V-Ray Render Optimization - an in-depth Guide (call for Before/Afters)

[Morbid Angel]

You DO NOT generally want to be using the RAW passes to diagnose noise!

While this makes sense to anyone who understands raytracing and sampling in general, go explain it to my supervisor who calls me in the office and opens each raw pass and says - WHY THERE IS GRAIN HERE! Go fix it.

[vlado]

Well, we could remove the raw passes from the renderer

Best regards,
Vlado

[joconnell]

You take that back! You take that back right now

[Dariusz Makowski (Dadal)]

Well, we could remove the raw passes from the renderer

Best regards,
Vlado

Nooo plz noooo Mi love raw!

While this makes sense to anyone who understands raytracing and sampling in general, go explain it to my supervisor who calls me in the office and opens each raw pass and says - WHY THERE IS GRAIN HERE! Go fix it.

Just make RAW passes to be hidde objects so that they are not displayed in ur superviz explorer : p

[joconnell]

That's actually a point Vlado,

You'd mentioned at one point (especially regarding the raw reflection) that the maths was somewhat off and you'd be better off making your own by dividing reflection with the reflection filter but that you were going to change the maths for this result to come out in the normal raw passes - has this been done?

[RockinAkin]

You'd mentioned at one point (especially regarding the raw reflection) that the maths was somewhat off and you'd be better off making your own by dividing reflection with the reflection filter but that you were going to change the maths for this result to come out in the normal raw passes - has this been done?

As of V-Ray 3.X this seems fixed and the normal non-raw passes now composite correctly... I got exactly the same result as when you manually create the passes using the RAW render elements.

[joconnell]

That'll be one less step for the compers in here, result. Also this forum needs some kind of "thanks" function for quick replies!

[Morbid Angel]

Nooo plz noooo Mi love raw!



Just make RAW passes to be hidde objects so that they are not displayed in ur superviz explorer : p

I got F...ED so hard for this one flickering pixel once, I had the settings so high, it was probably the cleanest cg render I ever did, but there was one highlite I don't know what from that was just flickering, it was less then 1 pixel big The f...ed my brains for weeks on it, where it would have taken a compositor less then 2 hours to paint it out.

[joconnell]

Mental ray used to come up with random not a number pixels (NaN values) and I reckon you could use the curve tool in nuke to get tracking data for where that was and do an auto clone on it. Compers are so high maintenance

[Morbid Angel]

Mental ray used to come up with random not a number pixels (NaN values) and I reckon you could use the curve tool in nuke to get tracking data for where that was and do an auto clone on it. Compers are so high maintenance

Yeah man, it really depends on a comper. I work with some who would go in and just roto and patch cg without even asking for a fix, and then I work with some that would open my render scrub through first half and say oh, all frames are black and just sit idle for days, not realizing that cg was appearing later on in the sequence

[leejk]

Just wanted to post my findings with my scene and was wondering why it turned out this way. The VRAY 3 settings in my scene prior to trying this optimization technique where:

• Global DMC noise set to .005, adaptive amount to .85
• Adaptive image sampler subdivs set to 1 & 4. Use DMC sampler threshold checked.
• Image sampler set to adaptive
• 1 dome light with it's subdivs set to 24
• For GI, using IR Map & Light cache as secondary. IR map uses low preset

My render time is about 1:55 for a 1024x768 image. There's a slight bit of noise in the lighting and reflective passes, but very much acceptable. There is a LOT of red tho in the sample rate pass, particularly around the shadows and dark areas.

Now when I try the optimization technique, I have to increase the light subdivs to 64 to get the same quality in the light pass. This is using the default noise value of .01 in the Global DMC. Now I also have a lot of noise in the GI pass, but the sample rate pass has much less red in it now. My point being is that the more I try and optimize the separate passes, the more my render times go up, and it all seems to stem from increasing the DMC noise from .005 to .01. So my question is, where in the optimization process do you determine the right DMC noise setting? It seems that it must be done first. Since everything else is so dependent on this, it sort of makes the optimization process irrelevant.

[RockinAkin]

When you're optimizing a scene, you're doing it according to the noise threshold you currently have set.

So if you go through the optimization process on your scene with a Noise/Color Threshold of 0.01 - and you get it to the point where you're getting the fastest render possible with the least amount of red pixels in your SampleRate render element - then you know that your scene is now optimized according to that noise threshold.

If you decide that you need less noise, then you have to lower your Noise/Color threshold. Now that you're asking more from V-Ray's samplers, you'll immediately get much more red in your SampleRate. In which case you'll have to go through your scene and adjust your sampling again to meet the lower threshold.

[leejk]

Ok that sort of makes sense I guess. So to reiterate the thinking process:

1. Using the default noise setting of .01, go thru the optimization process
2. If there is still noise present that can't be eliminated, such as I had in the GI pass, decrease the DMC noise setting and run thru optimization process again

But from my tests here the effort to get rid of the red in the sample rate pass is not reducing render time, so I'm not seeing how this is optimizing anything. When I decrease the DMC noise setting it fixes my noise in the GI pass and also let's me set my light's subdivs back to 24, but the sample rate pass has lots of red. But by working thru the optimization process to get rid of it causes renders times to increase a lot. The optimization tutorial is great I think and when I first read it I thought it completely makes sense, but my scene is proving otherwise. It seems in this case there is better performance by letting the AA engine do all the work. Is it possible I have some settings messed up or does this optimization process not apply to all scene setups?

[joconnell]

I woukdn't be a total slave to the sample rate element, it's only one bit of information. If you're getting render times and quality you're happy with then who cares if the sample rate is red? 4 is a pretty low max aa amount for a render anyway so there's a pretty good chance a lot of your scene will end up using something around it. If your render is slow, you've slightly higher aa and the result really dirty, then you can start using the sample rate to hunt around a bit, if you're happy with your results though then all good!

[RockinAkin]

It really depends on your scene whether or not manual optimization will benefit your render - could you share any images?
Some scenes are very Image Sampler heavy, some are somewhere in between, and some are very DMC Sampler heavy.
Optimization attempts to balance the load between V-Ray's two samplers to whats most appropriate to your scene's needs.

And as John said, the goal isn't to eliminate all red pixels from the SampleRate element - it's just a tool to identify if the Image Sampler is being stressed by trying to resolve noise that's better suited for the DMC Sampler to handle (like in shadows, reflections, gi, etc).