PDA

View Full Version : Rendering Speeds Too Slow


dragonfollower
01-18-2007, 08:27 PM
Hi,

Doing a test render with Mental Ray last night, I was discouraged by how slowly it renders, compared to my previous app (Lightwave). I don't know if I just simply need a faster computer or what. I was hoping I could get some advice.

Test scene:

. Single character on grid/groundplane, low-res sphere as a skydome.
. No textures
. Ambient Occlusion on character and groundplane
. Two lights....spotlight, pointlight, shadowmapped shadows
. No final gathering, GI or anything like that
. Motion Blur (this REALLY slows it down)
. 720X480 DV resolution

My system specs:

AMD Athlon XP 3200+ 2.19 Ghz
1 GB RAM
Nvidia Geforce 5950 FX

The final output is for an animated children's show. Even with such a simple scene, rendering at 720X480 was taking 1 min and 45 sec a frame. This was WITHOUT motion blur. Previously I used LW and render times were around 30 to 45 seconds for a single character (AND motion blur). For this show, there will be several characters and rendered at HD resolution. I'm worried, because render times will be astronomical.

Any advice would be greatly appreciated!

-Brian

CiaranM
01-18-2007, 09:06 PM
You need to post 3 very important settings before we can really help you out:

- Antialiasing (min, max, threshold, filter and filter size)
- Spotlight map resolution and samples
- Ambient occlusion settings (samples, distance etc. etc.)

I suspect therein lies your speed problem.

T4D
01-18-2007, 11:21 PM
your not going to get Mental ray to shot out 1-2 min render straight off the bat like LW
Mental Ray is so much more powerful then LW in every single way
there's a reason one MR render node costs more then a LW licence.
( tho i would love to have Fprime in XSI just for setup )

But all the power has to be learnt young jedi :D

check these posts out from saturn
http://forums.cgsociety.org/showthread.php?t=389990&page=12&pp=15

check out all the tutorials around softimage.com etc

Mental can do things HEAPS faster then LW and handle Tons and tons more data
I had a 15 million polygon scene in XSI ( LW could only load 2 million polygons )
MR 45 mins - LW 35 mins ( both win 64 bit )

after going over Saturn's trees and digging around MR i got my render now down to 5 - 9 mins a frame !!
you just would never get that performance jump from LW with that many polygons. ( Also LW just could not handle that many polygons )

MR is alot more then LW and you have to do alot more then play with the settings in front of you ( NO disrespect there I done it myself )

you have to learn the system and design how you want it to render,. like most of XSI nothing is on the surface you have to dig deep and connect the dots to get the most out of it.

Your not in Lightwave anymore,.. this is a much bigger and more powerful system so it has to be learnt in a much more deeper way.:buttrock:

dragonfollower
01-18-2007, 11:27 PM
Duh! You're right...my bad. I guess that shows my apparent lack of knowledge with rendering.

AA:
Min level: 0
Max Level: 1

Not sure where to look for threshold, filter and filter size (using 6.0 btw)

Spotlight:
Map Resolution: 1024,
Samples: 20

Ambient occlusion settings:
Samples: 30
Spread: 0.803
Max Distance: 0.292
Output Node: Occlusion using bent normals

Also, Mental Ray is set to Raytracing rather than Scanline or Rasterizer

Let me know if there's any other settings I'm forgetting.

Thanks!

-Brian

dragonfollower
01-18-2007, 11:31 PM
Thanks for the advice T4D. Point taken, I'll take a look at that link too. Your post is actually very reassuring. I was afraid LW was simply faster, but this gives me some encouragement that if I can figure out what I'm doing with it, MR has the ability to kick LW's butt.


-Brian

JDex
01-19-2007, 12:44 AM
First off... set the Primary Ray type to Scanline (you have it in Raytrace). If every ray is a raytrace right off the bat, you may be losing valuable cycles where raytracing isn't needed. If the scanline ray hits a sample that requires raytrace, it will change to raytrace and continue the trace.

There's hundreds of reasons that it could be taking so long... read up on the BSP, and how to tune it. Here's a good place to start http://www.lamrug.org/resources/chardiags.html (well, the link and all the resources through out that site are very helpful).

Is there a Shadow map on the point light, if so does it really need to be a point light? Point light shadow maps can cost quite a few unneccesary cycles.

Motion blur is the dog though... it's pretty much the breaker in most cases. Is this all multi-pass or beauty renders?

T4D
01-19-2007, 12:46 AM
Thanks for the advice T4D. Point taken, I'll take a look at that link too. Your post is actually very reassuring. I was afraid LW was simply faster, but this gives me some encouragement that if I can figure out what I'm doing with it, MR has the ability to kick LW's butt.


-Brian

Cool :thumbsup: I did the same thing and for the first few jobs i did render animation back in LW
only when i was pushed to HAVE to uses MR due to polygon count was when i had to sit down with MR,. I was freak'n out abit at the render times etc .. but It's worth learning it ,.But the cost is still a issues ( I pretty much render everything in XSI now )

But You can't just throw more CPU's at the problem like on a LW job,. ( cause they cost real $$ )
so you have to work alittle hard and smarter .. ( so I'm stuffed really LOL :D )

dragonfollower
01-19-2007, 01:42 AM
Okay...I've messed with some of the settings that have been pointed out and I have gotten renders down to 33 seconds. A little more comparable to my LW render speeds.

The big time killer seems to still be motion blur though. WOW. It's unreal how much time that adds. I don't remember this being much of a problem in LW. I wonder why MR has such issues with it?

-Brian

GeneralLethal
01-19-2007, 01:57 AM
I've found rendering a motion vector pass and using it to render motion blur in comp is a good way to save time, with results comparable to 3d motion blur. The motion vector pass renders very fast, plus if you have to rerender for whatever reason, you don't have to compute motion blur each time, just once when you render your comp. If you go that way you'll need the Reelsmart Motion Blur plugin, though, which is available for several 2d apps.

3DDave
01-19-2007, 05:56 PM
If you compare LW motion blur to MR you will see a hugh quality difference between the two, MR being much much better. I asume this is one reason for your move to XSI?

There are 2D motion blur solutions that can be done in post or right in XSI essentials or advanced using Real Smart Motion Blur. http://www.revisionfx.com/.

Also XSI Ambient Occulsion is slower than the Dirt Map shader, you may want to try that out.

dragonfollower
01-19-2007, 07:57 PM
The main reason I switched to XSI from LW was because of its character animation abilities. Also I found modeling, texturing, rigging and pretty much everything else more intuitive. So far Rendering has been my only dilemma....with motion blur looking like the main problem.

I'll look into Real Smart Motion Blur. It looks like it just might work for me. Also, I hadn't heard about the Dirt Map Shader. I'll check that out as well.

Thanks!

-Brian

dragonfollower
01-19-2007, 08:03 PM
Okay, quick question....

Where is the Dirt Map Shader located? I can't seem to find it in XSI. Is this something that was created for XSI or does it come standard with the software?

-Brian

JDex
01-19-2007, 08:06 PM
You can get it here:

http://www.xsibase.com/tools/shaders.php?detail=332

I have found Dirtmap to be faster if you are using low to moderate sampling levels... if you require very high, grainless samples, the XSI_Ambient_Occlusion shader seems to end up faster.

dragonfollower
01-19-2007, 08:16 PM
Thanks JDex!

adrencg
01-21-2007, 07:21 AM
Invest in some faster hardware, if you're making any money whatsover. That is a slowwww system. The Athlon XP 3200 was released 4 years ago I believe. I complain about renders on my Core2duo(which I just bought a few months ago), so I'm investing in a dual quadcore renderbeast.

Hardware is cheap, invest a small amount and it will pay off. Not only in rendering time, but just in the amount of time it takes to get things done. A simplistic answer, but no amount of fiddling with MR is going equal raw horsepower.

and also, Dirtmap is great. use it as an alternative to the default AO.

CiaranM
01-21-2007, 02:53 PM
I would also recommend using ctrl_occlusion. It has adaptive sampling and a few other tricks up its sleeve, which can speed up your renders.

mdee
01-21-2007, 06:45 PM
For characters with motion blur, I strongly advice to use rasterizer. I rendered a ton of characters that way for commercial work, and usually my render times were around 1 minute per frame per character (no background) on average xeon machine from 1-2 years ago.

Regular motion blur in mr is rather expensive rendertime wise.

mocaw
01-21-2007, 07:17 PM
Time to start thinking in passes maybe too? If you have something in the scene that is moving quickly, you should, NO MUST, check out using rasterizer instead of scanline rendering as stated by other users (mdee). Sometimes that can make MB times much quicker- and it should always be tried. As stated before DO NOT underestimate the power of the BSP tree. I'd learn how to tune it visually first (using diagnostic modes). That alone can sometimes speed a render up many, many fold.

I'm an x-waver too, so I understand/feel you pain. It took me a good two years to get my head around mr enough to use it for everything I do.

On another note: You're most likely doing a "lite" scene for XSI- so you're speed differences between LW and mr are going to be small. You'll notice fairly quickly though that with "heavy" scenes that you might not even have been able to render in LW WILL render in mr- and often much faster once you know a bit. I do a bit of rendering for larger size print work that LW would have taken forever to finish or not at all (all this in 1.5GB RAM!). Also, shaders play a larger role in mr it seems. They are often very flexible.

Check out this shader here (has great shading models not found in LW or native XSI, and speeds up several things, esp. blurry reflections.
http://www.tek2shoot.com/knowledge-base/softimage-xsi/t2s-illumination.html

ctrl.studios also makes some amazing shaders that, when used correctly, can make scenes render quickly/better.

CGTalk Moderation
01-21-2007, 07:17 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.