Diffuse Kernel Redundant?..

Newtek Lightwave 3D (exporter developed by holocube, Integrated Plugin developed by juanjgon)

Moderator: juanjgon

User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

Since Octane 2.0 I've been running comparisons between the Diffuse Kernel and Path Tracing. With the noise/firefly issue with the initial 2.0 release and subsequent fixes I've been keeping a close eye on the diffuse kernel to see how things have improved. To the credit of the Octane team the firefly issue seems to have been fixed but I can't help but notice that in every rendering situation I've tested, Path Tracing is faster than Diffuse. And now with GI Clamp built into PT I'm wondering what's the use of the Diffuse Kernel, it now seems redundant.

note: all testing done with the LW plugin which is why I'm posting it here. Will endeavor to test it with standalone asap.
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
geo_n
Licensed Customer
Posts: 350
Joined: Tue Feb 02, 2010 5:47 am

Pathracing is still 25% slower in my test.
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

hm.. will have to upload a sample scene
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

geo_n was your test in LW or standalone?
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
geo_n
Licensed Customer
Posts: 350
Joined: Tue Feb 02, 2010 5:47 am

Lightwave. Upload a sample scene.
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

The thing to remember when testing is the Diffuse depth should not be set the same for both. Diffuse Kernel requires less diffuse depth samples to achieve the same visual result. Here's my very basic and preliminary test. By that I mean it does not have any glass or refractions. It's the standard benchmark for LW. Here goes..

STANDARD LW BENCHMARK SCENE
DIFFUSE KERNEL
1000 Samples, Diffuse Depth 2, Specular Depth 4, Glossy Depth 4
bm test diffuse kernel.jpg
MS/sec = 19.20
Rendertime = 00:12


PATHTRACING KERNEL
1000 Samples, Diffuse Depth 3, Glossy Depth 4
bm test pt kernel.jpg
MS/sec = 19.14
Rendertime = 00:12

The difference is negligable so spec-wize I'd call it a draw. Visually though PT appears to resolve faster.
Last edited by MrFurious on Thu Aug 21, 2014 3:56 pm, edited 1 time in total.
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

Second test, my interior scene. I can't upload this scene but it'll do just to demonstrate.

TYPICAL INTERIOR SCENE
DIFFUSE KERNEL
1000 Samples, Diffuse Depth 3, Specular Depth 12, Glossy Depth 12
int scene diffuse.jpg
MS/sec = 2.86
Rendertime = 01:23

PATHTRACING KERNEL
1000 Samples, Diffuse Depth 4, Glossy Depth 12
int scene pathtracing.jpg
MS/sec = 2.86
Rendertime = 01:22

And the winner is.. PT! Only by a whisker but still faster. Whats more important here though is PT has clearly resolved better.

So..
faster (ok same speed more or less) by using more diffuse samples (?!@$?) and looks better.
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

Similar results in Standalone for the Benchmark Scene

STANDARD BENCHMARK SCENE OCTANE STANDALONE
DIFFUSE KERNEL
1000 Samples, Diffuse Depth 3, Specular Depth 4, Glossy Depth 4
ot std diff.jpg
MS/sec = 17.72
Rendertime = 00:34


STANDARD BENCHMARK SCENE OCTANE STANDALONE
DIFFUSE KERNEL
1000 Samples, Diffuse Depth 2, Specular Depth 4, Glossy Depth 4
ot std diff 2.jpg
MS/sec = 18.84
Rendertime = 00:32


STANDARD BENCHMARK SCENE OCTANE STANDALONE
PATHTRACING KERNEL
1000 Samples, Diffuse Depth 3, Glossy Depth 4
ot std pt.jpg
MS/sec = 18.80
Rendertime = 00:32

With the Diffuse Kernel at 2x diffuse samples it was the same rendertime as PathTracing at 3x diffuse samples. With the diffuse samples set the same PT was faster. I've compared apples with apples here just as with the LW test, tried to keep all other settings (other than the render kernel) the same. for some reason in standalone the PT test is a bit brighter, but once again it's faster than, or equal to diffuse.
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
geo_n
Licensed Customer
Posts: 350
Joined: Tue Feb 02, 2010 5:47 am

MrFurious wrote:The thing to remember when testing is the Diffuse depth should not be set the same for both. Diffuse Kernel requires less diffuse depth samples to achieve the same visual result. Here's my very basic and preliminary test. By that I mean it does not have any glass or refractions. It's the standard benchmark for LW. Here goes..

STANDARD LW BENCHMARK SCENE
DIFFUSE KERNEL
1000 Samples, Diffuse Depth 2, Specular Depth 4, Glossy Depth 4
bm test diffuse kernel.jpg
MS/sec = 19.20
Rendertime = 00:12


PATHTRACING KERNEL
1000 Samples, Diffuse Depth 3, Glossy Depth 4
bm test pt kernel.jpg
MS/sec = 19.14
Rendertime = 00:12

The difference is negligable so spec-wize I'd call it a draw. Visually though PT appears to resolve faster.

There's more noise I see in your pt render. Same thing I notice with mine. Needs more samples to get rid of noise which means more time.
For the interior let them resolve till there's very little noise and post render time.
User avatar
MrFurious
Licensed Customer
Posts: 529
Joined: Wed Oct 23, 2013 5:46 pm
Location: Melbourne, Australia

Are you sure have a closer look at the walls and in corner areas, its more course noise with Diffuse. Look at the rug, more detail can be seen in the PT render.

Will let it resolve more and see how it looks.
Dino Inglese
CG Artist
Melbourne Australia


Intel Core i7-4820K, 3x GTX 980ti
Windows 7 64bit, Modo 12.2v2 for PC
Octane build 4.04.0.145
Post Reply

Return to “Lightwave 3D”