Test scences GPU Path vs CPU BiDir

Post your tests, experiments and unfinished renderings here.
User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Test scences GPU Path vs CPU BiDir

Post by FarbigeWelt » Sat Sep 01, 2018 1:41 pm

B.Y.O.B. wrote:
Sat Sep 01, 2018 1:08 pm
FarbigeWelt wrote:
Sat Sep 01, 2018 12:35 pm
Somehow materials got stranger with every change or use in the blender file used for the scenes above.
A good reason, to start with a new blender file and define objects and materials from scratch.
...wat?
Well, to test my hypothesis of the screwed file I've appended the new objects to the old file and what do you guess? The new material did not work like in the new file. I cannot tell why but observe. The strange behaviour of the old file might have started with the use of homogen and heterogen volume for world volume. Or it is connected to copy materials and replacing their material node to avoid defining volumes each time you make a new material. I've observed in another study image texture has kind of memory, replacing the image did show the new image in the image texture icon but render showed the old image. The only solution then was to setup a new material with the new image. Just as if not all properties of a material instance got updated as defined in the nodes.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
B.Y.O.B.
Developer
Posts: 1824
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Test scences GPU Path vs CPU BiDir

Post by B.Y.O.B. » Sat Sep 01, 2018 2:14 pm

This sounds extremely strange.
Can you upload a scene that shows the problem, with steps how to reproduce/what is off exactly?
Support LuxCoreRender project with salts and bounties

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Test scences GPU Path vs CPU BiDir

Post by FarbigeWelt » Sat Sep 01, 2018 2:24 pm

B.Y.O.B. wrote:
Sat Sep 01, 2018 2:14 pm
This sounds extremely strange.
Can you upload a scene that shows the problem, with steps how to reproduce/what is off exactly?
Yes, I can upload the scences later and where possible describe reproduction steps.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Some success mirroring scattered beams

Post by FarbigeWelt » Sat Sep 01, 2018 2:30 pm

CPU BiDir Null and 10 Glasses different 1 Mirror.jpg
CPU BiDir Null and 10 Glasses different 1 Mirror
Glass looks transparent and mirror reflects the scattered beams. Nice if it rendered much faster.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Scattering Beams Files

Post by FarbigeWelt » Sat Sep 01, 2018 7:57 pm

Scenes.zip
(206.47 KiB) Downloaded 17 times
This file renders pretty well: Glass Scatter Clear World.
This file renders strange: Glass and scattering beams.

Load the file and start the render.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

wasd
Posts: 109
Joined: Tue Apr 24, 2018 7:20 pm

Re: Test scences GPU Path vs CPU BiDir

Post by wasd » Sat Sep 01, 2018 11:01 pm

B.Y.O.B. wrote:
Sat Sep 01, 2018 1:08 pm
...wat?
That's what I tell you in almost each of my posts – Carthago delenda ... er... there's bug that should be fixed.

Image
Image
Image
viewtopic.php?f=2&t=575
CPU Bidir + Metropolis | Log Power | Core i5-4570

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Glass and its fancy transparency

Post by FarbigeWelt » Sat Sep 01, 2018 11:18 pm

If one sets a glass cuboid in front of an objects so that one part of the object is covered by the cuboid and the other not one can expect part behind glass takes more time to be rendered. One can compare number of samples per pixel to get a similar pattern for an object aside or behind glass. The beams aside the glass take about 1 to 4 samples to be recognized well as beams. One can see the white matte object already with one sample and it takes about 10 samples for a similar look behind the glass. This means matte behind glass gets rendered ten times slower than be viewed directly. But scattered beams and even point lights behind glass require much more than 400 samples per pixel to be recognized as well as aside the glass. This at least 100 times slower. The glass in the scene has reflection 0.0 and its interior volume is clear air. Air's low IOR should limit refraction effects.
Why takes a ray from a direct light much more computing time than a ray from a matte object? The direct lights, there are three point lamps with radius 0.1, are just below the lowest of the beams.
CPU-BiDir_Laser-Matte-Glass-IOR-Air-Reflect.-0.0.gif
CPU-BiDir_Laser-Matte-Glass-IOR-Air-Reflect.-0.0
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
B.Y.O.B.
Developer
Posts: 1824
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Test scences GPU Path vs CPU BiDir

Post by B.Y.O.B. » Sun Sep 02, 2018 9:45 am

wasd wrote:
Sat Sep 01, 2018 11:01 pm
That's what I tell you in almost each of my posts – Carthago delenda ... er... there's bug that should be fixed.
Forum posts are hard to keep track of and remember for me.
viewtopic.php?f=4&t=2
You should make a bug report on github so it can be more easily found: https://github.com/LuxCoreRender/BlendL ... issues/new
Also, I'm not seeing a test file in your thread.
Support LuxCoreRender project with salts and bounties

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Test scences GPU Path vs CPU BiDir: Wooden Room

Post by FarbigeWelt » Tue Sep 04, 2018 5:05 pm

At the moment topic number one is glass' transparency. Again I don't understand the difference of a ray from different sources. There is obviously light from a source like point or sun or a scattered beam and there is another kind of light from HDRI or indirect light from an object made from matte or something roughly. At least material glass acts as there were different types of light.
Wooden Room open Door, Glass Windows_GPU Path, random, 800 Samples.png
Wooden Room open Door, Glass Windows_GPU Path, random, 800 Samples
CPU BiDir Daylight 1.jpg
CPU BiDir Daylight 1
CPU BiDir Daylight 2.jpg
CPU BiDir Daylight 2
GPU Path Daylight Arch. Glass.jpg
GPU Path Daylight Arch. Glass
Just to be sure, test with sun only (means no sky, no HDRI) and a single point lamp in the room.
CPU BiDir Sun without Sky, Glass.jpg
CPU BiDir Sun without Sky, Glass
GPU Path Sun without Sky, Glass.jpg
GPU Path Sun without Sky, Glass
GPU Path Sun without Sky, Arch. Glass.jpg
GPU Path Sun without Sky, Arch. Glass

Conclusion: In combination with GPU use architectural glass for rooms or buildings because it is transparent for direct light and also shows a mirroring image.
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

User avatar
FarbigeWelt
Posts: 362
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Metropolis, the choice for difficult room lights

Post by FarbigeWelt » Sun Sep 09, 2018 8:51 am

Actually, I was looking for an optimal openCL render for the completed wooden room.

First, I used random sampler also in combination with denoiser or post processing without getting a useful result. Then I switched through all available samplers in luxcorerender addon looking for an appropriate solution. The images are impressions on the way to a clean version of the image.

GPU Tile path rendered the scene quite fast but the image was noisy
Wooden Room 250 Samples GPU Tile Path clamped 10000.jpg
Wooden Room 250 Samples GPU Tile Path clamped 10000
or if clamped with lower value
Wooden Room 250 Samples GPU Tile Path clamped 100.jpg
Wooden Room 250 Samples GPU Tile Path clamped 100
the ashtray made from glass got dark and table's colors faded.
Random and Sobol showed similar results.

GPU Metropolis looked different from the first few samples and even more promising with 250 Samples then I increased the number of samples.
Wooden Room 1000 Samples GPU Metropolis not clamped.jpg
Wooden Room 1000 Samples GPU Metropolis not clamped
Wooden Room 1000 Samples GPU Metropolis clamped 2859.jpg
Wooden Room 1000 Samples GPU Metropolis clamped 2859
Already a quite clean image but clamping also lead to a dark glass.

Intermediate conclusion: Looks like openCL won't lead to a final version for this room scene.

Let's give Sobol on CPU BiDir a chance.
Wooden Room 72 Samples, CPU BiDir, sobol not clamped.jpg
Wooden Room 72 Samples, CPU BiDir, sobol not clamped
The ceiling looks more realistic if compared to path render but I stopped here wiht 72 samples because of my experiences with Sobol on path.

Again Metropolis proved its strenghts for this kind of scene. The only issue left has been the ashtray on the table.
Thanks to Metropolis' parameter maximum consecutive rejects I finally got what I was looking for.
Wooden Room 821 Samples, CPU BiDir, metropolis max cons. rej. 16384, not clamped.jpg
Wooden Room 821 Samples, CPU BiDir, metropolis max cons. rej. 16384, not clamped
I've noticed earlier there is an issue with ashtray object. It looks so easy to be modelled but it is a nasty one to get it nicely smoothed.
This and because table was over exposed (render starts with fitting exposure but gets over exposed with incresing number of samples) I started a second render. On my machine it took almost 11 hours to get the required 1000 samples, less are a halfway house.
Wooden Room 10h48m 1000 Samples CPU BiDir Metropois.jpg
Wooden Room 10h48m 1000 Samples CPU BiDir Metropolis
(The ashtray needs a further update. I know smoking is bad, but I like the test.)

Conclusion: Metropolis is the choice for difficult illumination in a room.

EDIT: Impressions during rendrering.
OpenCL tile path, clamped
https://youtu.be/-gwMqE-Eoyc
OpenCL, path, metropolis, clamped
https://youtu.be/OzDfO2sk7Kk
OpenCL, path, metropolis
https://youtu.be/9vOYWnespRQ
CPU, BiDir, random
https://youtu.be/TphfSHVZI3A
CPU, BiDir, metropolis
https://youtu.be/eaF9Ld5Z9r8
Microsoft Windows 10 Professional, intel i7 4770K, 32 GB, AMD R9 290x 4 GB, AMD R9 390x 8 GB
Instagrammer, please join #luxcorerender for your renderings.

Post Reply