Area Light Comparison

[atlantis23]

Here's an old indoor test scene, rendered using the old area lights (LightPlane shader), and a new render using the v8 area lights, photon maps, and gamma correction.

Render time for the old version was 22 h (at 1680x1050 on my iMac G5 @2.1 GHz). Render time for the new version was 2.5 h (same resolution, same computer).

Full resolution images are here

Old version www.atlantis23.com/bluebedroom6_4.png

New version www.atlantis23.com/bluebedroom7_3.png

[michel faubert]

Superbe !!!

Michel

[Mark Johnston]

Hi wow what a time savings . Nice work !

Mark

[Tomas Egger]

Ola Atlantis23,

first of all, excelent illumination...

Im intrigued, probably its possible to make the time render even faster.. Could you post some grabs of your settings? or your prj to do a test here in EIAS3d team?

Thanksss

Tom

[atlantis23]

Thanks for the comments.

Tomas, I PMd you a link to the prj. I'm sure there is some potential for speed optimization.

[Tomas Egger]

Ola Atlantis,

I just sent a optimized project to you..

my render times:

Mac Intel 3hz (8 cores).. 10.6.2 OSX

1 core render = 20min 6 secs

6 cores using Slaves = 5 min 6 secs

Any doubt or chat about How I did it.. lets share with our friends here...

Hope you like it.

Thanksss

Tom

[Igors]

Hi, Atlantis23, All

Few technical notes:

- Area Light is NOT "an array of (radial) lights" (as, btw, old LightPlane shader did). Such approximation is acceptable to simulate environments but creates artifacts closer to the array.

- Area Light and GI Sky Light use different calculation techniques but produced illuminations are essentially same. A window (where light comes from) can be considered as "a part of sky".

Glad to see EI8 features in use :shy:

[atlantis23]

Hi Igors,

Thanks for the clarification.

Hi Tomas,

Thanks a lot for the modified project. Render time on my computer is now 48 min.

www.atlantis23.com/bluebedroom7tom_1.png

Most of the time saving is probably due to your sampling level 1x1 instead of 2x2. I used 2x2 for a fair comparison --the old render was 2x2-- but also to get better texture detail.

You're using much larger lookup radii (and database segment length). Your lookup radii cover the entire scene. According to the manual, p. 258, larger radii increase render time. Why are your larger radii faster than my smaller radii?

I used the same lookup radius for "Direct" and "Bounced". You use a higher lookup radius for "Bounced". Does this mean your bounced light is brighter? Does a higher Bounced radius affect render time?

When I render your project on my computer, some of the ambient occlusion shadows are missing, for example curtains shadow on the floor, bed shadow on the floor, ledge shadow on the wall, lower part of the night-stand. There is bounced light between the night-stand and the bed which shouldn't be there, IMO. I will see if this is caused by your "Bounced" lookup radius.

BTW, what is the render time of my project on your computer? A single Intel core seems to be more than twice as fast as a G5.

Thanks again for the testing.

[Tomas Egger]

Ola Atlantis,

My Time:

1 core render = 20min 6 secs (so, less than 1/2 of your render time)

6 cores using Slaves = 5 min 6 secs, if I use Rama and slaves (6 in this case), my render was in 5 min.

These are my main parameters to my scenes, its a Time vs Quality balance, but you are correct, your image have more bounced light in the end of the room.

We could ask Igors and see if they could give us some tech light and explain whats the best in our case here.

Im posting your settings to Igors too, so we can think togheter... and analyze.

Thanksss

Tom

Atlantis:

My Settings:

[atlantis23]

Hi Tomas,

Thanks for the comment. Looks like the lookup radii are very important. Another question would be how area light ray count versus GI ray count affect speed and quality in this case.

[Igors]

Hi again

Our settings are slightly different :shy: Segment length is set to 0.1

We used 2 radii 0.05 and 0.1. Using second (bigger) one increases baking time and not absolute necessary. Just bounced photons are typically more disordered/scattered thus it makes a sense to average them in larger radius.

We also used one Area Light (with Block Distance) instead of 3 - easier to control. Attachments (from left to right)

- Setiing screenshot

- Direct light only

- Photon map

- Final render

[Tomas Egger]

Ola my friends,

Could you post the database values? please.

Which value was used in Block Distance?

So, its only a larger area Light than 3 in each window.

Thanksss

Tom

[atlantis23]

Hi Igors,

Thanks for the testing. Did you increase the ray count for your single area light? From 100 to 300, for example?

A single light is probably faster because of reduced baking time. I'm rendering now using 3 lights and your settings (Direct radius 0.05, Bounced radius 0.1), baking time for one map is about 15 min (about 1.5 million data points). I will post my render time in a few minutes.

[Igors]

Hi, Tomas, Atlantis23, All

Using a single (longer) light is more practical, easier to control. Note that 3 lights illumination (with same intensities and sum square) is not equal to produced by a single one.

-We set light's intensity 1.0 (was 0.4) it can be changed without re-baking.

-Block Distance = 0.3,

-Rays Count = 400.

-Segment length = 0.1.

-Our baking time is 7-8 minutes (Intel Xeon 2.66), database contains 600K points (18 Mb modest database)

The considerations are very simple: a large image is planned to render, most probably several times (or many times). Thus it makes a sense to prepare a careful baked photon map because it can be done once. It's absolute possible that wanted result could be achieved with only 1 million of photons or less yet (instead of 5 millions we used). But baking is a single/once step, so we can be a bit wasteful :rolleyes:

It's always helpful to see what's baked: turn GI off and set "Always Visible" light mode. With this settings it's very interested to "walk" inside baked room - just animate camera and render with few seconds per frame.

You can use more or less photons as you want, anyway the main idea is: first create a draft illumination (baked photon map) for your taste/needs. The final GI illumination will essentially repeat a mimics of the draft one.

[atlantis23]

Hi Igors,

Thanks for the explanation.

My first test using your settings on 3 area lights produced a lot of redundant calculations. Image quality was excellent but render time was exorbitant.

Here's the last test using 1 area light with slightly reduced settings

www.atlantis23.com/bluebedroom7igors_3.png

and the photon map

www.atlantis23.com/bluebedroom7igors_2_pm.jpg

Baking time is about 4 minutes. Render time on my computer is 1 h 24 min. Using 6 Intel cores, this means about 7 minutes render time.

The settings are:

- 1 area light, ray count 300

- photon count 1000K

- photon density 194

- segment length 0.2

- direct radius 0.1

- bounced radius 0.15

- GI ray count 600

- GI color tolerance 1

- sampling level 1x1

- "4x4 Average" antialiasing