LuxCore: issues, problems, regressions

Discussion related to the implementation of new features & algorithms to the Core Engine.

Moderators: Dade, jromang, tomb, zcott, coordinators

Re: LuxCore: issues, problems, regressions

Postby kalel » Mon Oct 24, 2016 1:29 pm

Piita wrote:This glass of water does not look right in luxcore. classic api is ok with the exact same setup, I only changed render api. Kalel's Oct 15 build
.blend file attached.


When I simply change to Classic API, I see this:
blender_2016-10-24_21-38-52.png


Could it be related to the difference?
kalel
 
Posts: 258
Joined: Sun Jul 17, 2016 6:35 am

Re: LuxCore: issues, problems, regressions

Postby jensverwiebe » Mon Oct 24, 2016 1:39 pm

Pita, change the normaldirection of your water/glass mat, then it looks right,
You did not used the new volumeprecedence system and built model the old way,
so the precedence cannot take effect.

Jens
User avatar
jensverwiebe
Developer
 
Posts: 3429
Joined: Wed Apr 02, 2008 4:34 pm

Re: LuxCore: issues, problems, regressions

Postby Piita » Tue Oct 25, 2016 10:24 am

Sorry but flipping the normals makes it still look wrong on my end.

luxcore flipped normals for water-glass material.png




kalel wrote:When I simply change to Classic API, I see this
Could it be related to the difference?


It shouldn't. I noticed the color difference earlier but that was no big deal, the render was good enough to show the difference in refraction between the two.
Last edited by Piita on Wed Oct 26, 2016 10:13 am, edited 1 time in total.
User avatar
Piita
 
Posts: 611
Joined: Sat Aug 06, 2011 2:09 pm
Location: Finland

Re: LuxCore: issues, problems, regressions

Postby B.Y.O.B. » Tue Oct 25, 2016 10:42 am

Your setup looks correct to me (even the normals), no idea why it's not working. :|
User avatar
B.Y.O.B.
Developer
 
Posts: 5180
Joined: Wed Nov 10, 2010 4:10 pm
Location: Germany

Re: LuxCore: issues, problems, regressions

Postby B.Y.O.B. » Tue Oct 25, 2016 11:25 am

I modeled a complete new scene from scratch and I get similar results.
I included also a new-style glass model (with volume precedence) which looks correct in LuxCore.

Left glass is old-style, right one uses new-style volume precedence.
The lighting is different because in Classic apparently it does not work to set both efficacy and power to 0 to control the light via gain only and get reasonable brightness, so I just hacked it together.
Attachments
classic.jpg
Classic
luxcore.jpg
LuxCore
glass_01.zip
(243.8 KiB) Downloaded 30 times
User avatar
B.Y.O.B.
Developer
 
Posts: 5180
Joined: Wed Nov 10, 2010 4:10 pm
Location: Germany

Re: LuxCore: issues, problems, regressions

Postby kalel » Tue Oct 25, 2016 11:41 am

I love how those glasses look. :) Basically a simple model, and such great results.

Edit: Just for comparison, here's what LuxCore Microkernel BiasPathOCL gives on the last scene (some clamping is on, so colors/brightness may differ, resolution also dec from 60 to 50% as path is in a serious disadvantage to bidir)
Attachments
biaspathglass.png
kalel
 
Posts: 258
Joined: Sun Jul 17, 2016 6:35 am

Re: LuxCore: issues, problems, regressions

Postby burnin » Tue Oct 25, 2016 7:18 pm

Using same scene on win7 with (kalel's latest build) i got an error every time setting Max Depth above 51 with Path OpenCL:
- CPU:
Code: Select all
The SampleData buffer is too big for Intel(R) Xeon(R) CPU           X5650  @ 2.67GHz Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=4091342848): try to reduce related parameters

- GPU1:
Code: Select all
The SampleData buffer is too big for Quadro M5000 Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=2013265920): try to reduce related parameters

- GPU2:
Code: Select all
The SampleData buffer is too big for Quadro 600 Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=268435456): try to reduce related parameters


Is it a bug? Because Biased Path works even if i set some high values.
ie.
Sampling - AA Samples: 64 (D:4, G:4, S:4) / Path depths - Max Tot. Depth: 128 (D:128, G:128, S:128)
Sample Estimates / AA: 4096 / Diffuse, Glossy, Specular: 65536 (with max. bounces 127: 8323072) / Total on Diffuse surfaces: 16580608
burnin
 
Posts: 287
Joined: Mon May 03, 2010 8:04 pm

Re: LuxCore: issues, problems, regressions

Postby neo2068 » Tue Oct 25, 2016 11:01 pm

burnin wrote:Using same scene on win7 with (kalel's latest build) i got an error every time setting Max Depth above 51 with Path OpenCL:
- CPU:
Code: Select all
The SampleData buffer is too big for Intel(R) Xeon(R) CPU           X5650  @ 2.67GHz Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=4091342848): try to reduce related parameters

- GPU1:
Code: Select all
The SampleData buffer is too big for Quadro M5000 Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=2013265920): try to reduce related parameters

- GPU2:
Code: Select all
The SampleData buffer is too big for Quadro 600 Intersect device (i.e. CL_DEVICE_MAX_MEM_ALLOC_SIZE=268435456): try to reduce related parameters


Is it a bug? Because Biased Path works even if i set some high values.
ie.
Sampling - AA Samples: 64 (D:4, G:4, S:4) / Path depths - Max Tot. Depth: 128 (D:128, G:128, S:128)
Sample Estimates / AA: 4096 / Diffuse, Glossy, Specular: 65536 (with max. bounces 127: 8323072) / Total on Diffuse surfaces: 16580608


Biased path is a tile based render engine. I think, it only stores the data for the current rendered tile in GPU memory. For path renderer the data for the whole image has to be stored in GPU memory. Dade, correct me if I am wrong.
i7 5820K, 32 GB RAM, NVIDIA Geforce GTX 780 + GTX 560, Windows 10 64bit, Blender 2.77a
neo2068
Developer
 
Posts: 742
Joined: Sun May 03, 2009 2:11 am
Location: Germany

Re: LuxCore: issues, problems, regressions

Postby kalel » Wed Oct 26, 2016 2:44 am

Biaspath definitely uses less memory, and the most affecting factors are tile resolution and AA samples.

I would also like to add that with Metropolis, there is a larger memory requirement than for e.g. Sobol when increasing path depths. That is, if you set a large path depth on Metropolis you need a large amount of memory, where with Sobol you can set anything you like without impact, from my experience, I can't claim anything as I'm not sure how they work internally. I think that in biaspath also path depth doesn't have much effect on memory (likely related to sampler used), but AA settings do affect memory usage and GPU saturation a lot, still if the tile size is smaller, you have more space to increase this.

So, it's worth also trying Sobol if you want a large path depth without too much memory consumption, to avoid such errors.

Another thing you can do to reduce memory on PathOCL regardless of sampler is send less data to each device. This can slow things down less or more depending on device, but will reduce memory usage, sometimes significantly.

In the custom luxcore commands field, write:
opencl.task.count = 64000
If still an error, try:
opencl.task.count=32000
kalel
 
Posts: 258
Joined: Sun Jul 17, 2016 6:35 am

Re: LuxCore: issues, problems, regressions

Postby Dade » Wed Oct 26, 2016 6:48 am

All correct, the PathOCL memory usage scale up with image resolution and, when Metropolis is used, also with max. path depth.

Memory usage is now the main reason to use BiasPathOCL over PathOCL.
User avatar
Dade
Developer
 
Posts: 8404
Joined: Sat Apr 19, 2008 6:04 pm
Location: Italy

PreviousNext

Return to Architecture & Design

Who is online

Users browsing this forum: No registered users and 2 guests