PDA

View Full Version : Why did this take over a day to render?



squeakybadger
08-14-2005, 12:13 PM
32 Hours 16 Minutes to be precise!

Its not exactly a complex scene so im quite baffled as to why it took so long (running on a G5 w/ 1GB of Ram)

These are the settings i used, can anyone offer up some tweaks i can fiddle with to make it shuffle along faster? This was only supposed to be a test render :D


Camera Settings

840x467
AA: Classic Low

GI Settings

Shading Noise Reduction
Radiosity Enabled:
Interpolated
100% Intensity
4x12 Rays per Eval
1 Indirect Bounce
0 Tolerance
18mm Min Eval Spacing

Render Options

Realistic Render Mode
Ray Trace Shadows/Reflection/transparency enabled
Extra Ray Trace Optimization enabled
16 Ray Recursion Limit
1 thread Multithreading


Cheers for any info fellas

Tiger
08-14-2005, 01:05 PM
My guess would be the glasstable. Radiosity and glass=looong rendertimes.
But I could be wrong, maybe it`s something else.

What kind of light did you use?
I really think you should consider F-Prime.
www.worley.com

gjjackson
08-14-2005, 01:15 PM
16 ray recursion is overkill. Lower that and it shouldn't take so long.

squeakybadger
08-14-2005, 01:23 PM
Tiger: Its just 1 area light outside the window. And believe me, if i had a magical cash cow, fprime would be first on my list of things to buy :)

Gj: I think that was the default setting, i have no idea what it does, but ill lower it next time :D

toby
08-14-2005, 01:52 PM
If you have a Dual G5, Multi-thread with 2 to 8 threads ( do test renders to see what's fastest ), if you have a single G5, that explains a lot!

Interpolated radiosity set to 0 tolerance will take just as long as Monte Carlo ( but tolerance set between 0 and .2 will take longer than MC, and look worse ) -

Start with a ray recursion of 4, then raise it if neccessary, 16 is never needed -

Set the Area light quality to 3 for test renders - 5 is achingly slow -

Use limited region to test render times -

If you're rendering stills, there's a technique to get really fast renders using Interpolated radiosity with motion blur. Set the tolerance to .232, eval spacing to about 1 m, set rays high like 7x21, turn on Normal motion blur, and use high or extreme AA. You have to turn off Adaptive sampling too.

You're setting the radiosity to give you fast but blotchy results, and the motion blur blends all the blotchy passes together. The more passes, the more accurate the final result - you'll have to adjust the settings for your particular scene.

Captain Obvious
08-14-2005, 03:05 PM
Never set the tolerance of interpolated radiosity to zero. It will look just like monte carlo, but takes much longer to render. Set it to a non-zero value and render with motion blur activated and you'll get much better results. Faster, too.

tektonik
08-14-2005, 07:00 PM
buy fprime and divide rendertime by 20

a+

toby
08-14-2005, 07:47 PM
Gj: I think that was the default setting, i have no idea what it does, but ill lower it next time :D
The ray recursion is how many times the rays will bounce off reflective surfaces or pass through transparent surfaces. For example if you have two mirrors facing each other and a red ball inbetween them, you will see as many reflections of the ball as you have ray recursions, give or take -

squeakybadger
08-15-2005, 01:38 AM
Thanks for all the replys fellas, majorly helpful as usual.

I take it that fprime is the new jesus then? :D

tektonik
08-16-2005, 06:19 AM
amen to that !

but it seems a bit pricey now (fprime) compared to the price of lightwave itself :stumped: especially with no network rendering...

Darth Mole
08-16-2005, 03:51 PM
With the right scenes (which is pretty much most of them), FPrime is like buying a whole renderfarm of computers.

With the image above, I reckon you could get the same result - or better - in an hour or two. Possibly much less.

Post the scene file and I'll give it a go!

Bytehawk
08-16-2005, 06:07 PM
off topic but hey, it's relevant

peeps looking at a big screen like that sitting so close will get motion sickness within 2 minutes. Believe me i know. Recently i visited a guy with a plasma screen mounted that close - i tell ya, it's not funny - .

Dave Jerrard
08-16-2005, 08:21 PM
Try this.

Increase the AA to something like medium or high. Stay with me here....

Now, lower the Radiosity level to 1x3 or 2x6. This will greatly speed up the rendering, but will be grainier. However,the higher AA level will smooth that out and give you a similar quality render, if not better, with a much faster render time. Also, kick that area light down to a quality of 3, or maybe even 2 if you're using a high enough AA. Finally, just use Monte Carlo.

And you really don't need that Shading Noise Reduction turned on. That will slow things down a bit and usually leaves the image looking a bit smudgy looking. It's a small render hit - about ten seconds per pass or more - but with higher AA, you really don't need it.

Using these tricks, I've taken images that wer taking over 12 hours dwon to 3 hours, and that was a scene that used four area lights.

You might even try using Ray Trace Transparency if you're not already. That's been known to speed things up nicely too.

And don't mess around with ray recursion until you know how it works. If you must, then don't go too low. Lower values start to introduce errors (no reflections, can't see through transparency), so only lower it a bit. I almost never go below 8. I've had to fix too many scenes that weren't working because someone turned this down because they hear it's faster.


He Who Has A Hard Time Getting Renders Over 10 hours Now.

toby
08-16-2005, 10:57 PM
And don't mess around with ray recursion until you know how it works. If you must, then don't go too low. Lower values start to introduce errors (no reflections, can't see through transparency), so only lower it a bit. I almost never go below 8. I've had to fix too many scenes that weren't working because someone turned this down because they hear it's faster.
How's he going to know how it works unless he tries it? Where I work the maximum is 4 until we know we need more. I'm working on a headlight render now that needs 7 to look good only because they get their distinct look from multiple bounces off chrome behind a glass cover. 7 takes more than twice as long as 4. I've yet to see a scene that really needed 16, including this headlight, so I'm betting he's losing a lot of time for nothing by leaving it set that high. Besides, test renders set too low quickly become obvious and render faster.

tischbein3
08-17-2005, 07:30 AM
Agree on this, also it may look even better and less distorted using less bounces. (Although less than 2/3 really could produce to some trouble / unexpected behaviour. )

I usually put all my glass stuff in seperate layers and exclude it from radiosity and most of the lights / cast shaddows (Sometimes even "unseen by rays activated", but I think in your case the visibility of the inner side of the glass is essential....)

Also looking at the image it simply cries for z-order errors.
If so, a subdivision of the ground / slightly rise of the objects could help fixing this and decrease time.

Celshader
08-17-2005, 10:18 AM
SqueakyBadger, try Dave Jerrard's advice before you muck around with ray recursion. The man's used LightWave professionally for 15 years (http://www.lwg3d.org/v3/interview.php?id=6), and when it comes to rendering, he knows where his towel is.

toby
08-17-2005, 12:10 PM
:rolleyes: Whatever.

Nothing at Digital Domain gets rendered with 16 ray recursions. Anyone who's raytracing reflection at 16 without testing to see if he needs it is wasting render time for nothing.

If you try it at 4 and it ruins your life, well you can just turn it back up, right? And you've learned something in the process.

Celshader
08-17-2005, 12:19 PM
:rolleyes: Whatever.

Nothing at Digital Domain gets rendered with 16 ray recursions. Anyone who's raytracing reflection at 16 without testing to see if he needs it is wasting render time for nothing.

If you try it at 4 and it ruins your life, well you can just turn it back up, right? And you've learned something in the process.

Hey, Dave once advised me to knock ray recursion down to 2 for a specific scene. He's not against lowering ray recursions...but it's not his first choice for optimizing a scene.

I also don't recommend learning ray recursion on a scene that "normally" takes 32+ hours to render. To specifically learn ray recursion, SqueakyBadger is better off experimenting with ray recursion on something that renders far quicker and really emphasizes the effects of ray bounces, like the RayTrace benchmark scene. That way he can better study the effects of ray recursion.

squeakybadger
08-17-2005, 04:30 PM
Well its nice to see this thread being so active :D

Thanks for all the help guys, nice and friendly as usual!

It does seem to me that fprime is the way to go. Unfortunately, ill have to put on the red dress and stand on the street corner before i can buy it.

:thumbsup:

Celshader
08-17-2005, 05:10 PM
It does seem to me that fprime is the way to go.

Outta curiosity...what image and render time did you get when you followed Dave Jerrard's suggestions? What image and render time did you get when you ignored Dave Jerrard's suggestions and merely dropped the ray recursion down to 4?

Post the renders if you have them on disk. Curious folks want to know. :)

Dave Jerrard
08-17-2005, 06:29 PM
How's he going to know how it works unless he tries it? Where I work the maximum is 4 until we know we need more. I'm working on a headlight render now that needs 7 to look good only because they get their distinct look from multiple bounces off chrome behind a glass cover. 7 takes more than twice as long as 4. I've yet to see a scene that really needed 16, including this headlight, so I'm betting he's losing a lot of time for nothing by leaving it set that high. Besides, test renders set too low quickly become obvious and render faster.

So you don't know how it works then.

Take a look at his scene. There's not a lot of areas where a surface will reflect in another surface, except for that glass table, and in the images you can see a lot of reflection recursion going on in the far edge of the glass, which makes it look like glass. Lowering it will almost certainly destroy the effect he has going on there.

Now, if he didn't have those recursions going on at all, then lowering the recursion level won't make a scrap of difference, in quality, OR time.

He Who Has Grown Tired Of Fixing Scenes When People Mess With This.

toby
08-17-2005, 06:31 PM
Outta curiosity...what image and render time did you get when you followed Dave Jerrard's suggestions? What image and render time did you get when you ignored Dave Jerrard's suggestions and merely dropped the ray recursion down to 4?

Oh please. Nobody suggested that he ignore Dave's advice or 'merely' lower the ray recursion. If you actually read the thread, you'd see a lot of different suggestions were offered, and nothing else that Dave suggested was argued.

I also don't recommend learning ray recursion on a scene that "normally" takes 32+ hours to render.
Obviously there's no need to render full size with AA to test render optimizing, and a 32 hour render needs all that much more. And if he uses the benchmark scene, he still won't know what to use in this scene.

toby
08-17-2005, 06:56 PM
There's not a lot of areas where a surface will reflect in another surface, except for that glass table, and in the images you can see a lot of reflection recursion going on in the far edge of the glass, which makes it look like glass. Lowering it will almost certainly destroy the effect he has going on there.

Now, if he didn't have those recursions going on at all, then lowering the recursion level won't make a scrap of difference, in quality, OR time.

And if he's made the glass double-sided? How do you know he doesn't have rays bouncing around in the piece of glass hour after hour? If you're saying that anything lower than 16 is going to ruin the glass I'm going to totally disagree, but it's silly not to try it lower when the render is taking so long.

Dave Jerrard
08-17-2005, 06:58 PM
One more thing I forgot to mention. I see that you're using Raytrace Optimization. That's almost always good in scenes with area lights or radiosity. Have you tried the same scene with it turned off. This is one of those settings that really needs to be tested for each scene. Sometimes it can make things take longer. And it should always be turned off if you're NOT using area lights or radiosity..


Also, try subdividing the floor and table a bit. If they're single polygons, they'll render pretty inefficiently. Subdivide them once or twice and you might see a faster render. :thumbsup:


He Who Is Fighting An Unruly Shelf Here Now.

toby
08-18-2005, 12:10 AM
Sorry to flog this :argue: :hijack: but I don't like the way the two of you were completely dismissive of everyone's advice and the 'don't touch it if you don't know what you're doing' attitude.

One of these took 40 seconds at 8 ray recursions - the other at 16 took 2 minutes 18 seconds, 350% longer. Can you tell which is which? The last image is the difference mode in Photoshop, 6 little spots.

I'm sorry if someone ticked you off by lowering your RR too much, but that doesn't mean it shouldn't be tried or tested.

Nigel Baker
08-18-2005, 02:33 AM
Hello SqueakyBadger,

So are you going to post your scene and lets see who can get the quickest render time with the best results.

I think a lot of people would find this most interesting.
It would also produce some other interesting facts.
FPrime users also welcome.
All renders must be posted with detailed render descriptions and renders.

Any Maxwell Render people here, aswell?

squeakybadger
08-18-2005, 11:14 AM
As requested, heres the scene files.

Now play nice. :D

Nigel Baker
08-18-2005, 11:29 AM
... and so it starts,

Thanks Squeakybadger,
I hope this catches on and some interesting issues arrise.
Already started 1 FPrime Interactive render.

Time will tell.

BACK SOON.

squeakybadger
08-18-2005, 12:15 PM
Time will tell indeed.

Let the games begin :rock:

Nigel Baker
08-18-2005, 12:18 PM
Hello Squeakybadger,

Here is the first FPrime Render Test.
Approximately 9 minutes.
Level 10
Monte Carlo Radiosity On
Intensity 100%
Bounce 1
Light Quality 0.6

All other settings as set within the scene.
Image size 1.5 megs

Back Son (whinnie the pooh)

Tiger
08-18-2005, 12:24 PM
I`m in...will post soon :jam:

tektonik
08-18-2005, 12:28 PM
lets get radiosity bounce to 3 instead of 1

knowing fprime it doesn't add much to render time but for interior scenes it is usualy better

a+

squeakybadger
08-18-2005, 01:03 PM
Hello Squeakybadger,

Here is the first FPrime Render Test.
Approximately 9 minutes.
Level 10
Monte Carlo Radiosity On
Intensity 100%
Bounce 1
Light Quality 0.6

All other settings as set within the scene.
Image size 1.5 megs

Back Son (whinnie the pooh)


9 MINUTES!

Sweet zombie jesus thats fast.

This fprime malarky is looking better by the second. :)

Tiger
08-18-2005, 01:12 PM
I tried to load the scene...but Lw. just crash :compbeati

toby
08-18-2005, 03:54 PM
9 MINUTES!

Sweet zombie jesus thats fast.

This fprime malarky is looking better by the second. :)

Well it may take a lot longer to get rid of the grain - FPrime gets to this quality very fast, but to get as smooth as your original takes much longer. I think a higher light quality will help, I'm going to try it when I get home. Thanks for posting the scene!

Captain Obvious
08-19-2005, 12:16 AM
About 11 minutes on a 1.6GHz G5, but I was using the computer for other stuff in the meanwhile. Otherwise it would've taken about 9 minutes, I think. Interpolated radiosity, 0.5 tolerance, 1 bounce, 6x18 rays, 50mm spacing. Enhanced medium anti-aliasing with motion blur activated.

Edit: something must've gone wrong while loading the scene... Ah well, at it again. ;)

Darth Mole
08-19-2005, 01:19 AM
I did this in FPrime and G2 - I used G2's Gamut Control to remove the over-bright area and tweak it so it looked more like the original. Oddly, there's still a lot of variation in shadows, color etc. I've done a little post work in PS.

LW Ray recursion is 4
FPrime MC bounces 1, Lighting Quality 1.0

I just ran it about 11:00 last night and left the sucker on till 8:00 this morning... (Dual 2GHz Mac).

Nigel Baker
08-19-2005, 02:29 AM
Hey Captain Obvious,

Have you noticed that there is a light missing on the far left near the window off the wall?

(and then some ... spotted but only much, much later)(Thanks, today is a bloody mess)

Captain Obvious
08-19-2005, 03:49 AM
Yeah, I noticed that. Something must've gone wrong while loading the scene, like I said, but I didn't have time to do it again immediately.

Darth Mole
08-19-2005, 04:31 AM
It's actually a wall-mounted speaker. Also, both floor standing speakers have gone as have both rear speakers and the large red paintings on the back wall!! Or did you not notice those? :)

squeakybadger
08-19-2005, 05:07 AM
bonus points for the mole man, this will be a home cinema room thing once i add the projector/blinds/other things :D

toby
08-20-2005, 12:12 AM
Here's mine :D

1 hour 10 minutes -
Interpolated, rays 6x18, tolerance 0.4, spacing 40mm, 8 ray recursions

I always like to compare interpolated to monte carlo, so I started one before work - 11 hours 28 min. But it shows how innaccurate the interpolated render is - no shadows in tight places like around the tv and paintings, and a big blotch by the window ... ah, back to the drawing board ... trying an fprime one now - nice room btw -

Darth Mole
08-20-2005, 12:51 PM
If you did two interpolated renders and averaged/merged them in PS, would you get a smoother result? I think the Interp'd image is pretty good (maybe do two with very slightly different values to get two almost-identical-but-not-quite images)

2 hours 40 would still be way better than 11 hours plus!

toby
08-20-2005, 01:39 PM
If you did two interpolated renders and averaged/merged them in PS, would you get a smoother result? I think the Interp'd image is pretty good (maybe do two with very slightly different values to get two almost-identical-but-not-quite images)

2 hours 40 would still be way better than 11 hours plus!

Well rendering on high with motion blur is already averaging 17 passes together. I'm sure it's the eval spacing, but times are going to go up.

Here's Fprime after 2 hrs, after 8 hrs. there was no visible difference. Better than I expected! Grain is pretty much gone, light quality was 3. Seems like the grain never goes away at lower quality settings. But it looks like Fprime doesn't do multi-bounce reflections. More tests...

edit - Also had to dim the area light by 30% for Fprime (?)