MR 3.10.1.4

[Hirazi Blue]

Let us try to put this OT-baby to bed! I completely agree that good communication is key, I personally, however, don't think good communication has to equal perfect language skills. That's why I responded like I did in the other thread, where you posted solely to correct the language of someone else, which could be construed as being offensive, even if you do not intend it that way. And as you feel strongly about this, you'll probably keep on doing it. I won't use my "utter awesome powers of awesomeness" (being an Administrator and all) to prevent this, but when applicable I will tell you if I find such a language oriented comment "just too much". So, truce?
;)

And thanks for you last comment!

[McNistor]

Sure man, I don't thrive on conflict even though I might have given you and others that impression and this because I put truth before anything else - diplomacy, tranquility, people's feelings (mine included), etc. I won't pick on people's communication again because as I've said, I'm sure I won't win any fights, not on the internet forums anyway and also I'll try to bury the urge to speak "from the heart" and be more of a diplomat.
So yeah, truce!

[McNistor]

Back on topic: I'm not sure how the argument of AD having apps with "lengthy" features holds water because having a small company with fewer people working on a small software is (or should be) the same with a big company with more people with more software. Doesn't this scale up proportionally? In this equation one parameter is skewed (like AD having too few developers) if the equality is broken. Or maybe I'm too simplistic in approaching this problem, but right now there's no reason to complicate it myself.

p.s. I didn't feel like editing the above post as I think that's a little too important not to be left as is.

[reberhart]

A missing apostrophe? Big deal. It's the people who don't know the difference between "loose" and "lose" that drive me bonkers. 90% of the time it's native speakers making that error.

[gustavoeb]

Must say Im astonished by this results!
Using Unified Sampling and Brute Force GI renders really see a huge bump in performance. I have reduce all local sampling to a minimum, in a technique some know as Unified Sampling Redux (due to the elementalray blog)
Rendertime on SI 2012: 36 minuts
Rendertime on SI 2012: about 16 minuts (and less than half the eye rays)



Most very glossy reflections look alot better, but very dark areas dont (see pants and wood). Which is consistent with this:
http://elementalray.wordpress.com/2012/ ... 3-10-more/

Very pleased with this, hope nVidia keeps pushing this, because they have much to do still...

Here are the render stats and images follow:
2012
# INFO : RC 0.4 info : type number per eye ray
# INFO : RC 0.4 info : eye rays 14722756 1.00
# INFO : RC 0.4 info : reflection rays 12505293 0.85
# INFO : RC 0.4 info : refraction rays 8000853 0.54
# INFO : RC 0.4 info : shadow rays 264843884 17.99
# INFO : RC 0.4 info : environment rays 39935323 2.71
# INFO : RC 0.4 info : finalgather rays 40016982 2.72
# INFO : RC 0.4 info : probe rays 102920370 6.99
# INFO : RC 0.4 info : fg points computed 20033619 1.36
# INFO : RC 0.4 info : wallclock 0:36:05.79 for rendering

2013
' INFO : RC 0.4 948 MB info : type number per eye ray
' INFO : RC 0.4 948 MB info : eye rays 6081724 1.00
' INFO : RC 0.4 948 MB info : reflection rays 7243697 1.19
' INFO : RC 0.4 948 MB info : refraction rays 8659735 1.42
' INFO : RC 0.4 948 MB info : shadow rays 121543655 19.99
' INFO : RC 0.4 948 MB info : environment rays 25238787 4.15
' INFO : RC 0.4 948 MB info : finalgather rays 16838102 2.77
' INFO : RC 0.4 948 MB info : probe rays 61285672 10.08
' INFO : RC 0.4 948 MB info : fg points computed 8427435 1.39
' INFO : RC 0.4 909 MB info : wallclock 0:15:54.00 for rendering

[Maximus]

What settings did you use for brute force GI? Exact mode? how many rays?

[gustavoeb]

Exactly, exact mode

I had all local samples at minimum, so I had exactly 1 ray
As you can see, from the render stats, that does not mean unified will shoot only one FG ray per eye ray, it seems like it tries to be smart... Anyhow if it were a very bright scene I might have had more samples in there...

[Kzin]

nice test.
but i think your scene is perfekt for 1 fg ray because of its darkness and but you can also see alot of noise. ;)

the problem with exact fg is the noise which needs alot of samples to get clean. yes you can get fast results, fast for unoptimized brute force path tracing, but you will need more rays for more brighter renders, something like 40-60 fg rays from my tests. rendertime is then about 3 times slower then brute force ip's and up to 5 times compared to ultra high fg interpolation settings.

but its great to see that the performance is getting better and better. lets say what new shadingmodels can do to further reduce rendertime.