My V-Ray concerns -- please tell me I'm wrong.

[BBB3]

Yes, I meant GGX.

[Dan Marks]

I've just come across this, and wondered if Vlado or Bertrand had considered the PC architecture Bertrand is using.

Over the past few years I have regularly use render nodes and occasionally a workstation similar to Bertrands machine, i.e. dual Xeon 32 threads, and although they work great in many occasions they have consistently given me problems in situations with high memory usage.

After doing some self initiated research into the cause a long time ago I resigned myself to the fact that this was possibly to due the motherboard architecture, in that only half of the memory is assigned directly to each CPU, to access the rest it has to cross the CPU bridge link... this is my suspected bottleneck and the cause of slowdowns.. and since I was out of time to investigate further I left it at that.

I now run watercooled i7s in a small but effective render farm now and I'm very happy with performance, and I've found that the latest Vray versions to be very good. When using a higher noise threshold, coupled with the denoiser and a good graphics card you can get amazingly quick results.

[vlado]

So I didn't have the GTR setup anymore but I just did a quick conversion, and the result is as expected. See below. I don't know but maybe I'm doing something wrong in terms of scene setup (lighting, exposure, color mapping). But after 20 minutes of progressive rendering and with active noise of 0.007 (30% left), it shouldn't look like this I guess.

The issue in your case is the VRayLightMtl material in the light fixtures. There are lights there already, but they don't cover the openings exactly and some GI rays manage to get inside and hit the bright light material.

I've seen this in a lot of scenes and in order to minimize this effect, for the next SP, I will be switching off the GI contribution of the VRayLightMtl material by default (although of course if can be enabled if one wants).

In any case, nothing to do with the GGX BRDF in this case.

Best regards,
Vlado

[vlado]

Dual Xeon 32 threads, and although they work great in many occasions they have consistently given me problems in situations with high memory usage. After doing some self initiated research into the cause a long time ago I resigned myself to the fact that this was possibly to due the motherboard architecture, in that only half of the memory is assigned directly to each CPU, to access the rest it has to cross the CPU bridge link... this is my suspected bottleneck and the cause of slowdowns.. and since I was out of time to investigate further I left it at that.

The shared memory between CPUs is indeed a problem, but we also found a bottleneck in our code that could slow things down on many-core machines quite a bit. Although it typically started to show with 64 cores and more. It's fixed for the SP, so hopefully things will run better on those machines.

Best regards,
Vlado

[francomanko]

This glossy Fresnel option... so currently i'm using a falloff curve on reflection map with the fresnel tickbox off, will this potentially be a new method to replace that?

[Jiri.Matys]

This shared memory problem with dual xeon motherboard has all motherboards on the market or new ones are ok without any problems? I bought two new xeons E5-2630v4 with great SuperMicro motherboard (each cpu has 10 cores, so 40 threads total). I have 64GB, but I dont have a 3D max scene which use more than 32GB, so... will there be some problem when I load up a scene which use more then 32GB RAM?

[jstrob]

Top one is Corona, bottom one is Vray. The point of interest here is the concrete floor. I could not use Corona PBR approach with V-Ray, because concrete would immediately light up and lose all the detail from grazing angle. Corona material is just IOR 1.5, 100% reflection and glossiness map, and V-Ray material is IOR 2.0, reflection color RGB 64, and same glossiness map.

No matter what combination of IOR, reflection color and glossiness map tweaking i tried, I could not get anything close to Corona result where I do not lose that glossiness texture structure at grazing angle.

Does it has to do with the fact that the GGX shader in many different renderers was not doing fresnel inside the BSDF so it was not taking microfacets into account for the fresnel and so when light is coming at grazing angle like in your image the GGX appear too bright? Now it seems every renderer are correcting this collective mistake and Grant Warwick just told me it's corrected in the latest V-Ray nightly.

I learned about that while watching that Mantra shading tutorial where we can even see an interactive vizualisation of the way fresnel and GGX are working together when the light is coming from varying directions, really cool to see, around 52:00 for the explanation (and comparison of doing fresnel outside or inside the BSDF) and 54:00 for the visualization :

[vlado]

Does it has to do with the fact that the GGX shader in many different renderers was not doing fresnel inside the BSDF so it was not taking microfacets into account for the fresnel and so when light is coming at grazing angle like in your image the GGX appear too bright?

Yes.

Now it seems every renderer are correcting this collective mistake and Grant Warwick just told me it's corrected in the latest V-Ray nightly.

It is corrected indeed, and it will be the default behavior starting from the next service pack.

Best regards,
Vlado

[Morbid Angel]

Its crazy how deep houdini is, however this also makes it slow to work in (I'm not a houdini expert) but the fact that you have to spend hours creating shaders its exhausting

[jstrob]

Cool to know Vlado!

And Dmitry, Houdini also has the Disney Principled shader, explained in the same previous video which is ready to use and faster to setup than any shader I think. But if you want to go deep one level (by diving inside the principled shader and playing with the fundamental BSDF), you can, if you want to go deep a second level (by plunging inside the BSDF and playing with even more basic nodes) you can and if you want to go deep a 3rd level (by writing your own basic node in VEX=vector expression as fast as C++ and unique to Houdini or VOP=vex operator= the visual equivalent of VEX) you also can! which is really cool. I love learning Houdini with sideFX masterclasses, it's like a pure fundamental CG courses I now understand things I used for many years and never fully understood.

[Mokiki]

Vlado do you mean the stable nightlies?

[vlado]

Vlado do you mean the stable nightlies?

Nope. Adding this feature to the stable builds will make them automatically un-stable

Chris compiled more information on glossy Fresnel with some nice exapmles here:
https://labs.chaosgroup.com/index.ph...lossy-fresnel/

Best regards,
Vlado

[Mokiki]

Gotcha

Which sadly means I have to wait but glad that this is coming.

[vlado]

Which sadly means I have to wait but glad that this is coming.

You won't have to wait for long

Best regards,
Vlado

[NicoC]

Chris compiled more information on glossy Fresnel with some nice exapmles here:
https://labs.chaosgroup.com/index.ph...lossy-fresnel/

Hello,
Please correct me if I'm wrong, but there's no way to get an RSS feed from Chaos Group Lab ?

Thanks.