Forum: Poser - OFFICIAL


Subject: To light a fire under the Fly

nerd opened this issue on Oct 06, 2002 ยท 14 posts


nerd posted Sun, 06 October 2002 at 1:35 PM Forum Moderator

This was buried in a thread below. Some people have asked me about it. So I'll post it again. I hope it helps people get some of the "Stoned" out of the "Sloth". * Use ray traced shadows, not "Displacement Mapped" Change the shadow setting on every light to "Ray traced". It takes nearly as long as the main scene to render each shadow map if you use mapped shadows. (Note: there is a bug with reflections and ray traced shadows, they don't show in reflections. CL has accepted this bug and will fix it next update. If you have mirrors you will need mapped shadows.) * Set the bucket size to match the memory of your computer. 32, the default is too much for an average system. If you hard drive is churning during a render set the bucket size down. A rough estimate 256MB = 12, 512MB = 16, 1GB = 24, 2GB = 32. Once you've set this it should stay the same unless you add memory. Don't set it too low you will just slow your self down. Set it as high as you can without causing memory swapping. NT/XP/W2K users can use Task Manager to help set this just right by watching Poser's memory utilization. * Forget Smooth Polygons. All that does is make your image look like some Salvador Dali thing. Anything straight gets a weird curve distortion. * Minimum Shading Rate (This is a big one) The Book's description of the settings for this parameter are wrong. The Book states that Higher numbers produce better quality and slower renders. WRONG. It's the other way around. So, set your Draft Shading rate to 4 or 8. Your quality will suffer, but it will render much faster. Even for Production I can't see the difference between a 1 setting and a .1 setting. The .1 takes forever to render. * Map size really effects memory consumption. You won't benefit much from setting it larger than the render size. Use 1/2 render size for draft. * Pixel Samples Set to 1 for your draft. This is like turning AnitAliasing off. * Remove Back facing Polygons. Well it would help, except it may spoil your render, which will waste way more time than it saves. The back face cull is a bit over zealous. It may remove you figures nose if it takes a notion to. (Fine for the sphinx :-P ) Probably better with it off. * GET THE PATCH. It made Stoned Sloth twice as fast... Well not fast. Let's say half as slow. Rooms switch less slow. Program loads takes only a couple of seconds longer than P4. This is all based on the tests I've run. Actual milage may vary... ngsmall02.gif
The Nerd


lemur01 posted Sun, 06 October 2002 at 2:01 PM

Thanks for this Nerd. I've copied and pasted this into a file in the P5 folder - for when I eventually get the f***ing thing working. Jack


the3dwizard posted Sun, 06 October 2002 at 2:03 PM

I knew I wasn't the only experimental scientist in the group! Thanks!


shadownet posted Sun, 06 October 2002 at 3:34 PM

Thanks Nerd. Definitely saving this thread.


lalverson posted Sun, 06 October 2002 at 4:48 PM

copy open note pad, paste.. save.


williamsheil posted Sun, 06 October 2002 at 5:25 PM

Attached Link: http://www.renderosity.com/messages.ez?ForumID=12356&Form.ShowMessage=899432

Link to another thread of current interest. I'm still using a default bucket size of 32x32 (lower values had no discernable effect), and my memory usage is maxing under 300MB during the rendering (oddly much more during the shadow mapping), with no disk thrashing on a 512MB system. Bucket size for a REYES renderer (supposedly like Firefly) is a compromise to allow z-buffereing to represent (numerous) multiple layers of transparency (and support a few other things that Poser cannot do). Unless you are rendering scenes with lots of transparency levels (eg. someone-or-other's fog prop ;)), there is no reason why a 32x32 bucket size should consume more than a few megabytes of memory. The architecture is intended to be extremely resource friendly. Firefly seems, however, to be at quite the opposite extreme and, with the presence of non-localised (non-REYES) effects and the persistence of unnecessary (in a REYES architecture) polygon shading and smoothing options, really makes me wonder if any at CL really had a clue what they were trying to do. Just for general information, REYES renderers stand head and shoulder over other rendering methods in terms of quality and (speed/resource) performance, so CL definitely made a good call in going for this technology. Its just seems to me that (by my own expections as well as all comparable standards and competitive products) they have really managed to screw the implementation, even if it promises some advantages over the P4 engine. Maybe they should have stuck to the P5 formula and gone for a bolt on third party engine instead. Bill

ScottA posted Sun, 06 October 2002 at 6:39 PM

Weird. Everything Nerd listed works just the opposite for me. 1.)Higher bucket = faster rendering 2.)Lower shading rate = faster rendering (as manual states) 3.)Backface polygons don't seem to influence render speed 4.)smoothing does produce strange warping on some body parts. But rather than tossing the smoothing function away completely. I just uncheck the smoothing option on that one body part. The smoothing works really nice with low res models. I wonder why we get opposite effects from the same settings? I'm currently running using the app patch only (SR-1b). ScottA


sirkrite posted Sun, 06 October 2002 at 7:49 PM

What they didn't know when they open the box and installed the software was, they were about to venture into the... Twilight Zone!


nerd posted Sun, 06 October 2002 at 7:54 PM Forum Moderator

Yes larger bucket = faster render, until you start swapping which depends on the complexity of the scene and your system. Simple scenes (1 figure and some hair) I can set it to 64 or even 128. Complex stuff, like a world scene, 16 or 32 yielded faster results. I assume this was due to disk thrashing. The Stock draft render shading rate is 2, Production rate is .5 Smaller numbers = slower render. At least that's the way it works on my SP1B. For example: Ball Prop on the ground 2.0 Shading rate renders in about 7 sec. Same scene with .1 Shading rate was 42 Sec. Yes, back facing polys have little or no effect on render speed, but if fire fly decides to cull the wrong polygon (which it has done to me) you have to re-render with that setting turned off. Actually I don't think it's done it since the patch maybe I'll get brave and leave it on to see if that bug is fixed. I hadn't tried turning the smoothing off on body parts. I takes so long to get a render I kind of inhibits experimenting with settings. The only difference is that I used the Full Patch. Would you double check that shading rate thing. If the settings are really different for different patches they need to straighten that out. ngsmall02.gif
See I told you milage would vary ;-)


nerd posted Sun, 06 October 2002 at 7:59 PM Forum Moderator

Actually Sirkrite I feel a lot like a Beta Tester. I give my beta testers a free copy of the finished product. I'll be waiting for my free copy of the release version when it's ready. CL can ship to the same address... ngsmall02.gif
Tired of wading through the land mines.


sirkrite posted Sun, 06 October 2002 at 8:04 PM

Don't hold your breath Nerd. ;) That's okay, my Poser five arrived at my house on... Friday the 13th. 8O


ScottA posted Sun, 06 October 2002 at 8:36 PM

Sorry. My Bad. I thought you were talking about Pixel Rate instead of Shading Rate. As far as Minimum Shading Rate goes. Rendering does speed up with larger numbers for me as well. ScottA


herr67 posted Tue, 08 October 2002 at 7:11 PM

Just did a small test in rendering speeds I had an image with 1 Vicky with P4 hair (Total Hair), 1 hirez square prop for a background, and 1 hirez square for water with reflect and refract maps plugged in. Displacment off and set to o.oo RayTrace on (needed for reflections) shadows on Smooth off remove back polys off min shade .5 max texture 4096 pixels samples 3 bucket 32 Raytrace bounce 4 post filter 1 gauss render time 19:56 set Raytrace bounce to 3, render time 20:07 Put raytrace bounce back to 4 and bucket size to 64 Time 20:00 Bucket back to 32 Min shade rate to 1.0 Time 10:30 The only quality differance I could see was the models lips did not have as much texture than with it set to .5. I did have to zoom in (4x) to see any differance. The render really slows down with reflect and refract, without refract it does speed up quite a bit (great for test renders to see different reflection settings) My CPU is P4 2.26 Ghz 1G RAM If you reproduce this test do not do it with a 20 minute render, it takes too looong :-(.


Spanki posted Wed, 09 October 2002 at 7:26 PM

"The only quality differance I could see was the models lips did not have as much texture than with it set to .5. I did have to zoom in (4x) to see any differance." If you have PSP, load your saved image1, paste image2 as a new layer and set the new layer to 'Difference'... this will show you exactly which pixels are different between the two images (I'm sure there's a similar method in PS).

Cinema4D Plugins (Home of Riptide, Riptide Pro, Undertow, Morph Mill, KyamaSlide and I/Ogre plugins) Poser products Freelance Modelling, Poser Rigging, UV-mapping work for hire.