Light tracing and pool

[FarbigeWelt]

How much time did this render take on what cpu?

It must have been 8-10 hours overnight on e5 2620 v4 cpu.
I play with the scene more, see if something better comes out of it.

Fox, thank you for answering!

Code: Select all

www.cpubenchmark.net
02. Sep 19		Intel Xeon E5-2620 v4 @ 2.10GHz		
			|		Intel Core i7-4770K @ 3.50GHz
			|		|		AMD Ryzen 9 3900X
Single Thread Rating	1547		2250		2918
CPU Mark		11354		10074		31827

Price			$363.45		$330.7		$669.99
Socket Type		FCLGA2011-3	LGA1150		AM4
CPU Class		Server		Desktop		Desktop
Clockspeed		2.1 GHz		3.5 GHz		3.8 GHz
Turbo Speed		Up to 3.0 GHz	Up to 3.9 GHz	Up to 4.6 GHz
# of Physical Cores	8 (*2)		4 (*2)		12 (*2)
Max TDP			85W		84W		105W
Yearly Running Cost	$15.51		$15.33		$19.16
First Seen on Chart	Q2 2016		Q2 2013		Q2 2019
# of Samples		101		9005		540

Intel-Xeon-E5-2620-v4-vs-Intel-i7-4770K-vs-AMD-Ryzen-9-3900X

[Fox]

This photon blotches are hard to get ride of

[epilectrolytics]

PhotonGi settings 10.09.2019.png

Now that's a maxed out setting table
But the result looks good though PathCPU & Metro together with caustic cache should render double caustics (+SDS).

I have been thinking a hybrid engine with PathCPU Metro for caustics+SDS only and PathOCL for the rest would be interesting

[Fox]

Getting closer, but there are this huge spots.

4h cpu hybridbackforward photongi

Code: Select all

screen.tool.type = IMAGE_VIEW
periodicsave.film.outputs.period = 7200
film.outputs.safesave = 1
path.pathdepth.total = 33
path.pathdepth.diffuse = 33
path.pathdepth.glossy = 33
path.pathdepth.specular = 32
path.hybridbackforward.enable = 1
path.hybridbackforward.partition = 0.5
path.hybridbackforward.glossinessthreshold = 0.0040000001899898052
sampler.sobol.adaptive.strength = 0
sampler.random.adaptive.strength = 0
sampler.metropolis.largesteprate = 0.10000000000000001
sampler.metropolis.maxconsecutivereject = 32768
sampler.metropolis.imagemutationrate = 0.10000000000000001
film.filter.width = 1.5
film.height = 3000
lightstrategy.type = "POWER"
film.width = 4000
film.filter.type = "NONE"
sampler.type = "METROPOLIS"
renderengine.type = "PATHCPU"
scene.epsilon.max = 0.35699999332427979
scene.epsilon.min = 0.00013830000534653664
film.opencl.enable = 0
path.photongi.caustic.lookup.normalangle = 9.9999997090929202
path.photongi.caustic.enabled = 1
path.photongi.caustic.updatespp = 25
path.photongi.caustic.lookup.maxcount = 1024
path.photongi.persistent.file = ""
path.photongi.indirect.haltthreshold = 1
path.photongi.indirect.usagethresholdscale = 2
path.photongi.caustic.merge.radiusscale = 0
path.photongi.indirect.lookup.normalangle = 9.9999997090929202
path.photongi.photon.maxdepth = 32
path.photongi.indirect.glossinessusagethreshold = 0.0040000001899898052
path.photongi.caustic.maxsize = 100000000
path.photongi.photon.maxcount = 100000000
path.photongi.caustic.lookup.radius = 0.001
path.photongi.indirect.maxsize = 0
path.photongi.indirect.lookup.radius = 0.019880000501871109
path.photongi.indirect.enabled = 1
path.forceblackbackground.enable = 0
filesaver.format = "TXT"
filesaver.renderengine.type = "PATHCPU"
native.threads.count = 14
renderengine.seed = 1
batch.haltspp = 0
batch.halttime = 0
film.imagepipelines.0.0.type = "NOP"
film.imagepipelines.0.1.type = "TONEMAP_REINHARD02"
film.imagepipelines.0.1.prescale = 0.80000001192092896
film.imagepipelines.0.1.postscale = 6
film.imagepipelines.0.1.burn = 10
film.imagepipelines.0.2.type = "CAMERA_RESPONSE_FUNC"
film.imagepipelines.0.2.name = "FUJICOLOR-PRO-400H.crf"
film.imagepipelines.0.3.type = "GAMMA_CORRECTION"
film.imagepipelines.0.3.value = 2.2000000000000002
film.imagepipelines.0.radiancescales.0.enabled = 1
film.imagepipelines.0.radiancescales.0.globalscale = 0.4
film.imagepipelines.0.radiancescales.0.temperature = 5400
film.outputs.0.type = "RGB_IMAGEPIPELINE"
film.outputs.0.index = 0
film.outputs.0.filename = "RGB_IMAGEPIPELINE_0.png"
scene.file = "scene.scn"

[epilectrolytics]

This looks great!
And I like the spots too (Laci was asking for lens flares anyways...)

[lacilaci]

Just by the way... here's how it looks faked in cycles

rendered in ~3 minutes...

It has few issues that could be solved, but often this would be good enough and fast enough... The flexibility rayswitching allows for is just amazing(not just for faking caustics..)

[Dade]

Just by the way... here's how it looks faked in cycles

They are not exactly hard to fake in LuxCore too:

with the not so small difference that they are true caustics instead of faked one.

[lacilaci]

Just by the way... here's how it looks faked in cycles

They are not exactly hard to fake in LuxCore too:


caustics.jpg


with the not so small difference that they are true caustics instead of faked one.

except you get no refractions and you get really weird darkening on meshes under water...

[Fox]

This is not a 3 minutes, but i found most interesting how much the HDR improves water.

[kintuX]

CPU PT
Halt Time: 90 s
Same geo, same IORs, colours HEX, default Sun & Sky...

LuxCore (w/ light tracing and all caches ON)

Corona ("Fast caustics solver" & "Generate caustics from environment")

About "Too Dark & Too Bright"... From real world I'd say LuxCores reflected caustics are more correct, it's only too much loos in refractive volume under glass (my old enemy). I suspect there's a mistake somewhere about how gamma is handled. Look at the corners how light bounced in Lux version gets darker Of course I can be wrong.
Also, Corona is much simpler, faster to set & reach the goal... while user also gets a caustic pass to fiddle with in post (if needed).