Different Works in Progress

Post your tests, experiments and unfinished renderings here.
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Multicubes, animated camera

Post by FarbigeWelt »

Render time per frame approx. 6 s, 160 frames, resolution 500*500 pixels
openCL, 2 GPU (Hawaii)
Objects are cubes made from 1000 sub cubes.
Material is a mix from matte and glass with different colors.
Blender animation, key frames for the camera, extrapolation mode is linear.

Animated GIF
Animated-Cubes_500.gif
Still far away from real time rendering, I like the short render time and wonder what could be rendered with a similar resolution but with about 1 s per frame.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

openCL PGIC

Post by FarbigeWelt »

LuxCoreRender openCL PGIC 1500 samples 20m52s init 7m40s
LuxCoreRender openCL PGIC 1500 samples 20m52s init 7m40s
:D
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
Sharlybg
Donor
Donor
Posts: 3101
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: Different Works in Progress

Post by Sharlybg »

20min + 7 on opencl GPU :o
Wonder how bidir render will deal with that.
Last edited by Sharlybg on Fri Jun 07, 2019 10:19 am, edited 1 time in total.
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

Sharlybg wrote: Thu Jun 06, 2019 10:19 pm 20min + 7 on pence GPU :o
Wonder how bidir render will deal with that.
Actually ‚only‘ 20m52s in total. Init is a part of it.

Well, I can tell you this evening.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

Sharlybg wrote: Thu Jun 06, 2019 10:19 pm Wonder how bidir render will deal with that.
LuxCoreRender BiDir 1001 samples OIDN 0.95 2h09m CPU 6of8
LuxCoreRender BiDir 1001 samples OIDN 0.95 2h09m CPU 6of8
(Due to a strange log in screen behavior since a while I use only 6 of 8 cores for BiDir at the moment. It seems that if CPU is on full load my Windows 10 cannot show the log in box for the password anymore. I will try with 7 cores for future renderings.)
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
Sharlybg
Donor
Donor
Posts: 3101
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: Different Works in Progress

Post by Sharlybg »

20 mn vs 120 mn

r9 390X + r9 290X VS 6core 4770k

One GPu = 60 mn = 3X faster

Meaning that if you get a CPU 3 time faster you can beat the GPU or be on par with it. According to this cinebench the 1920X can do the job :

bench.jpg
1920X cost = 350 £ Vs Rx 580 (alias R9 290X alternativ) = 180 £

But you can't add more CPU easly.

BUUUUUUUT !
teaser_small.jpg

An improved Bidir path Guided engine can bring on the table something like 6X to 8X faster convergence. meanning that in this case :


I7 4770k 8core enable will perform like 2X Rx 580 (r9 390X).

And TR 1920X will be then 3X faster than a Single Vega 64 ( wich perform like 2 Rx 580/r9 390)
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Bidir path Guided engine

Post by FarbigeWelt »

Sharlybg wrote: Fri Jun 07, 2019 4:29 pm
An improved Bidir path Guided engine can bring on the table something like 6X to 8X faster convergence.
Dade, did you read that? Is there something going on behind the stage?

Very good news Sharlybg! This means there is big hope for BiDir's revival. I like that. As you know it is the only pure stuff.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Bidir path Guided engine

Post by Dade »

FarbigeWelt wrote: Fri Jun 07, 2019 6:05 pm Dade, did you read that? Is there something going on behind the stage?
The problem is BiDir is worth using when, and only when, Path tracing is not able to rendering something. The linked paper can make BiDir 8x faster but it doesn't matter because it will still be maaaaany time slower than Path+PGI+GPU.

You can simply simulate a comparison: render something with BiDir than cut the time by 4x-8x and than render the same scene with Path+PGI+GPU (try to use comparable CPUs/GPUs in term of cost). Who is faster ?
Support LuxCoreRender project with salts and bounties
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Bidir path Guided engine

Post by epilectrolytics »

Dade wrote: Fri Jun 07, 2019 7:08 pmThe linked paper can make BiDir 8x faster but it doesn't matter because it will still be maaaaany time slower than Path+PGI+GPU.
But PGI is biased and the caustic cache has problems.

BiDir is unbiased and has problems with SDS paths but it can render those from intersectible light sources, albeit very slowly.
A 4x-8x speed boost would definitely help with that.

Many users want Luxcore for unbiased rendering!
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Bidir path Guided engine

Post by FarbigeWelt »

Dade wrote: Fri Jun 07, 2019 7:08 pm You can simply simulate a comparison: render something with BiDir than cut the time by 4x-8x and than render the same scene with Path+PGI+GPU (try to use comparable CPUs/GPUs in term of cost). Who is faster ?
Well, if one looks at floating point operations per second and tasks which do not need complex forecasting algorithms, that means a second pipeline and therefore more die space, every simple and shorter pipeline is faster. GPU have lot of the same, simple pipeline. Their construct is always cheaper for simple tasks. If BiDir is more complex than UniDir, sure GPU is the winner. If smart but complex preprocessing helps to achieve similar pictures or at least satisfying results for a specific use then it is clear what approach is to be supported, i.e. where to put more effort.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
Post Reply