glass..three years on!

[tricky]

I must've been using Vray for the best part of 3 or 4 years now. Most of my work is with images, so render times aren't really a big issue (as long as they get done overnight!) and glass is usually comped in in photoshop.

However, we are increasingly doing more and more animation of buildings - buildings with lots of lovely glass! ( )

Now Vray is superb in most cases, and to be honest, I can get thew render times down pretty low and still get a high quality finish. What really slows down the rendering though is glass - especially when its right up close to the camera. Does anyone have any tips/techniques for getting glass to render quicker? The solutions I create are based around IRmap and QMC, and also calculate very quickly. I always tell glass to be invisible to GI and not to create any GI - just receive it.

An example at the moment is a school we are working on. Most of the renders are very fast - 10 minutes or so. We go through the entrance and on to view an internal glazed atrium, behind which is a little gravelled garden with a couple of plants/low poly people. The renders here jump to 2 or 3 hours per frame! I've tried all sorts of things to speed it all up, and to be honest, I have confused myself! (easily done - very stressed at the mo!). IOR set to 1.0. Reducing the Max depth to 5 or 6. etc etc

What do you guys do?

My glass is very simple:
Diffuse - black
Reflect - falloff (27,27,27 - 107,107,107)
Refract - white
IOR - 1.6

[percydaman]

have you considered using standard mat and just use opacity instead of using any refraction at all? if you dont really need to see active refraction, give it a try.

[natty]

in my tests, standard mats seems to render slower when using opacity and a vraymap for reflection.

what i tend to do now is use planes for windows so its less complicated for vray to render. just did a quick test on a very simple Tpot scene and the plane rendered 10secs quicker than a box.

[Gijs]

If I remember well, Vlado told that checking fresnell reflection works faster than using a falloff map. I don't have any experience with it though.

[cpnichols]

Also keep in mind that if you really want to be "true," you would make would put the inverse falloff on the refraction from the reflection. As a surface gets more reflective, it gets less transparent. I don't always use that myself for simple window glass and stuff, but I like it every now and then for things like a vase.

[davision]

Also keep in mind that if you really want to be "true," you would make would put the inverse falloff on the refraction from the reflection. As a surface gets more reflective, it gets less transparent.

This brings up an interesting point -- one that constantly comes up when I'm developing materials. Fundamentally, the question is: does the renderer automatically (inherently) account for certain phenomena without the user having to manually recreate it?

For instance, if you are increasing the reflection of the material, isn't the higher level of reflectivity making the material more opaque intrinsically -- obviating adding an inverse falloff to the refraction?

Or another example was when Vlado had a post show that materials' glossy reflections become "sharper" as the normals become perpendicular to the camera. That is a natural effect as far as I know, but I don't know whether I need to put a perpendicular falloff in the glossiness slot, or if Vray creates a sharper reflection implicitly because the object's faces are effectively being "squished" together when viewed from a near perpendicular angle.

Vlado would be a good person to ask I suppose -- or anyone who has done extensive materials testing.[/i]

[tricky]

Also keep in mind that if you really want to be "true," you would make would put the inverse falloff on the refraction from the reflection. As a surface gets more reflective, it gets less transparent. I don't always use that myself for simple window glass and stuff, but I like it every now and then for things like a vase.

Thats fair enough, but really, when frames are taking 2-3 (even 4 hours now that I have come in this morning), truth flies out of the window. I need a faster way to 'imply' windows. Its annoying the more I think about it - we have always had problems with glass, regardless of the renderer we use. It always takes ages.

We always model our glass panels as 10 or 20mm thick boxes (extrusions) in AutoCAD. I had read on a number of occasions that Vray can have problems with no thickness on meshes. Is this true, even with the latest versions? Natty - when you say "10 seconds quicker than a box", what are the actual times?

[vlado]

V-Ray automatically tries to preserve enregy in the VRayMtl, so if a surface gets more reflective, it gets less refractive (more opaque) so there is no need to do anything to force that behaviour.

As for the glossy reflections becoming sharper at grazing angles - at present, V-Ray does not do that (mostly for compatibility with older versions), so you need to do this yourself with a map in the glossiness slot.

Best regards,
Vlado

[trick]

2-3 Hours is way too much. Even if I have 2 or 3 glass planes in front of me time is max. doubled. I do use "overlapping" as sample lookup method though but I didn't have any problems with that...

[tricky]

Your telling me. Check this out:



This is using a pre-saved solution as well. The odd thing is, rendering locally and creating a 'low' quality irmap takes only 15-20mins per frame - thats with solution calcuation as well!

The pre-calced solution was QMC (10 bounces) and Irmap (medium animation, but with blurGI set to 0).

[trick]

Have you tried 1st bounce Irr.map, setting "High" HSphere "70/40", sample lookup "overlapping", 2nd bounce Lightcache subdivs "2000" ? Since there are only two planes of glass you're looking through, you should be able to render these in 10 min per frame (without precalc) on an X2 3800+ !!

[tricky]

Initially we were using Lightcache as it seemed the fastest way to render with interiors. However, we had many problems trying to use LC with an animation - machines would crash out before finishing the LC bit. One machine did manage to create its LC and save it out before crashing out, so I tried using this to create the Irmap, but the machines we tried it with failed to calculate anything. I emailed Vlado and sent him log files etc but he couldn't help me. This project is a rush, so we went back to the 'traditional' method of Irmap and QMC - slower in calculation, but it works at least!

So, in a nutshell, I can't afford to risk LC at this stage. Our machines are all X2 4400+ with 2GB ram.

[trick]

Maybe the problem lies in the LC calculation on X2's. All pre-map calculations here are made on dual Xeon's after which the final render phase goes to slaves, among which there are X2's. I never did the map calculation on X2's. I will have to try it sometime to see if this crashes. I never had crashes with anything in the 1.47.03 release and I just rendered more then 6000 frames on the farm. Maybe your LC files become way too large !?!

[RErender]

Perhaps the boxes are overheating, causing them to throttle back.

[tricky]

Perhaps the boxes are overheating, causing them to throttle back.

An interesting thought - I'll need to find some way of checking. Is it simply a matter of right-clicking My Computer and checking the clock speed?