Mix shader layers in H16.5 slows render considerably is it normal behavior?

Forums Technical Discussion Mix shader layers in H16.5 slows render considerably is it normal behavior?

3271 18 1


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 18, 2018 5:03 p.m.

The question on the title.

I am using build 439.

Thanks.

Edited by mzigaib - May 18, 2018 17:19:11

https://vimeo.com/user2163076 [vimeo.com]


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 18, 2018 7:55 p.m.

Anyone?

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 18, 2018 8:13 p.m.

Was your material doing displacement before? Displacement tends to flip on when combining vops if you don't use the render property ‘shop_disable_displace_shader’ to kill it.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 19, 2018 1:47 p.m.

Thanks for the feedback.

No I am not doing any displacements, the simple fact that you mix 2 shaders in a clean scene with a shaderball and render it you can already notice the slow down.

I was wondering if anyone else noticed it and if this behavior is normal.

Edited by mzigaib - May 19, 2018 13:48:31

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 19, 2018 8:49 p.m.

I'm not seeing a large difference in shader performance between layered and non-layered. The layered render was about 10 seconds slower than the slowest layer. The three renders took 1:10 1:20 and 1:30, with the layered render taking the longest.

I would expect some difference considering the layered shader is more complicated than the non-layered one.

Attachments:
shader_ball.0006.jpg (431.4 KB)
shader_ball.0007.jpg (398.0 KB)
shader_ball.0008.jpg (442.4 KB)
shaderball_layers_perftest_example.zip (85.2 KB)


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 20, 2018 1:37 a.m.

Ok cool maybe it's just me I'll test again to be sure, thanks for the feeback.

https://vimeo.com/user2163076 [vimeo.com]


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 21, 2018 2:49 p.m.

After another test I am still getting the issue.

1-Mix shader ball - 2:45min
2-Just Material 1 shader - aluminium - 1:13min
3-Just Material 2 shader - Red coated - 1:10min

I've mix them with a sim x axis ramp no doubt if I use the mix shader it doubles the time even with similar materials.

And I made this test in a different machine.

Edited by mzigaib - May 21, 2018 14:51:06

Attachments:
mixed_shader.jpg (197.2 KB)
mat_1_shader.jpg (190.1 KB)
mat_2_shader.jpg (198.7 KB)
mix_shader_test.hip (3.0 MB)

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 21, 2018 3:10 p.m.

Your mix puts both materials everywhere, it would follow that the result takes as longer than with just one or the other. I was expecting a more extreme difference from your first post, like 10 minutes compared to one. Taking twice as long is well within the range of expected change when doubling shading complexity.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 21, 2018 3:23 p.m.

In fact the lone shaders examples are everywhere theoretically they should take longer, but I think it makes sense when you say that the double time is related the shader mixing complexity.

I think doubling the render time can be extreme depending on the complexity of the object or the shader now I am going to think more before mixing them.

Thanks for the feedback.

Edited by mzigaib - May 21, 2018 15:24:07

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 21, 2018 6:42 p.m.

There must be something up with the way you are mixing the layers. I redid the test with a ramp between two shaders, and I get the same time either way.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 21, 2018 9:43 p.m.

Really? I did it in the most simple way possible em 2 different machines, curious.

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 22, 2018 12:22 a.m.

Make sure you always add ‘shop_disable_displace_shader’ to the final vop if you are not building a shader with displacement and not using a collect vop/material builder. Shading networks that are ‘incomplete’ are automatically completed with both a surface and displacement shader. This might be part of the difference you are seeing.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 22, 2018 10:51 a.m.

Even with properties set I get the same thing.
The only difference between shaders are the colors.

Edited by mzigaib - May 22, 2018 10:53:27

Attachments:
net.jpg (110.2 KB)
compare_1.jpg (276.5 KB)

https://vimeo.com/user2163076 [vimeo.com]


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 28, 2018 6:20 p.m.

So no one with the same issue? Just me?

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 28, 2018 7:22 p.m.

Are you able to post an example scene? In my simplistic tests, I was unable to recreate the same amount of time discrepancy as you experienced.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 29, 2018 10:41 a.m.

There is the file.

Attachments:
mix_shader_test.hip (3.3 MB)

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 29, 2018 2:39 p.m.

You had ‘Add Baking Exports to Shader Layers’ enabled. This is very expensive when there are multiple layers.


mzigaib: Member; 948 posts; Joined: April 2008; Offline

May 29, 2018 5:02 p.m.

Really? I am going to test that.
I always put this parameter to export all my AOVs, without it some of them doesn't show in the channels.

Thanks for the feedback I'll take a look on that.

https://vimeo.com/user2163076 [vimeo.com]


jsmack: Member; 7759 posts; Joined: Sept. 2011; Offline

May 29, 2018 8:10 p.m.

Unchecking it in your scene dropped the render time from 1:15 to 0:22. I wouldn't expect it to be quite so costly, but it does have to do a lot of extra compositing for each sample.

Quick Links

                    
                        Search links
                        Show recent posts
                        Show unanswered posts