Rendering Hi-res relatively fast?

[Anders M?ller]

Hello

I've been looking around on he forums for som tips about relatively fast rendering settings hi-res images. I need a 5000x7000 pixels image for a presentation spread (A0 at 150) and it just takes forever (one of my trials took 36 hours but the textures were all in blotches and the model was gaping where it shouldn't)...

I have tried both QMC+QMC, QMC Light Cache, Irradiance+QMC and Irradiance+Light Cache but I can't seem to find the one that's faster...

I use the Vray sun and Physical Camera as it is an outdoor scene.

Any ideas/suggestions for settings would be much appreciated.

-Anders

P.S. I only have the one computer so DR is not really an option. (My computer is a MacBook Pro 2.16GHz with 2GB Ram)

[dalomar_cg]

Re: Rendering Hi-res relatively fast?

When it comes to HiRes most people don't approach the situation correctly. Most people choose to increase the quality of their settings because the images are typically final quality, but this actually this is usually what causes the drastic slowdown. If you understand whats going on at hi res, then you can render those images pretty efficiently without to much of a hit on render times.

Basically, at the heart of the issue, is that a high resolution image is going to be more accurate then a screen resolution image. This is because we're using more pixels to describe a given area than we were before. This actually means that high resolution images can actually be calculated with "less accuracy" then their low res counterparts and still look good. With most calculation methods in V-Ray, resolution will play a part in determining the final quality (how much depending on the method itself), so increasing the resolution automatically increases the accuracy. We actually need to counteract this so that our settings are basically for the same accuracy that we would use at a screen resolution, simply scaled to a larger one. The result might be settings that look like the might not do the job, but at a high resolution it works out.

So here's the breakdown...I'll go through it section by section.
DMC
DMC is a GI calculation method that works on individual pixels. So since high res images have more pixels, its not really isn't very efficient (for simple scenes it can still work though). If you are going to use DMC then keep in mind that increasing the resolution will make the results more accurate. That means that you may try backing off the Noise Threshold or possibly even decreasing the number of subdivisions a little bit (6 instead of 8 )

Light Cache
Light Cache is actually the only method where resolution really won't matter. I try to manage the sample size anyway, so decreasing that to .01 or .005 (screen scale) will help prevent any LC blotches that might appear.

Irradiance Map
Irradiance Map is very tied into resolution, and setting it wrong has a tendency to cause things to slowdown quite a bit. The easiest way to deal with IR at high resolutions is to have good settings at a screen resolution, render out the IR calc at a screen resolution, save it, and reuse that solution at high res.

The manual way is that for each time you double your resolution, you should subtract 1 from your min/max rate....So if you have a screen resolution of 800x600 and a min/max rate of -3/-1, then for 1600x1200 you would set it at -4/-2, at 3200x2400 it would be -5/-3, and at 6400x4800 it would be -6/-4...hopefully you get the picture with that example. At resolutions in between those, you have to judge where to set your min/max rate, but use those as a guide.

AntiAliasing
With AA, there's going to be less of a discrepancy from pixel to pixel, meaning that you don't actually need as many AA samples as you used before. I don't scale my AA samples linearly, I typically just back them off a little bit and it works out.


Hopefully those guidelines help. I would advise that you tests out a few settings using the RenderWindow command as this will allow you to test the results of those settings at the proper resolution without rendering out a whole image. I'd say to also test the settings that you were previously using to compare quality and render time.

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

Thanks for the quick and thorough reply.

How would you Setup you Image Sampler? At the moment min is set to Adaptive Subdivision with Min: -2, Max: 2, Threshold 0.01 Normals at 0.05
AA-filter is set to Area with a size of 1.

Is that allright ou reckon?

-Anders

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

So now I've tried the settings you suggested and the calculation of the Light Cache and the Irradiance Map are both quite fast but the rendering time is incredibly long... Maybe my textures are terribly bad or something?

Should I convert the whole scene to a mesh model before rendering?

Should I change the bucket size?

Any suggestions?

[dalomar_cg]

Re: Rendering Hi-res relatively fast?

The rendering is still going to take longer than a normal one is. Textures shouldn't add too much to the render time. You're image sampler settings look fine, although I would disable the Area Filer because it won't do much at this resolution, and any filter will add render time. What are the times that you're experiencing?

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

The one that actually finished but had all the strange blotches took around 36-40 hours. The latest trials i have done have all been stopped at 24 hours but at that time they hadn't rendered more than 20 percent of the scene.

It is worth noticing that i have a quite large displacement-map and some pretty hefty Bump-maps as well and from my tests i have gathered that it's the bumps that take the longest. Should i change them to displacement-maps or is there som magic switch that says "do it faster"? :P

-Anders

P.S. I just noticed that i had double-sided textures turned on and they are not at all nescesary. I could imagine they would take up some time as well? I also turned off reflections on the maps under options. Maybe this will yield a more matte look that way and save som time (as an added bonus).

[dalomar_cg]

Re: Rendering Hi-res relatively fast?

I believe that the displacement is where your issue is. Basically the maximum edge lengths for displacement is calculated in pixels. At screen resolutions, the default edge length is fine, but at large resolutions that makes for a very dense displacement mesh. A dense displacement mesh will take more time to calculate, require more "involved" raytracing and sampling all around. I'd increase the Edge length to about 1-2% of your final resolution and see what relief that will bring.

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

Are there any similar measures to be taken with the bump-map? As mentioned i have not experienced as many issues with the bucket-time for the displacement-map as i have with the bumps? Maybe it'll change now that i have removed he double-sidedness?

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

The displacement is a 512x512 pixel map (13,7 cm with it's current resolution), that has been tiled to fill a 6x6 meter area (43 tiles on each side). is it 1-2pct of the 6x6 or just one of the tiles (wich would make it 5-10-ish)?

I'm sorry I'm asking this many questions but i reltively new to this and my paper is due on friday.

-Anders

[Anders M?ller]

Re: Rendering Hi-res relatively fast?

I think I just made it work.

Thank you very much for all your help. If you ever come to Denmark I'll buy you a beer for sure.

Thank you again.

-Anders

[dalomar_cg]

Re: Rendering Hi-res relatively fast?

As far as bump maps there is no similar steps that need to be taken. For the displacement I was referencing the percentage of the image resolution not the map itself...sorry for the confusion on that.

Thank you very much for all your help. If you ever come to Denmark I'll buy you a beer for sure.

Glad you got it working...I think I've got to take a tour of Europe now ;D