Author Topic: what is the speed improvement Corona 2 vs 1.74??  (Read 4505 times)

2018-07-16, 19:28:19

rampally

  • Active Users
  • **
  • Posts: 208
    • View Profile
what is the speed improvement Corona 2 vs 1.7.4?? I mean overall render time taken  to achieve the desired noise level for example3%

2018-07-16, 23:01:03
Reply #1

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
In the 2.0 (NL=3%) the image is more clean that 1.7 for the same NL.
So, don't use NL to compare images between these two version.  Use time limit instead =) and use "your eyes" to see the noise level .

2018-07-17, 00:17:31
Reply #2

dj_buckley

  • Active Users
  • **
  • Posts: 876
    • View Profile
In the 2.0 (NL=3%) the image is more clean that 1.7 for the same NL.
So, don't use NL to compare images between these two version.  Use time limit instead =) and use "your eyes" to see the noise level .

So effectively what your saying is that 3% in 1.7 isn’t actually 3%?  Seems a bit silly - surely 3% is 3% regardless of version if you’re rendering exactly the same shot/scene

Actually, i’ve seen a number of posts saying NL isn’t useful for a variety of reasons but for me, if you want a consistently clean suite of images across a project - it seems the most logical route to take.

If someone uses NL as the baseline for the final images then why should they have to switch to a different method that they wouldn’t normally use in order to benchmark speed increases?

2018-07-17, 00:30:37
Reply #3

PROH

  • Active Users
  • **
  • Posts: 1219
    • View Profile
Well according to other posts from the Devs cecofuli is right. As I understand it it comes down to 2 things: how the noise is calculated, and how "visible" the noise is.

2018-07-17, 01:15:05
Reply #4

dj_buckley

  • Active Users
  • **
  • Posts: 876
    • View Profile
Sorry I wasn’t saying he was wrong.  I just find it hard to comprehend haha.  It’s either 3% or it’s not

2018-07-17, 02:23:02
Reply #5

PROH

  • Active Users
  • **
  • Posts: 1219
    • View Profile

2018-07-17, 03:06:24
Reply #6

Njen

  • Active Users
  • **
  • Posts: 557
    • View Profile
    • Cyan Eyed
Of what ever metric they using to calculate Noise Level.

2018-07-17, 03:37:24
Reply #7

PROH

  • Active Users
  • **
  • Posts: 1219
    • View Profile

2018-07-17, 06:19:07
Reply #8

dizartwork

  • Active Users
  • **
  • Posts: 25
    • View Profile
in my experience
1.74:
3 minutes 6 pass 7.06%
5m 8p 7.06%
7m 11p 4.78%
10m 14p 4,78%
11m 16p 3.97%
14m 19p 3.97%
15m 21p 3.48%
17m 23p 3.48%
19m 26p 3.14%

2.0:
3m 6p 6.59%
5m 9p 6.59%
7m 11p 4.48%
10m 15p 4.48%
10.2m 16p 3.71%
12m 18p 3.71%
14m 21p 3.25%

Faster clean. About 20-25% speed up

2018-07-17, 11:19:42
Reply #9

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
There are various speed improvements in v2 compared to 1.7.4:
-new embree
-new visual studio
-new way adaptivity works
-more at https://corona-renderer.com/blog/corona-renderer-2-for-3ds-max-released/

Unfortunately the noise level is a complex topic. I hope it will be simplified one day. Here are the facts:
-the same noise level (e.g. 3%) may give you different results in different versions of Corona
-the "%" sign at the end of noise limit value is added just for convenience, in reality it's not "percent of something" or any easy to measure unit
-the same noise level value may give you different noise in the rendering, even for the same scene, for example depending on what is visible in the frame - is it covered with objects? or is there a lot of solid color visible, and just one small object, or maybe a single object against sky?
-noise level will give you consistent quality for all frames if the scene does not change much throughout the animation (for example it will work fine if the camera is moving inside an apartment, but it may be a bit inconsistent if the camera is moving inside an apartment, and then goes out and you can see mostly the sky)

Generally if the noise limit does not bring you consistent results (for example the visible noise changes across different frames), I would recommend using pass limit instead. Pass limit means exactly the same quality for each frame.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2018-07-17, 11:51:43
Reply #10

rampally

  • Active Users
  • **
  • Posts: 208
    • View Profile
Thanks All And dizartwork this is what i am looking for And also thanks maru for additional information.
« Last Edit: 2018-07-17, 12:06:01 by rampally »

2018-07-17, 14:23:15
Reply #11

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
Maru, you wrote: "Pass limit means exactly the same quality for each frame. "
But, in my opinion, same Pass = same AA, but it's doesn't mean "same quality"

for example:
(*)  in an interior scene, you have a 50 Passes (in 30 minutes), perfect AA, but the GI isn't very good (NL= 10%)
(*)  in an exterior scene (easier for GI), with the same 50 Passes (in 5 minutes because you have less GI), you have obviously the same AA, and perfect GI(NL: 3%).

So, if you want the "same quality" (read as  same noise level) , Pass limit isn't the road.

You also wrote " if the noise limit does not bring you consistent results for example the visible noise changes across different frames"
But, Noise limit is a choice to have consistent result and it helps to have the same NL across the frames =)

Or am I wrong? =)




2018-07-19, 18:28:20
Reply #12

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
Maru, you wrote: "Pass limit means exactly the same quality for each frame. "
But, in my opinion, same Pass = same AA, but it's doesn't mean "same quality"

for example:
(*)  in an interior scene, you have a 50 Passes (in 30 minutes), perfect AA, but the GI isn't very good (NL= 10%)
(*)  in an exterior scene (easier for GI), with the same 50 Passes (in 5 minutes because you have less GI), you have obviously the same AA, and perfect GI(NL: 3%).

So, if you want the "same quality" (read as  same noise level) , Pass limit isn't the road.

You also wrote " if the noise limit does not bring you consistent results for example the visible noise changes across different frames"
But, Noise limit is a choice to have consistent result and it helps to have the same NL across the frames =)

Or am I wrong? =)

Well, in theory the noise limit should give you the same noise level for all frames. But sometimes it is not the case. Then I think using pass limit is the best solution.
I think in both cases it is the safest to find out which frame is the most problematic (has most noise for most time) and then set the pass or noise limit based on that. In worst case you will end up with some frames being unnecessarily over-sampled.
But with the noise limit it can easily go wild. I am attaching two examples. The scene is basically a noisy object, and the camera is moving along it so that on frame one there is sky visible only, and on frame 100 the frame is covered with the noisy object. If we set noise level to some high ("draft quality") value like 10, the first frames where only a small triangle is visible will render super quickly (5 passes only are enough), and then the rendering will gradually become slower (more than 5 passes will be needed), but it will be pretty fast, because the quality is low. This is expected. But if we set some "production" noise level like 2, then at the beginning we will render just 5 passes, which will render the object noisy (even though the reported noise level is <2 !), and then gradually the render times will become higher (as we need 5 passes, 10 passes, 15 passes, etc....) until the render time will become really huge. (I tried rendering the last frame where the whole view is covered and stopped after 30 passes because it did not make much sense to render further with the noise level still over 10).

Of course this is an extreme case, but it is similar to what I described before (camera moves from interior to exterior). With the pass limit you will get consistent quality per each frame in this case - as exactly the same GI and direct light samples will be cast (more or less - as we also have adaptivity, but that's a different story ;) ).

Also, I am absolutely NOT saying that the noise limit is useless! It is perfectly usable in cases where the scene doesn't change much throughout the animation - e.g. if the camera moves inside the interior only, or if the moving objects do not change their positions that much.

The scene + animation frames are here if anyone wants to play with it: https://www.dropbox.com/s/bbpvm3e5kip8caa/noiselimit.zip?dl=0
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2018-07-19, 19:11:53
Reply #13

cecofuli

  • Active Users
  • **
  • Posts: 1577
    • View Profile
    • www.francescolegrenzi.com
Ahhh.. maybe I have underatood!Yes, your example is an extreme situation.
It's like to render an object in the sky. In this image (see the attachment) , NL=2.
Usually is good, but because 80%is "sky" (it's obviously noise free), NL=2% isn't enough (the cloud is noisy)
So, it's better to use Passes.
Do I have understand correctly? =)

2018-07-20, 15:21:04
Reply #14

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12754
  • Marcin
    • View Profile
Do I have understand correctly? =)
Yup :)
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us