Thanks guys!
While testrendering with unified sampling, I found that
I get fastest turnaround if I take the time to manually
set sampling for glossiness or AO or lightsamples to a
reasonably clean (e.g. not too noisy) result first,
then rely on the unified sampling and pixel filtering filter
to clean out the result.
Hooked all AO nodes to the Renderglobals by setting rays to 0
for each shader´s AO node, then set it globally.
Glossiness samples in the floor and the chrome came out better
this way, I´d describe it as wheigting the input for the
unified sampling according to scene demands.
One caveat I found with the above is that regardless of the
realtively quick turnaroud for large, homogenous surfaces,
geometry edges will look noisy unless one cranks down the
Cuttoff Parameter to very low values (in the range of 0.03)
to prevent edge jagginess/crawling noise.
That is something that reminds of VRay when using a similar
(even if not at all identical) approach to sampling...
Most people leave the AA filter default of 1.5 in Vray,
which pronounces this even more (2 would be better) that´s
why I chose gauss 3 in mR, to blur the edges a bit more...
I´ll happily see how low I can push this, thought.
First, it´s trying irradiance instead of envAO for a comparison.
Cheers,
tim
On 26.02.2013 04:21, Jason S wrote:
True.. with unified sampling, it's better/faster to let it take care of
crunching through the noise (and crunches only if necessary)
Though in my experience, helping it a little (area lights at ~2, AO at ~4)
yeilded faster results, yet maybe I was missing something.
I heard Sony Imagework's version of Arnold does have a unified Sampling
equivalent.
Nice image though!
On 25/02/2013 5:10 PM, Ciaran Moloney wrote:
Looks great! You should be able to drop all of your shader samples (including
AO) right down to 1. Then push up the max samples and quality in unified
sampling. Keeping shader
samples at a minimum will allow for more efficient sampling of the scene, only
where it needs to be done.
As much as I like Arnold, I do miss having an adaptive sampling solution. Won't
make me use MR though...
On Mon, Feb 25, 2013 at 7:04 PM, Tim Leydecker <[email protected]
<mailto:[email protected]>> wrote:
Thanks for the hint to unified sampling!
Nice.
Here´s a mR unified sampling version with DOF.
*Switched to Gaussian (3) for Filtering.
*1/8/1/0.032 unified sampling settings...
*mia_lens_bokeh used 4 samples, should have been 6.
*global Ambient Occlusion Rays = 64, should have lived with 48 as before.
*mib_lens_clamp (0-1) added to get rid of some hard to filter hotspots.
MacPro2008//2.8GHz//7cores rendering in low priority
' INFO : RC 0.10 info : rendering statistics
' INFO : RC 0.10 info : type number per eye
ray
' INFO : RC 0.10 info : eye rays 15449600 1.00
' INFO : RC 0.10 info : reflection rays 61512093 3.98
' INFO : RC 0.10 info : refraction rays 1421904 0.09
' INFO : RC 0.10 info : shadow rays 4232441256 273.95
' INFO : RC 0.10 info : environment rays 3727286053 241.25
' INFO : RC 0.10 info : probe rays 3777812169 244.52
' INFO : RC 0.4 progr: rendering finished
' INFO : RC 0.4 info : wallclock 3:22:12.48 for rendering
I like the unified sampling, it is indeed easy to use and can be a real
speedimprovement!
The same image without DOF took ca. 45 minutes...
Cheers,
tim
On 25.02.2013 09:58, Arvid Björn wrote:
It's pretty easy, and it will change everything. Dof could even speed
up your render due to the more clever sampling scheme. I usually go with
mia-bokeh, stopped using
post-dof
quite a while ago. Similar thing with motion blur, it's a different
game with unified sampling.
