corona "vs" redshift

[piripi]

Dear everybody,

(jpg is compressed with 65%, but max file size is 1,5mb sorry)

just want to share my naive comparison test between amazing corona 7.2 and fantastic redshift V0.1.10 Alpha for Max.

Hardware:
i7 4930k @ 4,3ghz
Asus GTX 780 6GB (latest driver of NVIDIA)
32 GB Ram
win 8.1

first i want to point out, that this is NO “mac vs pc”,  “android vs iphone”  discussion…

its just a personal test i did this morning to see which render engine i could/should use for each project, or in which i should invest first in near future… (ciao ciao free public alpha)

i didn’t speed much time on the materials, i just tried to match it as close as i could in 5 min…
i’m very new to redshift, so the material conversion is not the best… and i don’t know how to translate the Fresnel from corona / vray to the “same” value in “facing bla bla ...” and “curve falloff”.

i used the same bitmap texture for “reflection color” and “reflection glossiness”.
i had the impression that, the corona looked more blurred, with the same texute, so i adjusted it a bit…

is there a way to convert this values better?

the lights are all corona / redhsift native area-lights, the size, color and intensity are equal.

hope this is interesting for you guys, again, BOTH render engines are amazing!!

i would be happy if somebody could point out, how to convert fresnel to redshift values, is there a simple rule / calculation?

my personal conclusion:
both are super cool, its really depending what you are going for…
after this test redshift would be my choice for animation and
corona for archvis, i’m sure both can be accomplished vis a vis

cheers
piripi

p.s. after finishing the renderpart, i realized its just 800px resoluion :( 
if someone is interested in the exr files let me know

[borisquezadaa]

What version of Corona did you use for comparation?.
Also very simple scene to be conclusive.

[piripi]

7.2 public

[Juraj]

Interesting :- ) Continue please.

Now you could move to some more complex (GI involving) scene.

Few websites btw state both fresnel IOR and Incidence angle reflectance (0° to 90°), I've even seen some calculator somewhere though can't remember where.
The most common non-metallic shader has frontal reflectance of 0.04, which corresponds to 1.52 IOR.

http://refractiveindex.info/ {angle of incidence is in top lowest graph}

Hey, can you post a screenshot of their shader UI in Max ?

http://docs.redshift3d.com/Default.html#I/Architectural.html

Interesting, they made their own Mix of Vray-ish (Roughness is dusty-ness and glossiness is roughness) with Maxwell (full fresnel n & K formula) down to PBR (Metal-ness which automatically decides whether to use color for specular or diffuse reflection).
It seems quite strange, but pretty robust. Everyone feels the need to re-invent wheel..

[piripi]

both materials

thanks for the explanation, i'll try to make it work

[piripi]

redshift related:
AHA, just found out, that if i change the IOR in refraction it has effects to the reflection, sorry this may sound stupid for people how know physics better than me ;)

or is there s simple way to convert ior to k? is this the same? confused...

[racoonart]

Those guys never think of the poor people who are doing conversion scripts. You can't convert material settings like these! But maxwell, fry, etc... are worse ;)

[piripi]

still confused ;)
but helps little bit:
quote of Adrian in reshiftforum:
https://www.redshift3d.com/forums/viewthread/3968/

By default RS the fresnel effect is computed as a dielectric index of refraction and of course by changing the index of refraction you change the reflection. I think the default fresnel should be conductor not dielectric since most of the time we create solid surfaces and metals which act as conductors not glass and water that are dielectrics.

Just switch the fresnel from dielectric to conductor and set the appropriate value for the coefficient of extinction.

Cheers

A

[Juraj]

That's pretty confusing explanation. Dielectric materials are also solids, not just liquids and gases and in CGI, we mostly created solids which are dielectric in nature (porcelain, plastics,etc..) not solids which are conductors (metals). Point is, both can be in solid form..

That's why "artist-friendly" developers (Like...Disney...) created far more logical division. Metals and Non-Metals. Metals (conductors) have specular reflection property that is best represented by complex Fresnel formula, including 'Refractive index' n and 'extinction coefficient' K. Dielectrics can be for CGI purpose represented completely fine with Schlick approximation (K is consider =1)

TL:DR Quick example: {Refractiveindex.org now has special category for 3D Artists !)

Plastic (PMMA): Dielectric material, IOR (n) = 1.49 {which is equal to frontal facing reflectance =0.38}
Aluminium : Conductor, IOR (n) = 1.07 K= 6.67 {cannot be input directly in Schlick models like Vray/Corona; Frontal reflectance = 0.9 }


All in, all out, the architectural shader in Redshift is extremely stupid and best example how engineers/programmers can't think of artist. Too bad because the engine looks great.

But let's go back to topic and post some comparisons ;- )

[Ondra]

As Juraj said.

Materials such as wood, plastic, plaster, paint, brick, concrete, fabric, etc. are from physics/optics standpoint MUCH closer to water and glass than to any metal.

[blank...]

Now do a comparison with a scene that eats 16 gigs of RAM.

[Ondra]

Now do a comparison with a scene that eats 16 gigs of RAM.

All i’m getting from those tests is that Corona needs something like 8x the rendertime to match the quality of Redshift, so i’m not at all sure why you’d want to bother with it.

These comparison threads are just doomed from the beginning ;)

[piripi]

Sharing is caring ;)

[Fibonacci]

Hi,

I think this comparison is not really equal, because the Redshift is GPU-base render engine. But looks good ! :)

[maru]

My car is fuelled by gas.

My car has electric engine.

Let's not compare their speed! :)

[Juraj]

My car is fuelled by gas.

My car has electric engine.

Let's not compare their speed! :)

Hi,

I think this comparison is not really equal, because the Redshift is GPU-base render engine. But looks good ! :)

And why would that be ?

Speed comparison is perfectly relevant measure. Does not necessarily mean it is ultimately better, but disregarding it because it seems unfair doesn't seem correct.
I wouldn't disregard comparison simply because some underlying cores don't match, BMW i8 (or Teslas) aren't compared to Toyota Prius either, but regular (fuel-driven) matching sport cars.

It's hardly ever objective, but that has to be taken with grain of salt for what it is. It still does provide informational value as long as it doesn't derided into "my dad can beat your dad" type of thing.

[maru]

(my post was an ironic reaction to Fibonacci's post)

[Juraj]

:/ sorry. The fact it's often used argument (almost in exact wording) led me to believe you agreed. My apology

[maru]

No problem. :) I haven't heard of such example, I just posted the first thing that came to my mind (which usually isn't a good idea).

Comparing speed of rendering engines sounds totally reasonable for me, whatever GI solutions or hardware they are using, as speed is one of the most important factors.

[Ondra]

Comparing speed of rendering engines sounds totally reasonable for me, whatever GI solutions or hardware they are using, as speed is one of the most important factors.

except that the speed differ so much based on scene and individual's preferences, that any comparisons I have seen were totally irrelevant. You would have to compare tens or hundreds of scenes to get any relevant data

[jjaz82]

i like these tests with new render engines :)
In my opinion for a true speed comparison you must use a complex indoor scene  with  various types of materials.
this scene is too simple.
congratulations and keep going  :D

[chopmeister]

Like Keymaster said, although I appreciate the effort, I find comparisons like these irrelevant. Testing render speed should be done with really complex scenes to get any decent results. What happens with difficult light scenarios or very complex materials? What happens with thousands of instanced vegetation objects? What happens with an interior packed full of glossy reflective materials? How does complex displacement impact the renderer? Lots of questions to be answered... :D

And also, while speed is an important factor, for a professional there are lots of other things to consider if you want to have a valuable comparison. Like pricing and licensing options, how good is the integration, etc. :)

[juang3d]

My question to you all that say that this type of tests are irrelevant...

How do you decide what render engine would you like to use for a project?

I'm not saying this tests are the perfect answer, but this could lead a lot of people to test things by themselves, it's like saying that I can't compare a Porsche vs a Smart.

Well I can, it all depends on what am I looking for, so probably if I want to spend very few fuel I will go to acquire the Smart, if I want to run and I want the luxury I'll go with the Porsche, it's not the definitive comparison, but i's ok for me, and maybe it may help others in their challenge of picking a car (or not hahaha )

So I find this tests pretty interesting, a more elaborated test may be welcome of course, with a real production scene like an interior or something similar, and of course every result in high resolution, but in the end, this is more info for the pocket, and I'll be doing my own tests for each project so I can decide the best engine for the job, but this is very welcome even when it's clearly biased :)

Cheers!!

[borisquezadaa]

Hmmm... a practical question. How was the learning curve between redshift and corona to you as user?.
One of the outstanding things in Corona was that for me the learning curve was "0". Thats speed!.
So even if redshift cames with lapdancers if i have to spend  a month trying to achieve decent quality and learning new workflow i would say thats a big NO.

[piripi]

Sorry for taking so long...

@ Boris
I think redshirt is more complicated, its a mix between mental ray, vray and Arnold i would say... Much more to tweak and to take care off...
This has pros and  cons... "Better" optimization -> shorter render time, but takes longer to setup... Corona vis a vis the opposite...

And to setting up materials feels more difficult in my opinion/current knowledge level... And in corona its just so much fun! But well, everything needs time to learn...

I really love the artist friendly corona, newer had so much fun rendering!

But when it comes to animation its a different thing for me, makes big difference waiting 6 days for rendering instead of 1-2 days...

But on the other hand, setting up a scene for still image like Speedy Gonzales, and than just let it render for the night...

Learning Corona felt like switching from mentalray to Vray... It was a big nice surprise how easy/fun thing could be...
This feeling i don't have so much with redhsift at the moment, but the speed is impressive, but it is also still in alpha and support is as great as in corona!

So again it really depends what the target is and and and ...
Unfortunately there is no "the one and only might holy cow" render engine... Sofar ;)

And competition makes a better product...

If i have more time i will make more complex test...

Cheers
Piripi

[juang3d]

@piripi One question, does RedShift works in a computer without GPU? (I know it's a GPU render engine, but iRay is also a GPU render engine but it can work with CPU, it can be slow, but it can work)

If the answer is yes, can you do the same comparison tests but just with the CPU?

One of the main reasons I abandoned GPU rendering is the cost, energy and hardware is a lot more expensive than using just CPU rendering, so no matter what the speed of redshift is for me, the GPU costs are too expensive to build a proper state of the art renderfamr with a proper amount of GPU memory :P
Anyways, i'm interested in knowing it's speed in just CPU mode if it can do it.

I think Corona needs some speed improvement for animation, my main projects now are animation related and while it's not crazy, it lasts aroung 30/40 minutes for an interior scene at 720p, and 60/80 minutes or more for a 1080p, wich is a lot, but the energy invoice is lower compairing it to when I rendered a project using the GPU's (same situation with GPU's was aroung 700€ of energy for two months, with just CPU's around 350€ for two months)

Cheers!

[Ondra]

If the answer is yes, can you do the same comparison tests but just with the CPU?

Even if possible, that would not be fair, as code for GPU is written entirely differently from CPU, so what works fast on GPU would not necessarily run fast on CPU.

[juang3d]

I don't want to be fair, I want to be practical, I'm not trying to criticize RedShift, I'm trying to compare two render engines under my case of use :)

Cheers!