[luis.hijarrubia]

One of the features that are better on the new GPU engine than in production traditional one: Caustics through a glass (sun light for the moment). With this pool renders will improve a lot.

CPU vs GPU

[Mihai]

Not to mention the caustics reflected in the mirror itself, and the little green streak of caustics from the glass, in the bottom of the render. Nice!

[Half Life]

This is what I have been waiting for (usable caustics) -- version 4 is already a huge improvement in my book based on this alone.

[Aniki]

This is pure niceness! Now get this working flawlessly with sss and translucent/single scattering materials and I buy within an blink of an eye;)

Yet I wonder why the right caustics are getting way more blurred towards the distant side of the mirror?

And why does the left mirror not show the caustics mirrored? What causes the difference of the caustic streak to be green and white plus changed position/angle?

[Nova66]

The Caustics in the GPU render are finally usable but whatever code magic has been done to the GPU code, really needs to be applied to the CPU code as well. As much as possible, the two render engines really need to output the same results.

I don't know if the MicroStation plugin will ever be upgraded to v4, so what are the chances of this Caustics fix being applied to one final release of Maxwell 3.2.x.


Andrew.

[luis.hijarrubia]

And why does the left mirror not show the caustics mirrored? What causes the difference of the caustic streak to be green and white plus changed position/angle?

Get caustics through a glass or reflected on a mirror is very similar problem. Both of them are new things on GPU engine.

[Aniki]

And why does the left mirror not show the caustics mirrored? What causes the difference of the caustic streak to be green and white plus changed position/angle?

Get caustics through a glass or reflected on a mirror is very similar problem. Both of them are new things on GPU engine.

I meant why the CPU side isnt reflecting the caustics in the mirror;)



Way less excited about caustics now...

[Half Life]

I notice the sun is larger (more diffuse) -- check out the sun reflection on the ball in the GPU render versus the CPU render.

I suspect this is part of why the caustics overall are working better on GPU, but also part of why the resulting caustics would be softer/more diffuse. This was a trick I used with v3 renders to get workable caustics (at the cost of sharpness). Without testing myself this is just a guess though.

[luis.hijarrubia]

I meant why the CPU side isnt reflecting the caustics in the mirror;)

CPU never reflected caustics on a mirror. BPT problems.

[JDHill]

CPU never reflected caustics on a mirror.

For small* values of never.







* relative to the age of the universe

[Aniki]

changing the sun radius, bigger or smaller doesnt change the blurriness for GPU.. so back to square one for caustics I assume..

[numerobis]

And why does the left mirror not show the caustics mirrored? What causes the difference of the caustic streak to be green and white plus changed position/angle?

Because the CPU engine is physically correct, but the new GPU engine is more physically correct...

[eric nixon]

So happy to see this news , Congratulations indeed.

Now I will direct my prayers to directional lighting, thank you Amen. (and a multishader ofcourse)

[seghier]

cpu still better

[PA3K]

Nice to see sun caustic through the glass. I was waiting for this feature since RC5 (did you implemented some kind of VCM algorythm for GPU render engine or is it secret?)
But ... why isnt there white line - caustic reflected from green glass plate on GPU image as it is on CPU image. Sun seems to shine very little from northwest rather then from direct West so it should be there. Also green line is nice on GPU image, but i am not sure it is physicaly correct. Reflected caustic bluriness as seen on GPU render from mirror should not be there as it is if there is sun of size 1. Shadows and caustic from planar surfaces should have blurriness of about 0.5 degree as is visual diameter of sun of size 1. Anyway it is far better even if the sun is bigger on GPU image, because there will be not visible reflected and refracted caustic on CPU image even if the sun will be of size 2.

Here is my first try with V4 without special caustic. Classic sponza scene. It is +/- the same time, but there is some white artefact on middle pillar on GPU image. Images are output from rhino fire draft CPU and draft CPU. Computer is g750JZ with i7 4700 and nvidia 880m gpu.

Patrik






image sharing sites