Acquiring caustics

[KierovNick]

Hello.
I am kinda fresh in here, as i have found Octane just about week ago.
I have rendered few scenes and results are good, but then i wanted to make a test on caustics... and then problems started.
First of all, i am unable to get caustics. I have tried many times in different modes, with different settings and different material settings, but all i get are normal, simple shadows as like it was all diffuse material, not specular.
I was wondering what i am doing wrong.
I run the laterst demo version of octane, and i am thinking about purchasingthe licence for it, but since i am unable to render even the simpliest scene with caustics.. well.. my guess is that i am making somethng wrong along in the process, yet i sit on the problem few hours looking for answers in google and even in random setiings changing, hoping for luck, yet with obvious results... and i am almost out of options.

So, the zip attached contains all the files i have, starting with SketchUp file i have exported it from to 4 images with the results i got.
As You can see in this case i've used mesh light but i've tried also with enviromental light.. all with no difference to lack of caustics.

Additionally i'd like to ask if, shown on the picture in attachment, amount of hot pixels with PMC is any normal or is it my 9600GT causing problems?

[t_3]

ok, a few things:
first off, your scene scale is wrong (guess 1/100 of what it was meant to be). octane units are meters, and thus your objects are only a few millimeters small.
of course caustics depends on the geometry; a hollow or cubic body for example won't create the effect you might look for.
as it lasts a while till caustics emerges, a slower card may need a good amount of minutes till it gets visible (a few hundred samples per pixel at least); and the fireflys will also vanish if enough samples were calculated.

here a quick example, roughly based on your scene (pmc, 4000 s/pix, 3min on a dual gtx 580):

[KierovNick]

Ok, thank you.
It really starts to make sense.
Yet this creates one more question.
How could one possibly check on the scale of the model within Octane?
Honestly, i just followed info seen somewhere, that while exporting from SU i should set units to meters, and so i did.
If this is wrong i'll have to set something else, but since programs may see it differently, i have to more-less know how big the model is.
Can you help me out?
(i know i could just scale the model up 100 times, but that solution in my oppinion is not quite a solution, because it messes up mu SU units.)

And another quick question.
PMC doesn't really cover those fireflys on my machine. In PathTracing it does, but in PMC they seem to build up untill after few hundred passes there's enough of them to almost completely white out the picture.
Another interesting thing is that this happens more-less at the same speed no matter what type of light i'll set or if i set any light at all.
Am i, again, doing something wrong?

Thanks.

[mib2berlin]

Hi KierovNick, you could change the default obj import settings in "File > Preferences > Scale". Hectometers?
The PMC kernel do not work with older hardware. My GTX 260 don`t work with PMC.
IIRC in linux but don´t know exactly.

Cheers, mib.

[KierovNick]

For the PMC, it will not be much of a problem, as i plan to change my graphic card in this month.. still don't know to what will i swich to, but my guess is that it'll be somewhere around GTX 560.

If it'll work nice with Octane, then i will buy myself a licence and conquer the world with it.

And for the scale.... Unfortunatelly, i can't set hectometers.. i have meters and then kilometers.
So i guess i will have to scale it within Su model anyway.
So this would mean that i have to scale it down 10 times and set centimeters or scale up 10 times and set kilometers. Am i correct?


EDIT.
Holly strawberry pie!
Sorry for this, i didn't read you carefully enough. I thought You wrote about exporting settings and not importing. The thing You described i will do for sure. Thank You, this solves my problem. All of them, actually.

But since i am writing. Have anyone of You any comparisson or knowlede about what nVidia GC are best for Octane?
Or is it just "the more expensive, the better" rule?

[t_3]

regarding scale: the default camera is 10 meters away from the scene center; depending on how large your scene is (should be), this should help to find out if the scale is right, if it looks way to small or to large if you just start rendering.

and regarding cards: the more expensive the better is a good rule of thumb
this might help also: http://dl.dropbox.com/u/10375307/octane_bench1.pdf

[KierovNick]

Thank You for Your help.
And so i will have to buy graphic card first and then play with caustics, for what i see, my current card not only refuse to calculate PMC correctly, but also doesn't do caustics.
The scale of the model is fine now (and for the sake of certainity, i have placed 3 identital scaled instances of the same object (simpliest magnifying glass) over the surface, and... it just doesn't compute.
Caustics should appear after few hundres s/px?
After over 2000 s/px i get almost perfectly evenly dark shadow with not a slightest mark of caustics.
My questions is.. is anyone interested in buying 9600GT with intelligence module that can all on its own take caustics out of the equation?
I'll sell it cheap or exchange to some other card! Say.. GTX580 will do.. It is still cheap as potatoes when you consider that this is the first known to humanity piece of hardware with naturally born, and not coded, intelligence!


----------------------
1000 s/px later
----------------------
I have noticed some fireflys apearing near the spots where caustics should be.
This confirms that my graphic card is intelligent, and maybe even self-aware! It wants to tell me something important, but i'm still not sure what.

----------------------
another 1000 s/px later
----------------------
Fireflys covered like 20% of the zones where caustics could be. I read it as "Hi, i am alive but i can only comunicate with 20% of ways you can comunicate with your kind."
Note to self: I am going to get rich selling this piece of equipment.

----------------------
5000 s/px
----------------------
Fireflys doesn't seem to multiply anymore. Graphic card still can comunicate only with 20% of possible ways. This makes me sad. (but rich none the less)
If i'd applied gaussian blur on it with radius around 50 px it still could pretend to be caustic effect.

---------------------
7000 s/px
---------------------
Dots keep showing randomly but veeeery slowly. This means i have to ask my graphic card another question so it could give some other answer.
This experiment will not continue. Even if those dots would someday grow up and became caustics, i just have no time to wait that long. Next week i have a bussiness trip scheudled.
GTX is almost knocking to my door.
Or it's someone from government to check if my super intelligent graphic card is still available.
Rest assured that i will not give it to them. I will exchange it for GTX580 to first one of you who will ask for it.


Conclusion (the purpose of which is to show you that this is not a spam) is that even though Octane is much faster with GPU than any other renderer CPU based i've tried, it still needs good piece of equipment to work properly.
To make it more scientific i'll write equation that describes everything to anyone thinking about purchasing Octane.

V ͪ *Log18(sqrtԒ[Σ1024ª])Ѡ ≤ 5Vˢ(sinѵ15‰/3.1415Ω5)

where
V ͪ - your hardware value
Vˢ - your software value
and everything else equals 1 because it's only purpose is to make equation look complicated. (I hope You're impressed.)
This translates to:
Don't buy software worth 5 times more than your hardware.

I'll let you know how are my caustics when i'll buy (or exchange) some GTX.

Thanks for Your help with scale dillema!

[t_3]

as keplers appear on the horizon, gtx 580 prices are dropping. maybe a good time to grab yourself one. btw, i still have a gt 9600 in my xp-based htpc - will try my caustics test on this machine...

[ayan13]

Dear t_3, I have one question. The glass shadow should have color, but your test image the blue one and the orange one have the same shadow color, also the shadow is not transparent as real world should.

[KierovNick]

Good point.
I've noticed it too, but i thought that i'll check on the results when i will have propper hardware to compute that on my own.
My first thougth was that the reason for this is that this scene was set to match my scene, but even then, with the thin layers of glass, shadow should be barely visible, yet in octane from what i see shadow is always 100% opaque at first, and then slowly caustics are built up. The render that t_3 introduced was baking only for 3 minutes... yet...
Would the shadows eventualy be as they should or is there a special trick to do in order to get the propper shadows (and thus, propper caustics)?

EDIT
Just checked in Kerkythea. As in Octane, the shadow in the scene is opaque in beginning, but...
In KT there are two modes suppurting caustisc, PhotonMap+Caustics and this gives fairly good result, and BPT (Bidirectional Path Tracing) and both give caustics instantly (they show up as one would expect, from the beginning). I know that it is completely different engine and it's CPU, not GPU based, but caustics should look like in real world no matter what processing unit program use, and in KT i have the option to turn on the mode that does what i want, is there one on Octane?
To show how does it looks like in KT i've attached file.
As you can see, shadows are not perfect, but even with this i at least can see where all the light gone. It bends in the glass and follow to different point. In Octane it seem to vanish in large quantities so in the end one needs few hundred s/px for caustics to show up.
The 3 minute rendering i showed had 8 passes (not sure if it's equivalent to s/px, though. Probably it is.)