True GI: Corona vs V-Ray 3.0

[fobus]

As we know Corona have nice speedy GI.

With release of V-Ray 3.0 and loud claims in speed increase it was interesting how really quick V-Ray 3.0 GI. As Chaosgroup claims they are sped up BrutForce GI in 3x-5x times against V-Ray 2.x. But is it enough to compare it to Corona GI (I mean PathTracing GI engine)? Let's take a look at tests:

Models: Simple scene with box with small pinhole and two teapots in it.
Materials: Grey material with blurry (0.2) fresnel reflections.
Lighting: Only white background with no lights.
GI: V-Ray - Bruteforce+Bruteforce, 25 bounces; Corona - PT+PT 25 bounces.
Software: 3ds max 2014, V-Ray 2.4, V-Ray 3.0 Demo, Corona A5.

[Alessandro]

...and the winner is...... CORONA!!! ;)

btw, very very basic test, to be correct maybe it need to be a little more complex, with different kind of material and lights, in works time i've never produced a gray box render ;)

[fobus]

I wanted to test only GI. Unfortunately Vray demo resetting settings time to time so it is nearly impossible to test real scene with it. Of course Vray has muche more materials and options, but GI speed in Corona (bruteforce GI of course) is better.

[andreupuig]

Don't know if Brute Force has been changed with the new version, but assume that 25 bounces are the same as 25 PT is total a mistake.
Brute Force's subds are totally depending on how you are setting the DMC. If you don't optimize the DMC this test is useless.
You can say that vray is longer than Corona in setting correct values, but you cannot compare speed in the way you made.

[Utroll]

My guess is that Vray manage a lot more situations (sss, etc) may be that makes the whole algorithm tracing more heavy. Although as scene is identical it should at the end not being so behind... shaaame !
Second point is that Vray comes from ages, and faking was mandatory according to processors performances at that time (who would have used Maxwell in 2005 for instance ahah) , and may be the whole software is now based on outdated design ? Rewritting in depth is may be needed, but actually I thought they did.

I know that little aperture is a good way to test a tracing engine but how would it be by using portals in both render with your scene ?

[fobus]

Don't know if Brute Force has been changed with the new version, but assume that 25 bounces are the same as 25 PT is total a mistake.
Brute Force's subds are totally depending on how you are setting the DMC. If you don't optimize the DMC this test is useless.
You can say that vray is longer than Corona in setting correct values, but you cannot compare speed in the way you made.

I mean exactly 25 bounces of light. PT samples and BF subdivisions are out of specs in table. If You can do this test "more correctly" do it please.

I know that little aperture is a good way to test a tracing engine but how would it be by using portals in both render with your scene ?

I don't need portals in this kind of test. I need real GI performance. In my real projects I see this difference a lot, but it needs to be tested to prove that it's not placebo. So I see it real now.

[gracelorn]

Faster or not, VRay still has crappy shading and colour mapping that haven't changed in years. All materials end up looking like plastic, and the lighting is bleak and flat. To me, it's VRay's fundamental flaw. They need to rewrite their shading core (if there is such a thing), not add another type of SSS material or funny buttons to the interface. I can always tell if a picture was rendered with VRay, its so recognizable (in a bad sense) and is never completely photo-real. Corona, on the other hand, produces stunning images in comparable times, and sometimes it's hard to tell whether you are looking at a photo or a render. That's why I love it so much ;)

[ecximer]

gracelorn +1.5

[cecofuli]

The big problem in VRay 2.0 was the sub-pixel mapping!!! I talked with Vlado so many times in the past! He told me that he could not do anything about this problem!  For interior rendering it wasn't possible render with sub-pixel mapping OFF (or the render time was 3 or 4X with this option OFF) . And, the difference, in many cases was really obvious! Because sub-pixel mapping try to remove the bright part in the shading. This is why VRay looks, sometimes, flat and not real!
With VRay you can obtain very nice, photo realistic renders, but you have to spend time with lights, shading and, the boring part, DMC, AA, subdivs tweaking. And you know, time is money. I prefer to buy 2 more PC and don't have to think about subdivs, IM problem, AA problems etc...

In the Chaosgroup forum there are tooooooons of threads about problems with noise, AA, sub-pixel mapping. I remember a thread from BBB3 ( 40 pages long) and no one official reply from Vlado about that noise solution (because there isn't a solution!!!)

http://forums.chaosgroup.com/showthread.php?54547-About-noise-in-general&highlight=sub-pixel

Also, look at that threads:

http://forums.chaosgroup.com/showthread.php?76373-V-Ray-Render-Optimization-an-in-depth-Guide-(call-for-Before-Afters)
http://forums.chaosgroup.com/showthread.php?71724-Understanding-DMC-Sampler&highlight=sub-pixel
http://forums.chaosgroup.com/showthread.php?68284-Cleanup-noise-in-Elements-Reflection-and-Shadow-Also-LOOONG-render-times&highlight=sub-pixel
http://forums.chaosgroup.com/showthread.php?71891-Sub-pixel-ON-OFF-phisical-INCORRECT-noise-and-rendertime&highlight=sub-pixel

Sorry, but I don't have time to read 200 pages about noise, magic AA, subdivs formula and, after that, we didn't have a good solution.

[fobus]

With so great GI in Corona it, unfortunately, has a flawn in direct illumination. Iа You want to use huge amounts of lights it is nearly impossible to deal with noise free picture. Unfortunately algorithm of lighting in Corona can't produce clean direct lighting as I understood from Ondra explanation (http://forum.corona-renderer.com/index.php/topic,2306.msg17090.html#msg17090). Even with no GI we can't get rid of noise with huge amounts of lights. No way. I hope Ondra will deal with it somehow in future...

[gracelorn]

The thing with VRay is that it is already used by many studios worldwide in their well established production pipelines, and ChaosGroup cannot change things radically, they have to stick to what people are used to. This is why they only add or try to improve certain features, but they cannot change or remove them completely because it may affect the usual workflow for some of their customers. Also, I think it sells pretty well as it is, so why bother changing things dramatically?

[maru]

What kind of image sampler (multisampling) did you use for you comparison? I think it would be best to use fixed@1 subdiv in Vray and 1 pass in Corona for an honest test.

[fobus]

... it would be best to use fixed@1 subdiv in Vray and 1 pass in Corona for an honest test.

It's fair. I've made a test with Fixed image sampler with 1 subdivision in V-Ray and 1-pass in Corona with 2000 PT samples. Roughly the same noise in 2x times quicker in Corona.

And another test with just changed material to steel.

[dfcorona]

I've been test Corona and Vray 2.4, I'm not sure what exactly I'm doing wrong in corona, but I can't really get even close to the speed of vray.  I am setting it up for biased rendering, because unbiased is still to long per frame.  I am testing an exterior scene. I set corona to progressive and path tracing & HD Cache, Because I can't find IR.  I set Vray to IR & Light Cache.  I set corona to  3min and receive quite a noisey image.  I let vray go and I get a really high quality image in about a minute.  Is corona fast only compared to vray brute force?

[fobus]

Corona is fast in brute force GI. It has not other GI options like Irradiance maps or light cache. It has HD cache to accelerate secondary GI only but GI itself is true brute force GI. In this scenario it is fast. You have to compare the same GI quality between Corona and Vray to be correct.

[lacilaci]

I wonder if irradiance caching will come back to corona or not...

[dfcorona]

okay, what happened to IR though in Corona, seems it use to be in there.  And I thought I read somewhere that it was going to be put back in.

[fobus]

I wonder if irradiance caching will come back to corona or not...

I hope it will not...

[Ludvik Koutny]

I wonder if irradiance caching will come back to corona or not...

I hope it will not...

It will, but quite simplified :)

[fobus]

Anyway I like Corona for brute force GI :)

[Juraj]

IR can be tremendously good tool, but people falsely consider it to be much faster than pure path tracing because they get "nice&clean" pictures fast. The fact is, most people
can't really see the huge quality difference you can get under different setup IR parameters. Once you have a complicated scene, with retrace thresholds and additional Detail enhancement,
you easily get render times that rival Brute Force in Vray because you need to solve a lot of artifacts appearing but still keep certail detail away from blurrying.

IR's strength is flexibility, not speed per se. Most people think it's fast, because they have it by default very low, that's what they're used to seeing, it's what gives them quite shitty results but they will never care to notice. It is only fast as long as it's extremely unprecise, and interpolated to avoid artifacts, but that's is exactly where it is best for animation purpose of static scenes.

My only qualm in Vray is that IR's samples are also governed by DMC, something that makes it harder to manage in complicated interiors because you weight in GI together with material shading,and all else. Also, maybe too much potato paramaters that nobody will ever understand so everyone just changes HSPV and interpolation leading. Detail Enhancement (local brute force bounce thingy) is nice touch.

To be honest, I lately change between Vray and Corona seemlessly depending on what I want in particular projects. Animations are of course, completely in domain of Vray because of IR, but even stills get evenly distributed. The biggest stregth of Corona is incredibly nice, precise and believable GI. Even pure Vray's BF+BF isn't such, the details are slightly unnatural in that they always go into dark tone regardless of soem secondary reflections or etc..

All in, all out, I wouldn's say you'll get better render times in Corona compared to Vray. I would say these overstatments come from overly enthuasiastic users getting such times from easy scenes with easy setups. I would say in similar quality, the times can be close to equal at best. But it's better to look at the whole setup time/quality/speed not just speed alone to make your complete opinion.

[Fatal1ty]

... you can not compare in such a way. Actually you put the V-Ray image sampler to such a stress. The noise will not go away fast. There is a shading rate parameter ... just make it 16 or higher and you will get up to 50% faster render with much less noise ... cheers :)

[gabrielefx]

I think that Vray 3.0 is not 100% re coded from scratch using Embree, is not yet optimized. Is not optimized for Cuda too.
Corona seems to use 100 % of the Intel kernel.
I read that a company fused Embree and Cuda to create a mix of these two worlds.
At this point we have to wait to see optimizations, faster renders and less noise only if Intel or Nvdia engineers will develop better algorithms.

[biolit]

fobus: which hardware did you use on the test?

[ondrike]

I´m really not sure about fobus´s render times

my example is vray3 BR+BR

[fobus]

fobus: which hardware did you use on the test?

Intel i-7 3930k @4.4GHz

I´m really not sure about fobus´s render times
my example is vray3 BR+BR

Can You post this scene to compare please?

[ondrike]

I´m on i7 2600k overclocked at 4,5Ghz and my time is somewhere around 3 mins +/-. I dont have scene already. That was just a quick test, so i deleted it

settings are BR+BR,  26 Samples

[fobus]

I´m on i7 2600k overclocked at 4,5Ghz and my time is somewhere around 3 mins +/-. I dont have scene already. That was just a quick test, so i deleted it

settings are BR+BR,  26 Samples

Is there any light source in tiny window in Your scene? I haven't had any lights except environment white color at background. If You placed light in window...it's not fair to compare to anvironment only (without any lights and portals).

[biolit]

Is there any light source in tiny window in Your scene? I haven't had any lights except environment white color at background. If You placed light in window...it's not fair to compare to anvironment only (without any lights and portals).

I'm sorry, but I think that this test is nonsense.

[fobus]

I'm sorry, but I think that this test is nonsense.

This test is for comparing Bruteforce GI only. No direct lighting, but GI. So light in window is cheating in this case.

[ondrike]

My mistake !! apologize... I didnt noticed that no light, just environment with white color... My bad, sorry one more time :)

[CHRiTTeR]

I'm sorry, but I think that this test is nonsense.

This test is for comparing Bruteforce GI only. No direct lighting, but GI. So light in window is cheating in this case.

Not if you make it a portal

[voltron7]

in the last 2 steel renders, why has the floor/wall line vanished from corona's image?