World Sampling in Cycles broken when rendering with GPU #47213

New Issue

Sebastian Koenig · 2016-01-20T12:14:28+01:00

Sebastian Koenig commented

2016-01-20 12:14:28 +01:00

System Information
elementaryOS 0.3.2, GTX970

Blender Version
Broken: 2.76.8 a6eae73
Worked: 2.76.5, 90250f8

Short description of error
When rendering with Branched Path Tracing on the GPU using an HDR the World Samples setting is ignored.

Exact steps for others to reproduce the error
Render the attached file. It has world samples set to 24. Now render with world samples set to 1. It looks the same. If you try it with CPU you get the expected result, 24 samples looking much better than just 1 (also renders much longer).
square_samples.blend

**System Information** elementaryOS 0.3.2, GTX970 **Blender Version** Broken: 2.76.8 a6eae73 Worked: 2.76.5, 90250f8 **Short description of error** When rendering with Branched Path Tracing on the GPU using an HDR the World Samples setting is ignored. **Exact steps for others to reproduce the error** Render the attached file. It has world samples set to 24. Now render with world samples set to 1. It looks the same. If you try it with CPU you get the expected result, 24 samples looking much better than just 1 (also renders much longer). [square_samples.blend](https://archive.blender.org/developer/F275306/square_samples.blend)

Sebastian Koenig commented

2016-01-20 12:14:28 +01:00

Changed status to: 'Open'

Sebastian Koenig commented

2016-01-20 12:14:28 +01:00

Added subscriber: @sebastian_k

blender-admin commented

2016-01-20 12:14:28 +01:00

#47474 was marked as duplicate of this issue

Thomas Dinges was assigned by Sebastian Koenig

2016-01-20 15:28:10 +01:00

Thomas Dinges commented

2016-01-20 17:38:39 +01:00

I cannot confirm this issue on Windows, with a Geforce 540M.
The render on GPU equals the one on CPU. Lowering the World samples changes the result, so does disabling World MIS.

Can you maybe attach screenshots of the settings + render?

I cannot confirm this issue on Windows, with a Geforce 540M. The render on GPU equals the one on CPU. Lowering the World samples changes the result, so does disabling World MIS. Can you maybe attach screenshots of the settings + render?

Thomas Dinges commented

2016-01-20 17:45:15 +01:00

Note, that the Lamp samples only have an effect when "Sample All Lights (Direct and/or Indirect) is enabled. Otherwise it will just use 1 sample. That should probably be reflected in the UI.

Anyway, Sample All Lights Direct is enabled in the uploaded file, so it works as expected here. I didn't change settings (apart from Render Tile Size).

Note, that the Lamp samples only have an effect when "Sample All Lights (Direct and/or Indirect) is enabled. Otherwise it will just use 1 sample. That should probably be reflected in the UI. Anyway, Sample All Lights Direct is enabled in the uploaded file, so it works as expected here. I didn't change settings (apart from Render Tile Size).

blender-admin commented

2016-01-21 09:02:49 +01:00

This issue was referenced by adcef2bd36

This issue was referenced by adcef2bd3686a887fba8f08192751ab6736b9782

Thomas Dinges commented

2016-01-21 09:03:03 +01:00

Changed status from 'Open' to: 'Resolved'

Thomas Dinges closed this issue

2016-01-21 09:03:03 +01:00

Sebastian Koenig commented

2016-01-21 09:52:44 +01:00

I get the same problem on OSX 10.11.2 with a GTX980.
Rendering the file from this bugreport renders the same as with just 1 sample in world settings, even though it is set to 24.
Rendering the same file with 2.76 behaves as expected and renders with 24 world samples.

I get the same problem on OSX 10.11.2 with a GTX980. Rendering the file from this bugreport renders the same as with just 1 sample in world settings, even though it is set to 24. Rendering the same file with 2.76 behaves as expected and renders with 24 world samples.

Sebastian Koenig commented

2016-01-21 09:52:56 +01:00

Changed status from 'Resolved' to: 'Open'

Sebastian Koenig reopened this issue

2016-01-21 09:52:56 +01:00

Sebastian Koenig commented

2016-01-21 09:53:24 +01:00

Could it be because I compile with Cuda 7.5?

Edit: Ah, there we go, with buildbot build it works fine. Hm... I didn't remember getting the same issue before. Did something change there recently?
But I guess using Cuda 7.5 is my own risk... ;)

Could it be because I compile with Cuda 7.5? Edit: Ah, there we go, with buildbot build it works fine. Hm... I didn't remember getting the same issue before. Did something change there recently? But I guess using Cuda 7.5 is my own risk... ;)

Thomas Dinges commented

2016-01-21 10:55:29 +01:00

Since it works with official builds and CUDA 6.5, will triage this as Low priority. Still we should look into it at some point.

blender-admin commented

2016-01-24 12:48:49 +01:00

This issue was referenced by blender/cycles@2c5b6dc54b

This issue was referenced by blender/cycles@2c5b6dc54b044225d44d5d0aecd851d4376d51cf

Sergey Sharybin commented

2016-01-24 13:43:07 +01:00

Changed status from 'Open' to: 'Resolved'

Sergey Sharybin closed this issue

2016-01-24 13:43:07 +01:00

Thomas Dinges commented

2016-02-18 11:45:21 +01:00

Changed status from 'Resolved' to: 'Open'

Thomas Dinges reopened this issue

2016-02-18 11:45:21 +01:00

Thomas Dinges commented

2016-02-18 11:45:49 +01:00

Now that we are on Toolkit 7.5, this should be fixed.

Brecht Van Lommel commented

2016-02-18 20:30:24 +01:00

Added subscribers: @mib2berlin, @ditos

Brecht Van Lommel commented

2016-02-19 00:51:11 +01:00

Added subscribers: @Sergey, @brecht

Brecht Van Lommel commented

2016-02-19 00:51:11 +01:00

Hey @Sergey, I tried some random things and found a stupid workaround that seems to work here. I went ahead and committed it so that it's included in todays buildbot build for testing, but feel free to revert or modify it if you find something better or if it gives issues.

Sergey Sharybin commented

2016-02-19 09:54:06 +01:00

@brecht, interesting. I've tried same workaround initially and it didn't work. However, your commit seems to be fine here on GTX980 in the studio. Maybe it should be both noinline and int which i didn't test yesterday, hrrrm.

@sebastian_k, mind testing latest linux builds?

@brecht, interesting. I've tried same workaround initially and it didn't work. However, your commit seems to be fine here on GTX980 in the studio. Maybe it should be both `noinline` and `int` which i didn't test yesterday, hrrrm. @sebastian_k, mind testing latest linux builds?

Sebastian Koenig commented

2016-02-19 10:32:33 +01:00

That does indeed fix the issue! Plus, it renders faster :)
With own build (cuda 6.5) i get 9.9secs without sample all indirect, and 45.7 with sample all indirect.
With buildbot (cuda 7.5) i get 8.5secs without sample all indirect, and 41 sec with sample all indirect.
Tested rendering with gtx970, 980 and 980ti
So, yay!
Thanks @brecht!

That does indeed fix the issue! Plus, it renders faster :) With own build (cuda 6.5) i get 9.9secs without sample all indirect, and 45.7 with sample all indirect. With buildbot (cuda 7.5) i get 8.5secs without sample all indirect, and 41 sec with sample all indirect. Tested rendering with gtx970, 980 and 980ti So, yay! Thanks @brecht!

Sergey Sharybin commented

2016-02-19 10:43:44 +01:00

Changed status from 'Open' to: 'Resolved'

Sergey Sharybin closed this issue

2016-02-19 10:43:44 +01:00

Sergey Sharybin commented

2016-02-19 10:43:44 +01:00

Finally it's nailed down! Thanks for the report and help!

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

World Sampling in Cycles broken when rendering with GPU #47213