Who can get this UNIVERSAL 2.4 LWF scene down to 5min render?

[AlexP]

Max ray intensity, dont mind me

[kosso_olli]

on my workstation, the rendertimes looked like:

5m 21s < scene as opened

Wow, for a single PC? What a machine is this?

[peterguthrie]

Does this not mean you end up with a non-linear result?

Also what is MRI ?

as the color mapping is set to 'none' (dont affect colors) its still the same result as far as i understand, the reinhard burn just makes the sampling a bit more efficient: http://forums.chaosgroup.com/showthr...579#post543579

[AlexP]

as the color mapping is set to 'none' (dont affect colors) its still the same result as far as i understand, the reinhard burn just makes the sampling a bit more efficient: http://forums.chaosgroup.com/showthr...579#post543579

Incredibly useful info.
The only thing holding my renders up these days are those pesky over bright highlights

[kosso_olli]

Very useful info indeed! Thanks Peter and Vlado.
Using these settings I am now at 3:21mins. Settings are as seen in the screenshot, plus 192 reflection subdivs on the materials and 256 subdivs for the Brute Force GI.

[Morne]

A few very nice results you guys are getting here.

So now, how to get this cleaner and faster in VRay 2.4?
I could drop the noise threshold more to get it cleaner, but then it pushes the render times up again

[glorybound]

yep! That's the game we play - quality vs. speed.

A few very nice results you guys are getting here.

So now, how to get this cleaner and faster in VRay 2.4?
I could drop the noise threshold more to get it cleaner, but then it pushes the render times up again

[Morne]

So far I got it down from my initial 11m 28s, to 8m 1s, but the noise is identical

Peter, were your tests with your Z820, or do you have a newer one now?

[Morne]

It appears that the noise is mainly GI related. The other elements are clean compared to the GI

[kosso_olli]

A few very nice results you guys are getting here.

So now, how to get this cleaner and faster in VRay 2.4?
I could drop the noise threshold more to get it cleaner, but then it pushes the render times up again

Did you try these settings here? It should be working in 2.4, too:

http://forums.chaosgroup.com/showthr...302#post622302

[Morne]

Did you try these settings here? It should be working in 2.4, too:

http://forums.chaosgroup.com/showthr...302#post622302

A little slower than my previous. This one comes in at 8m 19s, but MUCH cleaner

[kosso_olli]

Forgot to say, material subdivs were set to 192 and GI subdivs to 256 for this scene. If you can live with a little more noise, try raising the threshold for a little more speed.

[jstrob]

50.6 sec!

On my single workstation (dual xeon E5-2680 v2 @ 2.8GHz (40 threads) with 64 GB of ram):

-as opened: 4m18s
-with kosso olli fastest settings: 2m50s
-with my settings and some scene tweaks: 0m50.6s! The result is a bit different but I think it could be made more identical by bringing all the vray element in comp and working it a bit in comp.

note that my antialias is far better than the original scene and kosso olli's rendering on the edge of the ceiling. I can't stand jaggy edges and they look even worst in animation.

here is my 50.6 sec result. Can you guess what are the tweaks that I did to the scene?


the attached image at the bottom is different, I made a mistake and the lower right corner was too dark.

[Morne]

Can you guess what are the tweaks that I did to the scene?

Looking good!

I'm geussing you switched to reinhard with 0 burn, enabled maximum ray intensity on 10, embree
BUT, that's cheating :- )

You were suppose to use 2.4
Enlighten us about what you changed?

On a side note, do you have a seperate machine with less cores but higher GHz to calculate your phoenix sims, or you do it on this one also?

[jstrob]

Hi Morné,

Cheating? that's what I'm paid to do all day!!!

and no I don't have single core machine only dual xeons since many years. I heard that i7 can be faster to simulate phoenixfd but I am a bit sceptical or I don't understand something about that. Cause dual xeon are always a bit in advance in terms of speed and they also cost more. So it's difficult to compare then to i7 cpu. Maybe at equal cost the i7 can be faster. But if you shell out for the fastest dual xeon, I doubt any available i7 will beat it.

For sure I used embree. I also left almost everything at 8 sampling like I almost always do. I only increased a bit the reflection sampling of that anisotropic chrome material to 18 cause there was a lot of noise in some reflection of that specific material. But most of all I removed the vray sky and replaced it by a plane with a falloff and I deactivated GI entirely and faked it with some directionnal standard lights and a few vray area light. Some of them with shadow entirely deactivated. I also replaced the vray sun by a standard directional light.

I always try to avoid lights like dome light and vray sun cause they add time to render and are easily faked with area light and standard directional ones. I also always try to fake GI when possible. I do a GI render and then try to reproduce it by "painting" with lights. In this case I also used a negative ligth to add some darkness in the right upper side.

I used Reinhard (with burn at 0.05) for almost the first time in my life! I recently learned why to use that Reinhard type that is useful when you have that kind of archviz scene with a bright sun bringing light in a room. Usually I don't do archviz and I always use linear multiply and do tweaks in comp.

I used the image filter (mitchell-netravali) for better antialiasing. I always work in animation so it's always a must for me.

Max ray I left it at 20.

min shading rate to 2

adaptative image sampler: I always leave those at 2 and 16.

global dmc:
adapt: 0.85
noise thresh.: 0.02
global subdiv: 16
min samples: 8


Basically I always start from those setting and if i have a big noise problem I will then try to add sampling in some specific materials or lights. Looking for noise sources in the vray render elements.