PDA

View Full Version : Rendering with LW's native engine & HDinstance



Larry_g1s
07-07-2008, 02:26 PM
I have a project where I'm having to use HDinstance (presently v1.8.2). I primarily use fPrime & Kray for rendering. My problem is, the project I'm currently working on will require instancing due to the forest like scene. Obvious fPrime doesn't do/read any instancing, and Kray's instancing is not robust enough to handle such an environment. I'm not opposed to using HDInstance, which is why I purchased it in the 1st place, what my problem is, I'm very unfamiliar with rendering in LW's native render engine & particularly with HDInstance. I don't know how to optimize the render. I'm currently trying a 640x480 resolution test, and so far it's taken over 8 minutes with only 40% complete. That's on a Quad Core processor, no GI, and classic, low AA! I've gotta be doing something wrong here. The scene has 128196 polygons. I've got 3gb of DDR2 ram. Any help would be greatly appreciated.

Larry_g1s
07-07-2008, 03:46 PM
FYI...I'm using a tree exported out of Vue (see attached image). Would all the transparencies do that to the render time? If so, what might be an alternative? I've dumbed down the poly count as low as it'll go in Vue before exporting, and 2D clip maps will not do.

Cageman
07-07-2008, 04:55 PM
You don't need to raytrace the transparency, so make sure you don't have Raytrace Transparency turned on. IF, you for some reason need to have Raytrace Transparency, make sure you set Ray Recursion to a low value. Default is 16 and by lowering it to 4 you'll see an enormous speedup. In many cases you'll not need as high as 16.

Mr Rid
07-07-2008, 05:13 PM
You might also see if you can get away with clip maps instead of transparency maps.

Larry_g1s
07-07-2008, 07:27 PM
Thanks so much guys. When you're on a tight dead line, believe me, I really appreciate the suggestions. I'll try them out.

Larry_g1s
07-07-2008, 11:33 PM
You might also see if you can get away with clip maps instead of transparency maps.When you say clip maps, are you referring to just 2D images of trees? I don't think it will be all that possible because I'm going to eventually do an animation. I think the 2D clip map images would be very noticeable.

Mitja
07-08-2008, 01:03 AM
I think Mr. Rid was refferring to clip-map the leaves of the tree, not to use a 2d tree.

Cageman
07-08-2008, 01:35 AM
Yeah.. clipmaps would probably be the best way to go, unfortunately clipmaps are still an object property rather than a surface property. :/ I see a huge and messy workaround with an object like the one Larry_g1s has...

JonW
07-08-2008, 03:31 AM
I recently did an architectural scene with over 10,000,000 polys using a lot of Onyx trees and UV mapped leaves in the foreground. AA was 8 plus AS at 0.05, Radiosity 5, all settings were high, lots of glass (transparent) and reflections. Over a 1.3gb of images. About 4 hours for an A3 300dpi scene (5000 x 3500) about 6gb ram was used, on dual 5335 (8 x 2.0ghz).

Look at WTM and see how much ram you are using at your full size render. I find I need around 3 to 6 gb of ram for an A3, ram is so cheap just put 8 gb in your computer (you will need 64 bit OS), it is enough to cover most situations. If your rendering is using virtual ram it’s a disaster.

Flatten your objects in modeller as much as you can and put the the flattened version in Layout. I found with 9.3.1 you gain about 10% speed (for architectural imaging). It was much more of a gain in on older versions of LW so it was worth while flattening objects. Get rid of things that are not seen.

JonW
07-08-2008, 04:25 AM
Also..... As Cageman said.

I did some tests a while back on Ray Recursion Limit on one on my architectural scenes, I havenít got the exact figures but on a small test render it was something like:

1 1.00min:0sec
2 1:30
3 2:00
4 2:30
5 3:00
6 4:00
7 5:00
8 7:00
9 9:00
10 12:00
11 15:00
12 18:00

I hate to think what 16 is, or 24 for that matter, as RRL timing appears goes up exponentially.

I stopped at RRL at 12 because I had enough. look closely at your scene and see have low you can get away with (in my scene I found 7 was the absolute minimum and 9 was, to be honest enough fanatical, 11 was just been anal, and there was nothing more gained at 12, even though the timing went up more, I have also found that the lowest figure seems to be an uneven number). You may find in your overall image no one will ever notice that a reflection is wrong at 7 for example.

You will clearly see if you have a too low RRL as any transparent or translucent objects will go solid as per the background colour. Double up some of your test renders and do some of your test renders with low RRL during construction of your Layout tests, so that you already know the RRL for your final render. You will already know that say ď5Ē for example is enough, this will obviously speed up you test renders as well if you get on top of RRL early in your Layout construction.

Also donít put transparent or translucent object behind one another unless it is absolutely critical. You could double your RRL with two objects and then triple your rendering time.

Low RRL will be one of the best speed improvements you can make.

Cageman
07-08-2008, 10:43 AM
Great breakdown of how RRL affects a scene! Valuable info for those who are not used to optimize for LW-rendering! :thumbsup:

Mr Rid
07-08-2008, 01:11 PM
I dont think I have ever turned RRL above 2 for reflections. In most practical shots, more recursion is not noticeable. Cant recall needing to go above 8 for refraction or trace transparency.

animotion
07-09-2008, 11:57 AM
O K, so what's the speed of dark?

Thats very simple, zero. If your in the dark, you aint goin no where man!

Larry_g1s
07-09-2008, 12:31 PM
Does HDI 2.x render faster then 1.8.2?

pixym
07-09-2008, 04:08 PM
Does HDI 2.x render faster then 1.8.2?
Not really

Mr Rid
07-10-2008, 02:31 AM
O K, so what's the speed of dark?

Thats very simple, zero. If your in the dark, you aint goin no where man!

Actually, dark sucks the light along.
http://www.freakface.com/speedofdark/