Denoiser CPU and GPU different results.

Use this forum for general user support and related questions.
Forum rules
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Denoiser CPU and GPU different results.

Post by Dez! »

Path CPU
Image

Path OpenCL (CPU+GPU and only GPU)
Image

Lux settings
Image
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Denoiser CPU and GPU different results.

Post by Dade »

Did you used the same amount of samples for the test ? The GPU rendering looks sharper/better but it could just due to the higher amount of samples per pixel.
Support LuxCoreRender project with salts and bounties
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

the settings are the same. I just switch OpenCL to CPU.
In compositing, if you look at the node render layer, the images in the channels are the same. The results differ only in the channels of the Denoiser.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Denoiser CPU and GPU different results.

Post by B.Y.O.B. »

Dez! wrote: Wed Oct 31, 2018 6:02 pm the settings are the same.
But you did not specify the "Samples" halt condition.
So it could be that they have a different amount of samples.
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

B.Y.O.B. wrote: Wed Oct 31, 2018 6:17 pm
Dez! wrote: Wed Oct 31, 2018 6:02 pm the settings are the same.
But you did not specify the "Samples" halt condition.
So it could be that they have a different amount of samples.
And why does the processor and video card have different understanding of the noise threshold?
Image

Yes, indeed, the processor makes fewer samples with the same amount of samples installed.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

With an equal number of samples from the CPU and the GPU, the result is the same. The picture is blurry again.
Before denoise pictures look the same.
This denoise works differently between the CPU and the GPU.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

Yes, I am researching now.
Denuser settings work differently for the CPU and the GPU.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

everything i found the reason. value Histogram Distance Treshold works differently.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Denoiser CPU and GPU different results.

Post by Dez! »

Working with a long animation, I noticed another problem.
If in OpenCL mode you disable the use of the processor, the denoiser may hang.
The system resource monitor at the same time shows 100% CPU usage. But in fact the system is not loaded. And it lasts forever. I tried to wait the whole night, but no, this is a hangup forever.
This does not happen all the time. I have not figured out what it depends on. But it happens quite often.
I could not render more than 100 frames in one approach, the deoiser will definitely hang.
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
kintuX
Posts: 809
Joined: Wed Jan 10, 2018 2:37 am

Re: Denoiser CPU and GPU different results.

Post by kintuX »

Am witnessing somewhat small inconsistencies in final result statistics, while comparing GPU vs CPU usage.
I assume since GPU does computing far faster using different packaging than CPU, results may vary. Difference is very small and not visible to the naked eye.

Also, a small error can escalate faster and turn into a bug, a halt or a crash. In my case, sometimes, when denoiser doesn't do it's job, the frame is saved black, but the rendering sequence continues. I never had a hanging situation with LuxCore. Since i got ECC memory on GPU and CPU... I wonder, if it could be it, but i don't know how to test the assumption. There are also many other factors that could sprout bad behavior (applications running in back, faulty hardware... as in one case i had network card polluting the system environment, got some freezing & hanging).

Would you care to make a study scene for testing purposes?
Post Reply