PDA

View Full Version : Increased render time in 10.1, compared to 9.6 (w/ example attached)



cptwhite
10-28-2011, 06:41 AM
Hi all,

Just a quick question regarding render times. I've got a basic scene that was created in 9.6, it has radiosity enabled and rendered in 22.4s.

I've just opened the same scene in 10.1 and the scene is now taking 64s to render.

The change is entirely down to increased time in radiosity rendering (~50s in 10.1 as opposed to 15s in 9.6). It is using exactly the same GI, camera, lighting and material settings so I'm stumped as to what could be causing such a dramatic increase in rendering. The final rendered image is on first inspection identical so there's no major changes going on behind the scenes giving improved output results either.

Caching is disabled in both accounts.

Obviously I want to be using 10.1 but I need to understand what is causing this increased render time.

LW9.6 set-up notes and render:
http://i72.photobucket.com/albums/i172/cptwhite/LW9_6.png

LW10.1 set-up notes and render:
http://i72.photobucket.com/albums/i172/cptwhite/LW10_1.png

The scene and objects can be downloaded here:
http://www.mediafire.com/?pe2acw3vs5cvh4p

If anyone can shed any light onto what is happening here I would greatly appreciate it. I assume I'm overlooking some setting but I've exhausted where else I can think to look.

erikals
10-28-2011, 07:43 AM
sorry, couldn't find anything... :\

cptwhite
10-28-2011, 07:52 AM
Thanks for trying anyway :) Hopefully someone can solve this mystery :D

JustForFun
10-28-2011, 08:04 AM
I noticed the same rendering time increase when I moved to 10.x.

Sensei
10-28-2011, 10:31 AM
Is your LW v9.6 32 bit and LW v10.1 64 bit?
If so, use the same architecture version.
32 bit is FASTER, as long as no virtual memory is used.

Danner
10-28-2011, 06:41 PM
I downloaded your scene, it asks me for two plugs that I don't have, an ambient occussion node and an edge something. I didn't add them just ignored them, and guess what? LW 10 rendered faster..


both 64bits (ignore the color space difference I didn't want to mess up my configuration so I didn't move any CS setting on lw10)

erikals
10-28-2011, 06:45 PM
guess it's a plug-bug :]

cagey5
10-28-2011, 09:33 PM
Looks like the g.i. render time is longer though at 3.3 secs v 2.1 seconds. So about 50% longer. On a slower computer / more complicated scene that could still make a big difference.

cptwhite
10-31-2011, 04:05 AM
Just so you're aware my machine is a Q6600 @ 3Ghz, 4Gb RAM running at DDR800. I'm running Windows XP 32bit (don't laugh!)

Both my tests were run on the same machine, with plugins applied to both versions of LW.

Danner's test has the radiosity set to background only, hence the fast render times.

Danner
10-31-2011, 04:43 AM
that's how the scene was set up...

@Cagey
Yes I noticed that the total time was lower but radiosity calculations took longer. That is not always a bad thing, in my type of work (arch viz) that would mean lower render times in total because I use cached radiosity.

cptwhite
10-31-2011, 04:47 AM
Going to do some further tests without the plugins - will report back :)

lardbros
10-31-2011, 06:50 AM
Haven't had a chance to look at the scene files, but from the renders (not sure if it's an optical illusion) the 10.1 render looks slightly darker, and the lighting on the end at the top of the object looks more dull.

If you're using Linear Colour Space, have you checked that the HDR's are looking the same in the image viewer in both versions of LW?? I've had issues where HDR's from Dosch Design aren't actually linear, so I've had to linearize them in 10.1 before rendering! Nightmare!


EDIT:
----------
Sorry, just looked and it says you are using backdrop radiosity anyway (oh no you're not, someone else has changed that in their tests)! Sorry! Worth checking the HDR's then! :D

cptwhite
10-31-2011, 08:10 AM
lardbros:
I have no experience of HDR (I guess I'm a noobie in the whole field generally). Although from reading around the subject last week I understand HDR images used as a backdrop capture "intensity" values along with the raw image, which can be used to create natural hotspots when incorporated into a scene (in effect acting as an area light with variable intensity at points across the image)

However this scene doesn't incorpoate any HDR imagery, so I'm not sure if your point is still relevant?

Again I don't really have an understanding of what a linear colour space is right now, so i've no idea if I'm using it or not (is it something to do with gamma correction?)

This scene is effectively a metal pole sat in a white room (one of the plugins overrides the surface to pure white unaffected by lighting conditions, this is used on the room). The room is made visible to radiosity to get light bouncing around the metal pole but invisible to the camera so it isn't rendered. The pole is then sat on a simple floor polygon which is visible to the camera.

Radiosity should be Monte Carlo (attached scene is incorrectly set to Background)

Finally you are correct - the image rendered in LW10.1 is slightly darker (around 7-10 points in all three RGB channels)

erikals
10-31-2011, 09:11 AM
gamma / LCS
http://forums.newtek.com/showthread.php?t=122307

cptwhite
11-03-2011, 05:10 AM
Just a quick update on this. It seems my test was somewhat flawed (some of the materials weren't set-up exactly the same when I ran these tests in my original post, due to shaders been enabled in one of the tests and not in another).

Also when importing materials created in 9.6 into 10.1 the "exclude from vstack" was automatically enabled in 10.1 (whereas it wasn't in 9.6). Not sure if this made any significant different as I'm not clear as to what this does, but it was another difference I overlooked.

I was using the SG_AmbOcc_Exp plugin (set to surface override) for the room walls to give a high level of light bouncing with radiosity. Disabling this in 9.6 seemed to have no effect to the render curiously (bug?) but behaved as you would expect in 10.1 giving a much darker render as per Danner's test above

With the surfaces corrected the render times *are* a lot closer (albeit the final render has a very different light intensity).

However it does still seem the GI render time in 10.1 is around 30-50% longer than in 9.6.

I'm going to rebuild the scenes in both versions to ensure I've not overlooked something and post the results.