Author Topic: 1.3 vs 1.4 speed  (Read 5858 times)

2016-05-07, 01:49:51

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
I was expecting 1.4 to be much faster when rendering ( with adaptive, not Denoise ). I set the scene to render at 100 passes, same settings but 1.4 has adaptive. 1.3 did it in 4min09sec vs 1.4 at 8min06sec. Now maybe it's something like you can get a clearer image at 50passes or something, but this is very confusing it just seems much slower, would of thought it would of rendered way faster.

2016-05-07, 07:26:31
Reply #1

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
I don't know what I'm doing wrong, but I still can't match what Vray is capable with. Vray 1min32sec, and with less noise. Especially check the highlights on the chrome hose object.


2016-05-07, 10:28:27
Reply #2

vansan

  • Active Users
  • **
  • Posts: 47
    • View Profile
Hi.
Latest Corona with adaptiveness show more constant noise levels all over the image and +15% sampling speed improvement on my scenes.
75 passes is almost enough for any interior render now, as denoiser is really good even at 0,6 value.
I'm happy with all the improvements done by developers.
Using Corona instead of Vray is not about speed, it's about much simplier workflow and much more predictable results with less time spent on tests.

2016-05-07, 11:08:42
Reply #3

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 9032
  • Marcin
    • View Profile
I was expecting 1.4 to be much faster when rendering ( with adaptive, not Denoise ). I set the scene to render at 100 passes, same settings but 1.4 has adaptive. 1.3 did it in 4min09sec vs 1.4 at 8min06sec. Now maybe it's something like you can get a clearer image at 50passes or something, but this is very confusing it just seems much slower, would of thought it would of rendered way faster.
The proper test would be setting time limit and checking the quality of both images rendered within the same time. According to all of my tests, I always got better results with adaptivity on.

2016-05-07, 11:10:35
Reply #4

Juraj Talcik

  • Active Users
  • **
  • Posts: 3611
  • Tinkering away
    • View Profile
    • studio website
I was expecting 1.4 to be much faster when rendering ( with adaptive, not Denoise ). I set the scene to render at 100 passes, same settings but 1.4 has adaptive. 1.3 did it in 4min09sec vs 1.4 at 8min06sec. Now maybe it's something like you can get a clearer image at 50passes or something, but this is very confusing it just seems much slower, would of thought it would of rendered way faster.
The proper test would be setting time limit and checking the quality of both images rendered within the same time. According to all of my tests, I always got better results with adaptivity on.

But is there any catch ? Why would it also do less passes ? Does it mean the noise is better but less AA has been done ?
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2016-05-07, 11:36:25
Reply #5

sprayer

  • Active Users
  • **
  • Posts: 606
    • View Profile
I don't know what I'm doing wrong, but I still can't match what Vray is capable with. Vray 1min32sec, and with less noise. Especially check the highlights on the chrome hose object.
vray always do less noise that why i dont like it, it filter image do more plastic, it's ok for interior but it's not realistic, as for chrome in corona you set less glossiness than vray, and highlight at teapot not realistic, also i think in vray turn on absorption. Vray plus is only speed. But then you will set scene parameters corona gives you more speed with simplicity and IR, actually there is nothing to setup ))
As for speed of coronas can you tell how many pass was, in 1.4 image less noise at shadows, and do you try denoise feature? it will boost speed if you render in 1.3 for example to noise limit 2, you can render in 1.4 to 4-6 and use denoise, it will be faster and cleaner

2016-05-07, 13:49:37
Reply #6

SairesArt

  • Active Users
  • **
  • Posts: 676
  • Pizza | The Cheesen One
    • View Profile
    • SairesArt Portfolio
I was expecting 1.4 to be much faster when rendering ( with adaptive, not Denoise ). I set the scene to render at 100 passes, same settings but 1.4 has adaptive. 1.3 did it in 4min09sec vs 1.4 at 8min06sec. Now maybe it's something like you can get a clearer image at 50passes or something, but this is very confusing it just seems much slower, would of thought it would of rendered way faster.
The proper test would be setting time limit and checking the quality of both images rendered within the same time. According to all of my tests, I always got better results with adaptivity on.

But is there any catch ? Why would it also do less passes ? Does it mean the noise is better but less AA has been done ?
There is the adaptivity recalc penelty, but only very small. The main diffrence comes from the priority as far as I see it. If you look at the CInfo_SamplingFocus Render element you will see hard to sample part of the image (glossy refractions, glossy reflections etc...) get some times 200% of the sampling focus of the average of the scene. Since these are harder to crunch, the "preseved" render speed goes down, as you sampling these hard to sample places extra hard. But in reality without adaptivity, the image would get it's fixed samples per pixel and those hard to crunch areas are not actually crunched harder than the easy parts. Thus the loss in rendered passes, but overall a uniform and thus faster to render image.

edit:Funny enough just as I typed it, I found the dev confirming with my words in the daily builds thread.
https://corona-renderer.com/forum/index.php/topic,7239.msg76367.html#msg76367
« Last Edit: 2016-05-07, 13:53:02 by SairesArt »

2016-05-07, 15:54:39
Reply #7

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 8889
  • Turning coffee to features since 2009
    • View Profile
But is there any catch ? Why would it also do less passes ? Does it mean the noise is better but less AA has been done ?

It actually even improves AA. If half of your image is just directly visible environment map that requires only single sample (because of image filtering) to get the final result, it will now not get unnecessarily sampled.  Oversampling environment map does not improve image quality, but it is fast. So diverting the effort elsewhere will make the image converge faster, but it will be slower according to some internal technical metric (passes) that does not show directly in the image. The solution is to simply stop using passes, and use render time and noise level to argue about image quality.
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2016-05-07, 15:57:54
Reply #8

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 8889
  • Turning coffee to features since 2009
    • View Profile
There is the adaptivity recalc penelty, but only very small.
just looking at rays/s it actually seems that 1.4 is faster even in this brute force manner (including recalc in 1.4) because ryuu did some microoptimizations ;)
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2016-05-07, 16:33:15
Reply #9

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
Hmmm... Don't know getting same problem with very large scene, takes vray about 10min a frame and corona about 18min.  Tried denoise, but This shot has a lot of grass and denoise makes it look like a horrible oil painting, tried different settings.  Quality wise both look the same. I would love to use corona for an animation and see how it goes but at the huge hit of time and money it would take to render an animation we cannot.  Unless I'm missing some setting here. The interactivity is really great.

@Sprayer, not sure I agree, I rendered this with Vray using Brute force + Brute Force, I didn't use IR or even Lightcache. Even on the huge scene I used. If I used IR I'd probably be looking at 30sec for the vray render. so far I cannot get denoise to look good on a heavy vegetated scene. works okay, on some of the less detailed stuff.

2016-05-07, 17:17:08
Reply #10

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 5711
  • Let's move this topic, shall we?
    • View Profile
    • My Models
It's not very fair to compair Corona vs Vray, when your Vray version clearly shows that its ray depth is much lower and it doesn't show internal reflection at all. Who knows what else is "cheated" there? Just saying, hopefully this doesn't triger yet another flame war :]

p.s. it looks like it's inevitable after each major release to have those v1.x vs v1.x+1 topics spawning :]
I'm not Corona Team member. Everything i say, is my personal opinion only.

2016-05-07, 17:26:13
Reply #11

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
It's not very fair to compair Corona vs Vray, when your Vray version clearly shows that its ray depth is much lower and it doesn't show internal reflection at all. Who knows what else is "cheated" there? Just saying, hopefully this doesn't triger yet another flame war :]

p.s. it looks like it's inevitable after each major release to have those v1.x vs v1.x+1 topics spawning :]

I can't see choosing one without having a 1 vs 1. when I buy a car I compare. So your bringing up good points, let me try to adjust ray depth and such to match, I'm not favoring one, but want to compare each, just like that car. if it performs the same and looks just as good for way less time and money...... well.... to many people get to sensitive when it comes to comparing....

2016-05-07, 17:30:22
Reply #12

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
@romullus

How do I turn off internal Reflections like you say vray has, I want to match vray and see the rendertime.

2016-05-07, 17:32:42
Reply #13

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 5711
  • Let's move this topic, shall we?
    • View Profile
    • My Models
I don't think that it's possible to turn off internal reflections in Corona. You have to turn them on in Vray, i guess.
I'm not Corona Team member. Everything i say, is my personal opinion only.

2016-05-07, 17:33:11
Reply #14

denisgo22

  • Active Users
  • **
  • Posts: 675
    • View Profile
I was expecting 1.4 to be much faster when rendering ( with adaptive, not Denoise ). I set the scene to render at 100 passes, same settings but 1.4 has adaptive. 1.3 did it in 4min09sec vs 1.4 at 8min06sec. Now maybe it's something like you can get a clearer image at 50passes or something, but this is very confusing it just seems much slower, would of thought it would of rendered way faster.

it may well be that enabled Adaptivity render of the one separate pass some slower, but on the other hand there is no need to render at additional amount of passes to obtain high-quality render
without visible noise, and perhaps the overall rendering time remains
same as in 1.3 version, BUT Adaptivity greatly reduces render time in situation's with Hard DOF and MB effects, and in cases where it was impossible to get rid of visible noise on highlights
with DOF and MB effects in previous versions, which is very important for close-up's render's///

2016-05-07, 17:45:50
Reply #15

denisgo22

  • Active Users
  • **
  • Posts: 675
    • View Profile
Hmmm... Don't know getting same problem with very large scene, takes vray about 10min a frame and corona about 18min.  Tried denoise, but This shot has a lot of grass and denoise makes it look like a horrible oil painting, tried different settings.  Quality wise both look the same. I would love to use corona for an animation and see how it goes but at the huge hit of time and money it would take to render an animation we cannot.  Unless I'm missing some setting here. The interactivity is really great.

@Sprayer, not sure I agree, I rendered this with Vray using Brute force + Brute Force, I didn't use IR or even Lightcache. Even on the huge scene I used. If I used IR I'd probably be looking at 30sec for the vray render. so far I cannot get denoise to look good on a heavy vegetated scene. works okay, on some of the less detailed stuff.

About TIME and MONEY vs Quality and Realism this is very old question:)

2016-05-07, 17:54:39
Reply #16

dfcorona

  • Active Users
  • **
  • Posts: 116
    • View Profile
What is the equal setting to changing Color Space Inverse Gamma in vray HDRI to intensify lighting of the HDRI?