Vray, lets get progressive.

[Scott Schroeder]

I've been fiddling around with progressive rendering in Vray 3 and I'm pretty impressed by it. Especially since I decided to test it out on a project that is on a severe deadline.

 

The positives I've seen so far are.

Super easy set up. No more fiddling with individual subdiv values. Keep it at 8 and let the render refine.

Super easy render settings. No more guesswork. Again, just let the thing render until you are happy.

Predictable render times if you use the render time limit. I can only give 2 hours to the render, set it to render for 2 hours and stop after that. On the big render slave, it probably doesn't need the full 2hours but I'm getting much more detail and sharpness in those 2 hours versus 2 hours using the old IP+LC way.

The min shading rate really excels at this one.

 

The potential cons I can see are

How do you deal with rendering in a studio setting with variable computer speeds? A 2 hour time limit on the big render box will yield different results than 2 hours on the slower user box. I would guess using the max subdivs would be the way to go as the limiter.

Breaking your existing habits of individual subdiv rates and/or adopting this on older scenes. Thank you Vray Jesus for the VMC script.

The noise threshold limit doesn't seem to work as I had expected it to when using that as a limiting factor to stop a render submitted via network.

 

Thoughts, discussions?

[braddewald]

I'm also really impressed with this feature's potential. I'm using it a lot for test renders. I can grab a cup of coffee, come back to my desk and I've got a pretty good idea of what the final render is going to look like -- plus it supports render elements. Coupled with the new "use local host" option, I can uncheck that box and let the slaves do all the work. My workstation remains quiet and I can keep working with email, Photoshop etc.

 

I've done comparisons on the same scene between it and what's now called "adaptive" and it seems like it's really close in terms of quality but there are some big differences in the way it handles edges -- even with AA turned off in both cases. Better in some areas, worse in others, it's really hard to put my finger on what's going on there.

 

When you say you're getting better quality than the old IM+LC way, do you mean you're using BF+BF with the progressive sampler?

[Scott Schroeder]

I'm using BF+LC when doing progressive. Brute force is set to 8 and LC is set to 1000 at .02 screen size. Which makes sense as to why I'm getting much more detail out of it since I'm not using the imap. But even doing brute force + LC the old way still I can get better results in the same time frame using progressive using the min shading rate set to around 4-8 depending on the scene.

 

I'm pretty hopeful about adopting progressive for most of our rendering situations, but I do need to find that sweet spot of max subdivs as a time limiting factor across the user boxes we render on at night. Or maybe it's just better to hook up 9 machines to the Xeon render slave and let them go at each still render for about an hour. That should provide enough speed to get really clean solutions in that time.

[braddewald]

Why put a max subdivision in place (rather than leaving it at 100) if you're putting a time limit in place anyway?

[Scott Schroeder]

I realize I wasn't clear in the description. In lieu of a time limit, the max subdiv is probably the better way to go across our farm of varying speed machines. Maybe a case to use both would be if you want to put in a safety catch so you just in case don't get a 20 hour render trying to get the max subdiv.

[braddewald]

Ah, that makes sense.

[RyderSK]

I think it's also quite excellent :- )

 

Every time I used it, I did so with time limit. I sort of evaluate that time by hand though based on previous tests.

 

It's worth noting that Progressive sampler (but even adaptive Universal solution) works well with IM+LC too, not only BF+LC.

Of course, the BF speedup in Vray3 makes BF quite convenient staple but sometimes IM simply gives that 'clear' look faster and doesn't provide any stuck noise. It only needs bit lower adaptivity in DMC (ie. not 0.9-1 but 0.8-0.85 for example) so it won't waste all the IR subdivs and also global multiplier needs to be at 1 and not 0 (for those who don't use the excellent and robust VMC plug to manually reset the subdivs back to 8).

 

Some things I found: Default ray bundle size is too small for any 'decent' render outside of very early tests. I kept at least 256 to avoid nodes idling.

I also mostly kept Min. Shading Rate at 8 by default, except for shots that feature DOF/Mblur, where I took it back to 2 to get higher amount of passes to clear the AA.

With BF+LC in progressive, I keep Retrace threshold now ON by default in 100perc. times.

[Scott Schroeder]

Does anyone have experience doing an animation with it? It seems like it could have both positives and negatives for animations.

[heni30]

Can you keep going after it's stopped at your designated time limit?

 

You can just save at different intervals but it would be nice to have a pause feature.

Edited July 30, 2014 by heni30

[Justin Hunt]

I tried it with an animation, it was a simple fixed camera but with lights moving, in this case an animated light material.

 

I ruffly worked out that I would need 5 minutes per frame to meet the deadline, needless to say that 5 minutes was OK for a draft but no way near enough for final.

 

jhv

[Chris MacDonald]

Haven't really had a proper chance to experiment with it, but I've found the brute force speed-up to be enough to stop using the irradiance map entirely now, which is wonderful!