Irradiance vs Brute Force

[glorybound]

Is BF getting close to the speed of IRR, or is IRR always going to be faster? If IRR is still faster, by how much? If I set me IRR to a high preset, after the 4 passes, and the actual render, it doesn't seem that much faster. With the last couple V-Ray releases, at least, BF is the default. Why is this? Is it just because it's more. Accurate?

[Alan Iglesias]

With the last couple V-Ray releases, at least, BF is the default. Why is this? Is it just because it's more. Accurate?...

As I remember, at the time Vlado was saying that the BF routines were recently upgraded and were significantly faster than they previously were - and that's when I noticed that that BF was the new default for the primary GI engine. ...And certainly an unbiased rendering is going to be more accurate, ray-wise, than a biased one.

But to my knowledge math-wise, biasing in general should typically be faster than (at least relatively) pure path tracing, so I'd imagine that in general IRR mapping is still going to be faster pixel per pixel than Brute Force - at least most of the time, anyway. ...And this is what I have seen in the tests I have done since that Upgrade....

I'd like to do some more testing though...

Generally, I get the feeling that the general movement is to eventually do away with biasing altogether, along with all the time it takes to get it right, especially with animations. And I guess that's another reason you see these defaults change over time...

Just my observations...

-Alan

[glorybound]

As I remember, at the time Vlado was saying that the BF routines were recently upgraded and were significantly faster than they previously were - and that's when I noticed that that BF was the new default for the primary GI engine. ...And certainly an unbiased rendering is going to be more accurate, ray-wise, than a biased one. But to my knowledge math-wise, biasing in general should typically be faster than (at least relatively) pure path tracing, so I'd imagine that in general IRR mapping is still going to be faster pixel per pixel than Brute Force - at least most of the time, anyway. ...And this is what I have seen in the tests I have done since that Upgrade.... I'd like to do some more testing though... Generally, I get the feeling that the general movement is to eventually do away with biasing altogether, along with all the time it takes to get it right, especially with animations. And I guess that's another reason you see these defaults change over time... Just my observations... -Alan

gotcha, thanks. I still see photon mapping and remember when that was the engine of choice for interiors (I think it was interiors).

[Donfarese]

Well BF technically isn't unbiased unless your talking BF + BF. Since we use lightcache as secondary as default it's all Biased, but can your really tell a difference ever. Close enough in my book. As far as IR vs BF, I find IR to be around 3x faster. That's a great jump from before which it was like 5 or more times faster. Now BF in RT is really fast, and just imagine IR in RT.

[glorybound]

Well BF technically isn't unbiased unless your talking BF + BF. Since we use lightcache as secondary as default it's all Biased, but can your really tell a difference ever. Close enough in my book. As far as IR vs BF, I find IR to be around 3x faster. That's a great jump from before which it was like 5 or more times faster. Now BF in RT is really fast, and just imagine IR in RT.

you say that IRR is 3X faster than BF, however, IRR has several presets. I just waited 2 hours for the IRR high preset to calc prior to rendering. When people give their times, I wonder if it's with, or without pre-pass? BF doesn't have pre-passes. I guess my question is, 3X faster using what preset?

[Donfarese]

I would personally never use High preset, I put it on low and adjust the settings accordingly. Also using low with good settings and using detail detection give you much better quality closer to the look of BF. Honestly if you are having trouble figuring it out use the Vray quick settings. They take a lot of the guess work out. Try this in using the vray quick settings, click interior, make sure it's IR + LC, go to IR and turn on Detail Enhancements. Quite of a lot of situations you don't even need Detail Enhancements, and if you can do without it will help reduce render times quite a bit. Leave GI quality at 0%, set shading rate to 6. Change bucketing to Progressive and set the AA max sub to 6. From there up it to reduce the noise. See the times and quality you get, let me know if that helps at all.

[Donfarese]

Your asking about prepass but that is for animation normally. I'm talking about calculating IR at each render since working on it or single image you might be changing things, so leave the mode at default single frame.

[glorybound]

I'll give it a try.

I would personally never use High preset, I put it on low and adjust the settings accordingly. Also using low with good settings and using detail detection give you much better quality closer to the look of BF. Honestly if you are having trouble figuring it out use the Vray quick settings. They take a lot of the guess work out. Try this in using the vray quick settings, click interior, make sure it's IR + LC, go to IR and turn on Detail Enhancements. Quite of a lot of situations you don't even need Detail Enhancements, and if you can do without it will help reduce render times quite a bit. Leave GI quality at 0%, set shading rate to 6. Change bucketing to Progressive and set the AA max sub to 6. From there up it to reduce the noise. See the times and quality you get, let me know if that helps at all.

[Morbid Angel]

You have to ask your self, do you like the biased look of irmap? for example if you set it to -3/-1 and in the areas where objects are close to each other you see blurry shadows (where they should be defined). If you raise irmap settings until its crisp, it will be almost same as bf, and therefore the render time will be similar.

[glorybound]

Gotcha. It makes sense.

You have to ask your self, do you like the biased look of irmap? for example if you set it to -3/-1 and in the areas where objects are close to each other you see blurry shadows (where they should be defined). If you raise irmap settings until its crisp, it will be almost same as bf, and therefore the render time will be similar.

[super gnu]

You have to ask your self, do you like the biased look of irmap? for example if you set it to -3/-1 and in the areas where objects are close to each other you see blurry shadows (where they should be defined). If you raise irmap settings until its crisp, it will be almost same as bf, and therefore the render time will be similar.

Well in theory (not quite in practise) reducing the min rate will sharpen those contact shadows while still allowing the imap to undersample the big flat areas. That is where the speed is gained. However a couple of things I'd note: 1, the imap engine has not been updated for ages. I'm sure if some of the techniques used to improve BF recently were applied to the imap engine, it could be improved further speedwise. And 2: in the case of multiframe incremental for fly through animation, the speed advantage over BF is still extremely compelling, and you can afford to do an extremely sharp Imap (min rate of 0 or even 1, with high subdivs and minimum interpolation.

Wrt. Setup, I usually switch to the medium preset, then, for a print resolution (5-6k horizontal) Image, I set the min rate to - 10 or 11, and the max rate to - 1 or 0. Then I reduce the interpolation to 10 or so, and crank the subdivs until the splotchies are gone. Problem with the presets are they are not designed for high resolution work. The min rate is much too high (low), meaning lots of unnecessary samples and low speed (plus more noticeable blotches) on the flat areas.

[glorybound]

What do you mean by reducing the interpolation?

Well in theory (not quite in practise) reducing the min rate will sharpen those contact shadows while still allowing the imap to undersample the big flat areas. That is where the speed is gained. However a couple of things I'd note: 1, the imap engine has not been updated for ages. I'm sure if some of the techniques used to improve BF recently were applied to the imap engine, it could be improved further speedwise. And 2: in the case of multiframe incremental for fly through animation, the speed advantage over BF is still extremely compelling, and you can afford to do an extremely sharp Imap (min rate of 0 or even 1, with high subdivs and minimum interpolation.

Wrt. Setup, I usually switch to the medium preset, then, for a print resolution (5-6k horizontal) Image, I set the min rate to - 10 or 11, and the max rate to - 1 or 0. Then I reduce the interpolation to 10 or so, and crank the subdivs until the splotchies are gone. Problem with the presets are they are not designed for high resolution work. The min rate is much too high (low), meaning lots of unnecessary samples and low speed (plus more noticeable blotches) on the flat areas.

[glorybound]

Okay, you are talking about the Interp. samples. Got it!

[Morbid Angel]

Its true about multi frame advantage, I actually asked for a feature to cache bf in same way. In my tests though you can get a sharp clean irmap which will be cleaner then bf in about same time. But these days its not just gi that slows things down. For example reflections and glossy reflections and multiple light sources also slow down rendering. Even without gi. So even if you speed up gi, you still get bogged down by other areas.

[glorybound]

I read something about exteriors with an HDRI not needing GI enabled for a blazing fast and clean render, but it didn't make sense, or I missed something.

Bobby Parker www.bobby-parker.com