Thea Render Revision 568

Development versions, documentation, tutorials and application resources may be found here.

Moderator: patricks

Locked
User avatar
notareal
Thea Support
Thea Support
Posts: 6558
Joined: Sat Jul 19, 2008 3:12 am
Location: Oulu, Finland

Did take a quick look to the scene. I have not tested this, but I think complex Crome objects may be explaining factor. They probably do generate challenging caustics to slow down render. Also use of instancing (or SU components, and exporting as intanced, if you use SU) could help too. Have you tried to illuminate scene with HDR?
Zacchia
Licensed User v1.5
Licensed User v1.5
Posts: 825
Joined: Tue May 12, 2009 7:11 am

notareal wrote:Also use of instancing (or SU components, and exporting as intanced, if you use SU) could help too. Have you tried to illuminate scene with HDR?
I allways export scenes from SU, never tried to export components as instances, will do that! Would be great if this speed up!!
Never illuminate scenes with HDR, do not know this... you mean IBL instead of sky??
Thanks for your interest!!
Windows 7 64 bit, Thea Render, SketchUp Pro
User avatar
patricks
Thea Technical Artist
Thea Technical Artist
Posts: 2375
Joined: Fri Jul 18, 2008 2:51 pm
Location: Germany & Palma de Mallorca/Spain

Zacchia wrote:
Zacchia wrote: version v1.0.10.505 rendertime with tr1 1745 seconds means 100%
version v1.1.01.568 rendertime with tr1 1589 seconds means 91%
or
version v1.0.10.505 rendertime with tr1 1745 seconds means 110%
version v1.1.01.568 rendertime with tr1 1589 seconds means 100%

version v1.0.10.505 rendertime with BSD 880 seconds means 100%
version v1.1.01.568 rendertime with BSD 710 seconds means 80%
or
version v1.0.10.505 rendertime with BSD 880 seconds means 123%
version v1.1.01.568 rendertime with BSD 710 seconds means 100%
This is the scene I made the test: http://dl.dropbox.com/u/38782963/Scene.pack.thea
This is the BSD Preset: http://dl.dropbox.com/u/38782963/12.%20 ... ryHigh.xml
This are the results:
http://dl.dropbox.com/u/38782963/v1.0.1 ... 12m25s.png
http://dl.dropbox.com/u/38782963/v1.1.0 ... 1_9m49.png
http://dl.dropbox.com/u/38782963/v1.0.1 ... 14m40s.png
http://dl.dropbox.com/u/38782963/v1.1.0 ... 11m50s.png
The TR1 renders have been limited to 10 passes

I have seen now that I made with the TR1 comparison a simple mistake: The seconds are not 1745/1589 they are 745/589 :crazy: That means the speed up is the same with TR1 and BSD, about 80/100% or 100/123%. Sorry for this!!
Hi Zacchia

how many threats are you using for rendering the scene ? and at what resolution (i see that the images you link to are smaller then the settings in the darkroom).

Now something about unbiased render times comparison .... the only "correct" way to compare them is to compare "noise free" renders ..... in other words ... how long does it take to get a clean image.
the reason for this is because unbiased renders don't clean up linearly in time ..... its more like this .... if a render is not clean after 30 minutes, then it can take 1 hour to get the noise away ... and so on.

so the only way to really see a speed difference is with noise free renders (or nearly noise free renders).

please try the "Jotero Cornell Box" from the Thea example scenes and see how long it takes to get it clean with the old Thea version and the new one (try TR1 or TR2 .... i guess that with TR1 it will be faster)

about BSD ... i need to know the render resolution you made the render with and how many threads ..... as here the new tracing depth per material will make a difference

Greetings Patrick
" Poets say science takes away from the beauty of the stars - mere globs of gas atoms.
I too, can see the stars on a desert night, and feel them. But do I see less or more ? "

Richard P. Feynman
User avatar
notareal
Thea Support
Thea Support
Posts: 6558
Joined: Sat Jul 19, 2008 3:12 am
Location: Oulu, Finland

notareal wrote:Did take a quick look to the scene. I have not tested this, but I think complex Crome objects may be explaining factor. They probably do generate challenging caustics to slow down render. Also use of instancing (or SU components, and exporting as intanced, if you use SU) could help too. Have you tried to illuminate scene with HDR?
Actually now thinking, caustics are not probably the explaining factor... but anyhow I can confirm that with this particular scene the speed increase is just about 20% (using exactly same settings and 10 pass). As in my test the render time was relatively short, it's hard to tell if there would be greater difference in noise if RV568 scene where rendered long enough it to be noiseless vs RV505 noise. Saved file size after 10 pass was almost identical (png).
User avatar
patricks
Thea Technical Artist
Thea Technical Artist
Posts: 2375
Joined: Fri Jul 18, 2008 2:51 pm
Location: Germany & Palma de Mallorca/Spain

Unbiased renders clean up very quick in the beginning .... so its normal that a "short" render will not show much difference. Its on long render times where the difference is much more evident.

Greetings Patrick
" Poets say science takes away from the beauty of the stars - mere globs of gas atoms.
I too, can see the stars on a desert night, and feel them. But do I see less or more ? "

Richard P. Feynman
User avatar
notareal
Thea Support
Thea Support
Posts: 6558
Joined: Sat Jul 19, 2008 3:12 am
Location: Oulu, Finland

patricks wrote:Unbiased renders clean up very quick in the beginning .... so its normal that a "short" render will not show much difference. Its on long render times where the difference is much more evident.

Greetings Patrick
True, perhaps if DallasRT had time to run some longer tests with his monster machine...
btw noticed that most materials had no reflectance set. Wonder if it can affect?
User avatar
patricks
Thea Technical Artist
Thea Technical Artist
Posts: 2375
Joined: Fri Jul 18, 2008 2:51 pm
Location: Germany & Palma de Mallorca/Spain

a big PC is not needed ... just a "normal" scene that would normally take 2 or 4 hours .... it should get to less nosy result in 1 or 2 hours ....

simple scenes or short render times will not show the speed increase that much. that's why i suggested to try the Jotero Cornell Box scene from Thea Example scenes.

But there is something Zacchia say that makes me wonder ... the CPU only get loaded 80% and that's strange as i get always a 99 - 100% load ... so here we could have something else going on too ...

Greetings Patrick
" Poets say science takes away from the beauty of the stars - mere globs of gas atoms.
I too, can see the stars on a desert night, and feel them. But do I see less or more ? "

Richard P. Feynman
ThomasHelzle
Licensed User v1.5
Licensed User v1.5
Posts: 642
Joined: Wed Feb 16, 2011 2:03 pm
Location: Berlin
Contact:

I had the same - are there Render Channels active?

Cheers,

Thomas
User avatar
DallasRT
Licensed User v1.5
Licensed User v1.5
Posts: 1147
Joined: Fri Jul 09, 2010 7:12 am
Location: Carrollton, Texas

notareal wrote:
patricks wrote:Unbiased renders clean up very quick in the beginning .... so its normal that a "short" render will not show much difference. Its on long render times where the difference is much more evident.

Greetings Patrick
True, perhaps if DallasRT had time to run some longer tests with his monster machine...
btw noticed that most materials had no reflectance set. Wonder if it can affect?

If me and my "monster" machine can help I am here. ;)

EDIT: 09/23/2011 18:45

I ran the scene with his preset and it took my machine 2.38 minutes to render out. I watched my CPU usage and it was at a constant 100% The scene itself used about 2 gig of ram.


EDIT: 09/23/2011 18:53

Running the JoteroCornellBox and I am watching my CPU fluctuate between 70% and 90%
Xeon 5600 Series 4 Twelve Core Processor
192GB DDR3 1333 MHz
4 - Quadro 6000
User avatar
DallasRT
Licensed User v1.5
Licensed User v1.5
Posts: 1147
Joined: Fri Jul 09, 2010 7:12 am
Location: Carrollton, Texas

I did the render of the JoteroCornellBox. Processor flutuates between 70% and 96% tops. Memory was 1.8k

TR1 129 passes 8 min 48 sec

Image
Xeon 5600 Series 4 Twelve Core Processor
192GB DDR3 1333 MHz
4 - Quadro 6000
Locked

Return to “Versions & Resources (Studio)”