Page 1 of 1

Maxwell 5 test

Posted: Thu Oct 31, 2019 9:11 am
by seghier
Hello, after testing V5 :
- Random color don't work with GPU or all procedural textures don't work with GPU
- Pymaxwell help is missing
- Embed materials from Web library don't work

Re: Maxwell 5 test

Posted: Thu Oct 31, 2019 12:29 pm
by Daniel Kerbler 20191031100659
yeah, procedurals are not yet supported on the gpu.

here you find a clear table of what features are still missing concerning CPU vs GPU

https://nextlimitsupport.atlassian.net/ ... GPU+engine

Re: Maxwell 5 test

Posted: Thu Oct 31, 2019 12:36 pm
by luis.hijarrubia
seghier wrote:
Thu Oct 31, 2019 9:11 am
Hello, after testing V5 :
- Random color don't work with GPU or all procedural textures don't work with GPU
- Pymaxwell help is missing
- Embed materials from Web library don't work
Procedurals are not supported on GPU.
Brany is working on it, I hope next patch.
We changed server and it seem that materials gallery is not working. IT is working on it, it should be possible to fix it from here without have to wait to a patch.

Re: Maxwell 5 test

Posted: Thu Oct 31, 2019 3:58 pm
by seghier
Thanks
Render region don't work , is this limitation of demo version?

Re: Maxwell 5 test

Posted: Thu Oct 31, 2019 5:34 pm
by seghier
Another problem i think is from v3 and still exist
In Local browser when choose outdoor ies folder studio blocked and force closed

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 12:14 am
by seghier
Test Glass , in GPU render some areas are dark compared to CPU render

CPU:

Image

GPU:

Image

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 12:49 am
by T0M0
raydepth limitation ?

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 1:43 am
by luis.hijarrubia
It can be. If my memory is correct GPU is limited to 18 bounces. This can be easily changed. We need to check performance. Anyway, dielectrics may have issues on GPU yet, so maybe it's not only the max bounce.

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 2:21 am
by T0M0
I was always curious how ray depth works in Maxwell (speaking of CPU engine).
Is there some value hardcoded for each ray type or maybe some "system" that automatically adjust ray depth based on energy of light reached after certain bounces ? ( i.e. stop computing if value reaches certain threshold )

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 4:16 am
by luis.hijarrubia
T0M0 wrote:
Sat Nov 02, 2019 2:21 am
I was always curious how ray depth works in Maxwell (speaking of CPU engine).
Is there some value hardcoded for each ray type or maybe some "system" that automatically adjust ray depth based on energy of light reached after certain bounces ? ( i.e. stop computing if value reaches certain threshold )
That code is really old, and I may be wrong. But as far as I know there is no max depth. If you hit sky is game over. If not we have some criteria, like amount of energy that is going to be give to the final pixel or roughness acumulated (try not to kill rays if they are bouncing on mirrors or through glass), and we do a random russian roullete (weaker rays have more chances to be killed). If you kill a ray you adjust the energy that you may loose, giving the other samples a little extra energy.

Is something like that, not very precise explanation, but you can make an idea with this. Maybe we need a place to talk with the most entuthiastic people in private :roll: I don't want to post in public all our secret souce :lol:

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 12:12 pm
by T0M0
luis.hijarrubia wrote:
Sat Nov 02, 2019 4:16 am
I don't want to post in public all our secret souce :lol:
I absolutely understand that :) Thank you for your time and answer Luis.

Re: Maxwell 5 test

Posted: Sat Nov 02, 2019 12:23 pm
by seghier
I test MERL Material (TableBRDF)
the render or preview is very slow ; there is this tool from Disney to preview BRDF materials and it is fast
https://www.disneyanimation.com/technology/brdf.html

https://nextlimitsupport.atlassian.net/ ... +TableBRDF
-----------
In Maxwell every change in material settings affects BRDF material , i think this must be added to the manual.
To show the real BRDF , reflactance 0/90 must be black and roughness 100

Image

Image

Re: Maxwell 5 test

Posted: Sun Nov 03, 2019 3:23 am
by bobross
What is all this "ray depth" talk? I thought MR is the "ground truth" render engine. Should it just not look right no matter what?

Re: Maxwell 5 test

Posted: Sun Nov 03, 2019 4:11 am
by luis.hijarrubia
bobross wrote:
Sun Nov 03, 2019 3:23 am
What is all this "ray depth" talk? I thought MR is the "ground truth" render engine. Should it just not look right no matter what?
Ground Truth (what in real world means we don't fake a lot of things others do) is true on CPU, that's why we don't have a maximum ray depth there. We have it on GPU. But we have to re-check that, make that number bigger or (if performance is not really affected) have no maximum ray depth on GPU either.

Re: Maxwell 5 test

Posted: Sun Nov 03, 2019 8:42 pm
by bobross
luis.hijarrubia wrote:
Sun Nov 03, 2019 4:11 am
bobross wrote:
Sun Nov 03, 2019 3:23 am
What is all this "ray depth" talk? I thought MR is the "ground truth" render engine. Should it just not look right no matter what?
Ground Truth (what in real world means we don't fake a lot of things others do) is true on CPU, that's why we don't have a maximum ray depth there. We have it on GPU. But we have to re-check that, make that number bigger or (if performance is not really affected) have no maximum ray depth on GPU either.
Ah, I understand now. Thank you!