Cornell box testing, how to speed it up?
Cornell box testing, how to speed it up?
After a long time, I finally got some time to play with Kray (2.0) again.
To test it vs LW I made a Cornell box (only after to find out Matt made a way more "official" one a while ago.. doh ) But for some reason I can't get Kray to render faster than LW yet, which I would expect with some tweaking.. so perhaps one of you guys could help out perhaps?
So, for starters the render straight from LW native, in 2min 50 sec (i7 @ 3.5 ghz): with settings:
Followed by my current Kray render in 4 min 58s: with settings: And the used Photon Map, which rendered in 18 seconds:
Any pointers perhaps?
To test it vs LW I made a Cornell box (only after to find out Matt made a way more "official" one a while ago.. doh ) But for some reason I can't get Kray to render faster than LW yet, which I would expect with some tweaking.. so perhaps one of you guys could help out perhaps?
So, for starters the render straight from LW native, in 2min 50 sec (i7 @ 3.5 ghz): with settings:
Followed by my current Kray render in 4 min 58s: with settings: And the used Photon Map, which rendered in 18 seconds:
Any pointers perhaps?
Re: Cornell box testing, how to speed it up?
added to above: (I can't type well in long messages on forums, dunno why, some bug in my IE or so. Therefore I add it in new msg)
-I forgot to turn on "use transparancy" in LW for that render. It matters, but doesn't make that much of a difference. It goes from 2m 50s to 3m 10s. See: -Also, I couldn't get rid of the "caustics" like thing in Kray. Even turning off "caustics" in the main options panel or setting photons to 1. Not sure why that is. Doesn't matters rendertime wise though.
-I read the userguide front to back regarding these things, and tried and tweaked settings for about 6 hours total.
-If I use "passes=1", I get very ugly splotches, especially on the ball. Can't seem to get rid of them in any way except upping passes so it finds and refines them. Increasing rays or decreasing distance did not help much.
-The PM is shown "precomputed", and is made in Photonmap mode instead of lightmap. Since I have no lights and that mode gave bad results.
-The LW render has a bit more splotches under the ball and next to the right cube. The Kray one has the ball less smooth.
-Both renders, but the LW has it most pronounced, have a very weird "glitch"? going on in the dark shadow left of the cube on the red wall. Do you see how it becomes more bright there for some reason, with a sharp edge? odd..
-I forgot to turn on "use transparancy" in LW for that render. It matters, but doesn't make that much of a difference. It goes from 2m 50s to 3m 10s. See: -Also, I couldn't get rid of the "caustics" like thing in Kray. Even turning off "caustics" in the main options panel or setting photons to 1. Not sure why that is. Doesn't matters rendertime wise though.
-I read the userguide front to back regarding these things, and tried and tweaked settings for about 6 hours total.
-If I use "passes=1", I get very ugly splotches, especially on the ball. Can't seem to get rid of them in any way except upping passes so it finds and refines them. Increasing rays or decreasing distance did not help much.
-The PM is shown "precomputed", and is made in Photonmap mode instead of lightmap. Since I have no lights and that mode gave bad results.
-The LW render has a bit more splotches under the ball and next to the right cube. The Kray one has the ball less smooth.
-Both renders, but the LW has it most pronounced, have a very weird "glitch"? going on in the dark shadow left of the cube on the red wall. Do you see how it becomes more bright there for some reason, with a sharp edge? odd..
Re: Cornell box testing, how to speed it up?
Continuing with my monologue
Got rendertime down to 3m 39s, but setting min distance to 100%. It made hardly a difference (200% makes the touch shadows blur out though). So slowly getting there
On the whole I like the Kray render better, just wondering if more can be done.. Next step of testing will be much more blurred reflections in the scene and animation with moving objects, cached. Edit: just saw that that can't be done, animated and cached.. hmm, bummer. I hoped the time interpolated mode allowed such a thing in some way.
Got rendertime down to 3m 39s, but setting min distance to 100%. It made hardly a difference (200% makes the touch shadows blur out though). So slowly getting there
On the whole I like the Kray render better, just wondering if more can be done.. Next step of testing will be much more blurred reflections in the scene and animation with moving objects, cached. Edit: just saw that that can't be done, animated and cached.. hmm, bummer. I hoped the time interpolated mode allowed such a thing in some way.
Re: Cornell box testing, how to speed it up?
Is there a scene file?
- Janusz Biela
- Posts: 3265
- Joined: Mon Mar 13, 2006 10:39 am
- Location: Finland
- Contact:
Re: Cornell box testing, how to speed it up?
Please Post this scene on forum.
...but REMEMBER this is only syntetic TEST.
this is no speed test - this is test of GI.
For comparing You need something heavy: big scene ....
...but REMEMBER this is only syntetic TEST.
this is no speed test - this is test of GI.
For comparing You need something heavy: big scene ....
Re: Cornell box testing, how to speed it up?
ah yes, scene file, sorry. Will put it up when I'm home again.
Will move to a way heavier scene in a bit, I guess this one is too light to see much difference perhaps.
Will move to a way heavier scene in a bit, I guess this one is too light to see much difference perhaps.
Re: Cornell box testing, how to speed it up?
I agree with Johnny, not the best scene for performances
Re: Cornell box testing, how to speed it up?
edit:
Perhaps it is not the best scene indeed. But it is a simple one and good to start with I suppose.. Initially I just wanted to learn Kray (how PM's work etc) and tried using LW native to compare if I was getting decent rendertimes or needed more tweaking. I should have kept LW out at first and just asked my questions first, now LW renders are up front and my questions in the back and no one reads all that
Still those questions stand a bit, about the shading in the red, the caustics and for instance if it is common to see the splotches line up horizontally when using scanline? I use renderworm now and like it actually. Ideally there should be no splotches ofcourse, but when they align they show up faster I guess.. user error I'm sure.
Perhaps it is not the best scene indeed. But it is a simple one and good to start with I suppose.. Initially I just wanted to learn Kray (how PM's work etc) and tried using LW native to compare if I was getting decent rendertimes or needed more tweaking. I should have kept LW out at first and just asked my questions first, now LW renders are up front and my questions in the back and no one reads all that
Still those questions stand a bit, about the shading in the red, the caustics and for instance if it is common to see the splotches line up horizontally when using scanline? I use renderworm now and like it actually. Ideally there should be no splotches ofcourse, but when they align they show up faster I guess.. user error I'm sure.
Re: Cornell box testing, how to speed it up?
You should try with the scene in the tutorials. This test scene is too far from what you will do in Kray, so learning with it is really not the best idea.
Te type of light, quantity, number of polygons / photons is too far from what you'll do.
Believe me, when I started, I tried with really small scenes to render faster and I've wasted my time, I had to learn everything from the ground up when I was ready to try with Interior Architecture...
Hope you find answers anyway, not so much time today, sorrry.
Te type of light, quantity, number of polygons / photons is too far from what you'll do.
Believe me, when I started, I tried with really small scenes to render faster and I've wasted my time, I had to learn everything from the ground up when I was ready to try with Interior Architecture...
Hope you find answers anyway, not so much time today, sorrry.
- matt gorner
- Posts: 78
- Joined: Thu Apr 13, 2006 11:39 am
- Location: United Kingdom
- Contact:
Re: Cornell box testing, how to speed it up?
Thanks for the link again Matt.
The scene I used:
The scene I used:
- Janusz Biela
- Posts: 3265
- Joined: Mon Mar 13, 2006 10:39 am
- Location: Finland
- Contact:
Re: Cornell box testing, how to speed it up?
This is not good scene for test. Kray show much more speed in bigest scenes.
I will try prepare , maybe from evermotion.
In Cornell box is around 3 times faster then LW, but I know in big scenes with complex reflections, blurr etc this difference can be more bigger with more biger quality It is imposible compare LW engine render with Kray render becouse it is big tehnology space between both
Of coarse tonemping from Kray kill everthing around (I mean LWF, reinhard ) .It is perfect.
I will try prepare , maybe from evermotion.
In Cornell box is around 3 times faster then LW, but I know in big scenes with complex reflections, blurr etc this difference can be more bigger with more biger quality It is imposible compare LW engine render with Kray render becouse it is big tehnology space between both
Of coarse tonemping from Kray kill everthing around (I mean LWF, reinhard ) .It is perfect.
Re: Cornell box testing, how to speed it up?
Could you upload the scene that renders it in 53 seconds like you have there please?
Would be interested to see how you changed it.
It has more blurry contact shadows and some light leaks on the bottom of that right cube, but over all it looks fast and nice.
Would be interested to see how you changed it.
It has more blurry contact shadows and some light leaks on the bottom of that right cube, but over all it looks fast and nice.
-
- Posts: 152
- Joined: Sun Jan 22, 2006 5:22 pm
Re: Cornell box testing, how to speed it up?
wow, the Cornell box! I made this scene around 2001 (?) to test it in Arnold and Messiah:render. FPrime was not born yet, that was released March 2004. LW GI was very poor for this sort of scene until LW9.2 much much later.
In december 2005 I had a go with the just released kray 1.6 demo. Time flies when you are having fun
Will load this scene in kray 2.0 and see what happens, this one closely resembles the original Cornell photo, yes the original is a photo, not a render.
I made my scene from this data: http://www.graphics.cornell.edu/online/box/data.html
Now see kray 4 years ago!! time is on a AMD64x2 2.4 Ghz working on a single core, multithreading came in the summer of 2006 (after I whined to G for 6 months at least hehe)
In december 2005 I had a go with the just released kray 1.6 demo. Time flies when you are having fun
Will load this scene in kray 2.0 and see what happens, this one closely resembles the original Cornell photo, yes the original is a photo, not a render.
I made my scene from this data: http://www.graphics.cornell.edu/online/box/data.html
Now see kray 4 years ago!! time is on a AMD64x2 2.4 Ghz working on a single core, multithreading came in the summer of 2006 (after I whined to G for 6 months at least hehe)
Re: Cornell box testing, how to speed it up?
nice one. erwin's 'vintage render' is looking the best (most natural one) btw.
site: http://www.3dworks.com
reel: http://vimeo.com/threedeworks/demoreel
behance: https://www.behance.net/3dworks
reel: http://vimeo.com/threedeworks/demoreel
behance: https://www.behance.net/3dworks