Test scences GPU Path vs CPU BiDir

Post your tests, experiments and unfinished renderings here.
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Test scences GPU Path vs CPU BiDir

Post by FarbigeWelt »

CPU BiDir, Path and GPU Path
CPU BiDir, Path and GPU Path
Maybe I have to check the CPU and GPU scene, because today they rendered Path differently, although all pictures

Unfortunately, latest ee4d881 crashes on my PC Windows system if Photon GI cache is enabled. This is why there is currently no GPU PGIC image.

This time, it does not matter if Sobol or Random sampler is chosen, the the software crashes rather evil with mouse freezer, screen flashes/flickers and best thing is to reboot after that kind of crash.

Build ee4d881 crashes with PGIC

Code: Select all

[LuxRays][19.828] [Device Hawaii Intersect] BVH mesh vertices buffer size: 180Kbytes
[LuxRays][28.953] [Device Hawaii Intersect] BVH nodes buffer size: 361Kbytes
Traceback (most recent call last):
  File "C:\Users\Michael\AppData\Roaming\Blender Foundation\Blender\2.79\scripts\addons\BlendLuxCore\engine\__init__.py", line 85, in render_final
    final.render(self, scene)
  File "C:\Users\Michael\AppData\Roaming\Blender Foundation\Blender\2.79\scripts\addons\BlendLuxCore\engine\final.py", line 40, in render
    _render_layer(engine, scene, render_slot_stats)
  File "C:\Users\Michael\AppData\Roaming\Blender Foundation\Blender\2.79\scripts\addons\BlendLuxCore\engine\final.py", line 63, in _render_layer
    engine.session.Start()
RuntimeError: clCreateBuffer
ERROR: clCreateBuffer
Error: EXCEPTION_ACCESS_VIOLATION
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Test scences GPU Path vs CPU BiDir_fix

Post by FarbigeWelt »

Indeed the scenes have not been set up the same way.

After exchanging the point lights by emitting (irregular spectrum) planes (emits only in one direction), GPU Path and CPU Path look almost the same (PGIC seems not to matter here).
CPU BiDir, Path and GPU Path_fixed
CPU BiDir, Path and GPU Path_fixed
As expected CPU BiDir looks different, quite interesting is that with BiDir one can see the whole cubes, i.e. the part behind the emitter gets light too, probably because of light mirroring in side the cube. I wonder if path BiDir (Eye 10, Light 10) compared to Path (Total 8, Diffuse 4, Glossy 4, Specualr 8) explains the mirrored light. However, the light from the plane must be brighter in case of BiDir what explains why the fonts are darker (tonemap linear, auto brightness).
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Test scences GPU Path Sobol, horizontal patterns?

Post by FarbigeWelt »

The pattern of Sobol on GPU looks unknown to me. Where do the horizontal patterns come from?
Animated GIF, click it if it does not play at the post's window.
The animation shows 25 to 2500 Samples over 25 frames.
NewSobolPattern.gif
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
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. »

FarbigeWelt wrote: Sun Apr 28, 2019 9:42 pm Where do the horizontal patterns come from?
I've also wondered about this, seen it a lot of times.
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

A Cascade of Prisms

Post by FarbigeWelt »

A Cascade of Prisms
Setting: openCL, UniDir Path, with Sobol, two GPU (AMD, Hawai).
Animated GIF, from a few samples to 1000 and de-noised.
(The first few frames are prolonged as well as the last frame.)
The attachment Prism_openCL_PGIC.gif is no longer available
After 4000 samples there are small improvements in the 'rainbow' parts.
Prism_Path PGIC_OIDN
Prism_Path PGIC_OIDN
The same image on CPU with BiDir and Metropolis (about 6000 samples).
Prism_BiDir_OIDN
Prism_BiDir_OIDN
I like both versions of the scene, the BiDir looks rather static like some future monolith things hovering whereas the UniDir looks dynamic almost in a rush to jump at the viewer.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
Sharlybg
Donor
Donor
Posts: 3101
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: Test scences GPU Path vs CPU BiDir

Post by Sharlybg »

The first one (PGI + OIDN) look amazing ;)
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA
Fox
Posts: 437
Joined: Sat Mar 31, 2018 11:17 am

Re: Test scences GPU Path vs CPU BiDir

Post by Fox »

B.Y.O.B. wrote: Sun Apr 28, 2019 9:47 pm
FarbigeWelt wrote: Sun Apr 28, 2019 9:42 pm Where do the horizontal patterns come from?
I've also wondered about this, seen it a lot of times.
Sometimes when i render with luxcore external ui at 3840x2160 pix, render can be even half an hour later stripy, but at 6000x4500 pix, seems like sampling algorithm is more even (closer to like random gaussian was in LuxRender), even 1 minute later no black spots or stripes. But the color lines are new to me.
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Test scences GPU Path vs CPU BiDir

Post by Dade »

B.Y.O.B. wrote: Sun Apr 28, 2019 9:47 pm
FarbigeWelt wrote: Sun Apr 28, 2019 9:42 pm Where do the horizontal patterns come from?
I've also wondered about this, seen it a lot of times.
It is the GPU rendering: each thread receive strip of pixels to render (16 pixels). This is done to reduce some inter-thread synchronization overhead (by 16 times with 16 pixels).

At the beginning of the rendering, if the samples have very long paths, it is easy to see the threads progress of the strip (there is a huge difference between 0, 1, 2, etc. samples over the same strip while it is not perceptible with 100, 101, 102, etc.).
Support LuxCoreRender project with salts and bounties
Post Reply