|
|
|
|
Van Nuys California
923 posts
|
Thanx for the info. I suspected you had a reason for setting the bucket size lower rather than higher, so I got the answer I kind of anticipated.
If Modo continues to render faster with smaller bucket sizes, that will be a nice feature you can use to promote the product. Smaller buckets give us a better chance of dividing up the hot spots among several threads/cpus. Of course the more cpu power you have crunching on the hot spots the faster they get done.
The greater efficiency of big buckets under other engines puts the render rangler in a nasty position. Set the bucket big and you get a wicked increase in render time if you have a really hot spot in the frame. Set the bucket size small and you loose efficiency all the time. Neither is a good option.
You may have just destroyed the dilema for us in Modo: Just set the bucket size small and forget about it. :D
|
|
10/28/2005 - 3:06 AM
|
|
|
London, UK
7354 posts
|
I would be very interested in learning more about modo's (and others') render architecture. I understand if you feel it's either too damned technical or a "trade secret" of sorts, but I figured it won't hurt to ask. So, what can you tell us about it? ;)
Oh, while we're at it, I have a question about Lightwave's renderer and how it compares to modo's. How does Lightwave's interpolated radiosity work? It seems to behave an awful lot like the irradiance caching method in YaFray (it produces similar results). Is Interpolated in F9 a form of irradiance caching? modo uses irradiance caching as well, yet in the "brick torus" test render at medium quality (http://adh.best.vwh.net/Tori150.jpg), there seemed to be an awful lot of radiosity noise. I haven't been able to produce noise like that with either F9's interpolated or YaFray's irradiance caching. Additionally, YaFray's irradiance caching produces the same result in every rendering, but F9 produces sort of random results. How does modo's radiosity work?
|
|
10/28/2005 - 3:35 AM
|
|
|
Martian Megalopolis, Olympus Mons, Mars
7780 posts
|
Quote from Captain Obvious :
I would be very interested in learning more about modo's (and others') render architecture. I understand if you feel it's either too damned technical or a "trade secret" of sorts, but I figured it won't hurt to ask. So, what can you tell us about it? ;)
If you do a Lux forum search for 'bucket render' I bet you will find the answer there....
For example...
http://forums.luxology.com/discussion/topic.aspx?id=3049&show=bucket%20allen
"Rommel, you magnificent bastard. I read your book!" -George Patton
http://aoleonthemartiangirl.com
Message edited by Marinello2003 on 10/28/2005 - 10:30 AM
|
|
10/28/2005 - 10:23 AM
|
|
|
974 posts
|
Quote from Captain Obvious :
I would be very interested in learning more about modo's (and others') render architecture. I understand if you feel it's either too damned technical or a "trade secret" of sorts, but I figured it won't hurt to ask. So, what can you tell us about it? ;)
We'd like to keep its inner workings a secret for a while, although I did reveal some stuff in that topic Marinello linked to (about halfway down the second page).
Oh, while we're at it, I have a question about Lightwave's renderer and how it compares to modo's. How does Lightwave's interpolated radiosity work? It seems to behave an awful lot like the irradiance caching method in YaFray (it produces similar results). Is Interpolated in F9 a form of irradiance caching? modo uses irradiance caching as well, yet in the "brick torus" test render at medium quality (http://adh.best.vwh.net/Tori150.jpg), there seemed to be an awful lot of radiosity noise. I haven't been able to produce noise like that with either F9's interpolated or YaFray's irradiance caching.
LightWave's "interpolated radiosity" is indeed irradiance caching, but I never got it to work as well as I wanted, so I shifted my attention to noise reduction as an alternative. Starting over at Luxology, I decided to try a new take on irradiance caching. So far it seems to work a lot better (as the radio car demonstrates) and it's also simpler (with no need to adjust tolerances or evaluation spacings). You're very observant about the Tori150 image -- it didn't use irradiance caching (which may have been broken at the time) but rather 10 indirect rays at every point being shaded, which is bound to cause noise.
|
|
10/28/2005 - 12:55 PM
|
|
|
London, UK
7354 posts
|
Thanks for the replies :) but I still have some more questions:
I assume modo's anti-aliasing, DOF, etc, works by taking samples either by following a set pattern or in a quasi-random way. Would it be possible to create a refining bucket renderer by having the renderer render each bucket area in the final image with one of the samples, then the next sample in all the bucket areas, then the third, etc? This would make it, in essence, a refining renderer, and you can still get some of the memory advantages of bucket rendering (though to a much smaller degree). It should perform worse than rendering all at once, but it would provide better feedback.
Would it be possible to get some photon map GI support functionality for modo? Basically, the theory is that you to a photon map of only the nth bounce, and only the final of the irradiance caching bounces respect the photon map. This way, you can render with a large number of bounces and still not get most of the artifacts produced by photon mapping. Rendering with a large number of bounces would be great for some types of work (architectural visualization comes to mind). Take a look at Kray if you haven't already. It's a photon mapping/final gather renderer. The interesting thing about Kray is that you can render with an insane amount of indirect bounces without a big speed hit. I've seen images rendered in Kray with 100 bounces. If you could calculate, say, 50 indirect bounces to a photon map and then render the scene using irradiance caching (say three bounces, where the first two bounces work just like normal, but the third also samples the photon map in addition to any normal lights), you can render with huge amounts of bounces with fewer of the disadvantages of photon mapping. Any flaws in the photon map will be invisible anyway, since it will be so diffused out by the irradiance caching, and you don't need to do time consuming final gathering.
As for the torus, 2.5 minutes for that that image indicates if anything that modo's per-pixel radiosity sampling is also fairly fast. Sure, the quality isn't all that high, but less than three minutes is still impressive indeed.
Thanks in advance for further replies.
|
|
10/28/2005 - 3:39 PM
|
|
|
Martian Megalopolis, Olympus Mons, Mars
7780 posts
|
I was one of the first Kray testers, and it is SLOW!
And all the various options are way way confusing!
Lux has the right idea - keep it simple stupid and fast to boot without loosing quality.
I would go with a Maxwell before I would use Kray.
"Rommel, you magnificent bastard. I read your book!" -George Patton
http://aoleonthemartiangirl.com
|
|
10/28/2005 - 7:47 PM
|
|
|
64 posts
|
Hi everybody,
I am very interresting by modo 201. This video is very impressive... but I would like to see the same with blured reflexive, textured surfaces and an casting shadow area light to be fully convinced.
Please?
Message edited by Eddy MARILLAT on 10/28/2005 - 10:48 PM
|
|
10/28/2005 - 10:34 PM
|
|
|
Van Nuys California
923 posts
|
Yep, I see you point. what happens when we combine features? I can get a quick GI clay bake out of finalRender, Advanced Renderer, VRay, Mental Ray. They all go quick with no textures or special features (caustics, SSS, displacements, etc).
P.S.
Who will be Poster 666 in this forum. The magic number is approaching... >:{-}
|
|
10/29/2005 - 12:32 AM
|
|
|
London, UK
7354 posts
|
Quote from Marinello2003 :
I was one of the first Kray testers, and it is SLOW!
And all the various options are way way confusing!
Lux has the right idea - keep it simple stupid and fast to boot without loosing quality.
I would go with a Maxwell before I would use Kray.
It may very well be slow (I haven't had a chance to try it yet, since the Mac version is still to be released). However, from what I've gathered, it can use a whole lot of bounces without getting slower, which I think is interesting.
|
|
10/29/2005 - 3:53 AM
|
|
|
|
|
|
|
|
|
|
|
|
