrid mode will develop more which will bridge the gap between the features unsupported by the GPU. I’m wondering, does iray support everything that full blown mental ray does?
Nope.
Best regards,
Vlado
Ideally, I’d imagine that the hopes would be that everything could be done in real-time eventually, without the need for pressing the “render” button as we currently do. I think that’s obvious - but the question is whether or not we’ll encounter an acceptable solution relatively soon (by the time the official release comes out.) Convergence. It seems that a lot more has been implemented with RT GPU than plain RT. Now it comes down to working out/circumventing the technical limitations. I don’t think all this work would be invested if Chaos didn’t think RT was the future. I have a feeling my question is getting avoided but maybe it’s pretty obvious where this is headed lol. 
I would be really happy just to see even CPU RT able to act fully as a preview render and can live without it entirely as a production engine. For my type of workflow just being to see everything Vray can do at interactive speed (albeit lo-res) would be more than enough time saving to make RT worth it several times over.
I’ll be interested if they can get rid of the noise.
I’ve actually had clients like a little noise in their images… go figure.
That’s cool. I’m very critical of noise, maybe to much so. But I bet they would not want not an RT amount of noise in an image produced in a reasonable amount of time. There are almost no examples that look clean even with the latest hardware. I hope to see some examples of noise free images produced in a reasonable amount of time using the latest and greatest hardware. This will require a substantial (by my terms) investment by people willing to show results and describe their setup.
What do you consider a reasonable amount of time?
I have recently decided that my definition of a quick render time is NOT the same as others.
Something measured in minutes as opposed to hours. I currently get production quality, noise free stills in minutes using a small farm that’s getting on in age.
Currently it looks like RT is more of a previewing tool than a replacement of the vray production renderer.
Great to be able to preview lighting and materials interactively!
It will be interesting to see how it progresses as the hardware gets faster.
GPU looks interesting, but for the start I’ll still be investing in CPU/render power primarily. Would be nice to get GTX480 or similar in the near future though
This is a nice example of a high-res RT image on a GTX 295 http://www.chaosgroup.com/forums/vbulletin/showthread.php?51270-good-news-and-and-a-quot-3-hours-test-quot
I saw that. But three hours??? We might as well use Maxwell.
Yeah, it is a long time, and scene doesn’t look massive, but shadows and refractions DOF must push the rendertimes up a lot, and it looks good.
I’m guessing a similar scene and settings might take longer to get to the same resolution and smoothness in Maxwell. Maybe days… A couple of tests i did a while ago with it took ages.
The thing that really interests me with RT is being able to tweak material and lighting in realtime. It looks like in most cases it won’t be the best option for production/final renders, but it offers a more immediate workflow for setting up and fine tuning materials and lighting (the magic ingredients;).
I’m looking forward to being able to work in in the RT viewport! maybe one day;)
We’ll see how the technology develops over time, as the hardware speeds up.
I would be interested to see how that scene refined over time. Different people might have different cut-off points where they’d be happy to present it to clients. For most projects my final rendertimes are 30min-3hours, for 2000-3000 pixel width images, and before that preview/test renders are around the range of 5-15min, but i seen a lot of people have a lot higher settings/rendertimes for finals. In a way it would be great to have the option to just leave render to go as far as it can, rather than trying to calculate settings around a particular rendertime required then rendering.