Forums » CentiLeo for Cinema 4D
Pages: 1
Loss of performance.
Performance loss of about 30% in the version up to 0.680
 
User  Posts: 9
Oct 5, 2024 23:18
I would like to know the experience of other users if the same thing happens to them as to me.

From version 0.680 onwards I have experienced a considerable performance drop of around 30% in render speed and in some scenarios even 40%.

Forcing me to give up the following versions since the speed/quality ratio was the main virtue for which I chose Centileo for my projects. And with this loss it is no longer an attractive option in favor of other options.

Is the same thing happening to you?

On the part of the creators of Centileo, any explanation that justifies it?

Thanks in advance.
 
Administrator  Posts: 933
Oct 6, 2024 09:08
Hi Jiv21! Did you lost performance in 0.69 compared to 0.68? Can you compare them side by side? Have you tested the very-very last version of 0.69 that was updated a week ago?
What kind of scenes did you test? Maybe some very
In my comparisons 0.69 and 0.68 are almost equal, with just small difference.
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 6, 2024 11:40
Hi Kirgman.

Thanks for your quick interest.

While it is true that the render quality improves in 0.697 with respect to 0.680, I leave you some sample images in which the loss of performance (in some cases) reaches twice the time.

I noticed the performance loss immediately with version 0.690 and the situation has been repeated in all intermediate versions up to 0.697.

I also observed some time ago that when exceeding the resolution of 4000X2000, not always, but it can be the case that the engine makes errors in part of the image. At the moment, it always happens on the right side.

In my case I have to render about 50,000 images on average.

Therefore the importance of what I am saying. A difference of 5 seconds between one version and another, implies almost 70 hours more rendering time.

It is an eventuality (relatively assumable) but above that ....

I'm sorry I can't share the scenes so you can investigate it.

In the name of each image you will find the resolution and the Centileo version used.

The computer configuration is:

AMD Ryzen 9 7950X3D
Nvidia RTX 4090
128 GB DDR5 at 6,400Mhz running at 3,600Mhz
MB MPG X670E CARBON WIFI

I have another station of similar characteristics with the same results.
Edited: jiv21 - Oct 6, 2024 12:11
 
Administrator  Posts: 933
Oct 6, 2024 12:22
I have fixed the bug with right side artefact on highres image (over 4096) just 10 minutes ago and now compiling an update with the fix.

As for the speed loss then nothing special was done that could influence the speed. Just more shaders were added that don't influence on performance in my tests.
The issue can be some tricky small thing that was done in some internal configurations that resulted in some issues for your particular scene.
Just anything could make influence.

Can you please collect some minimal scene maybe without your production staff that reproduces the issue? In that case I can investigate and fix very quickly. I can sign the NDA and guarantee no any leaks.
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 6, 2024 13:02
Thanks again Kirgman.

I am quite busy at the moment finalizing a project. But as soon as I have a moment, I'll check a scene for you to study it.

Comment that the performance problem has been reproduced in each and every one of the scenes I have created. So that is the reason for my first question in the forum. If I had detected that it only affected one scene, I would have thought about it before publishing in the forum (sowing the doubt).

I respect your work very much.
 
Administrator  Posts: 933
Oct 6, 2024 13:57
Thank you! Btw before I upload the fix with artifact to fix it at the moment: either switch off the bloom or make resolution a multiple of 128.
Your scenes are similar - they are stadiums. Nothing was changed internally with geometry handling. But probably something is wrong on my side with multi-thread waiting that was minimized (that's internal thing not for user thinking).

Currently I have run several 5K-10K renders on 0.68 and 0.69 with one or two RTX3080. And the older 0.68 is not always faster. But thats the render of the car interior and spherical camera. The usual difference is 2%.
Will continue some other benches.

Also what's your render Iteration Size and number of render Iterations?
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 6, 2024 15:02
Thank you for the clarifications.

Attached is the Centileo configuration.
 
Administrator  Posts: 933
Oct 6, 2024 15:08
jiv21, btw, do you use Bump maps or Normals maps connected to CentiLeo material? And what type exactly?
This can be the reason because I have fixed the issue of using the bump map with gradient node. And this makes the bump map more expensive and also do the same for normal maps whic could be avoided if we split them to different ports.

But anyway working with a scene itself helps better because we may open internal performance counters for every little aspect of the system and compare with the previous version.
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 6, 2024 15:31
I usually use normal maps. I do not usually use relief maps and on rare occasions displacement maps.

As soon as I have some time available I will send you a scene.
 
Administrator  Posts: 933
Oct 6, 2024 15:40
Ok, for the moment the first candidate on this problem is normal map which has become more expensive. Will test it once get a resource from you. Thank you for reporting an issue!
CentiLeo Chat: https://t.me/centileochat
 
Administrator  Posts: 933
Oct 7, 2024 18:47
Quote
jiv21 wrote:
I usually use normal maps. I do not usually use relief maps and on rare occasions displacement maps.

As soon as I have some time available I will send you a scene.
Hi Jiv21, I have tested the scene with heavy use of normal maps and get the adequate slowdown of 5% compared to 0.68 because of more correct precise computation of bump map and normal map. So now we need to explore the scene itself that reports this. There are some extra candidate idea but need to check it using your resource.
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 7, 2024 23:27
Hi kirgman.

The scenes are very large because they use recurring resources such as seats, trees and HD animated characters, and the project can reach 60GB.

On the other hand, it has an integrated tool for the creation of stadiums that I have been developing for a year and a half. So in order to send it to you, I would have to remove the whole part of the tool. Then I would remove all the characters to check if the loss of performance is still occurring and thus do not create a project so large to send it.

I beg your patience because right now I am closing a project and the number of hours I dedicate daily does not leave me time for anything else.

What I can tell you is that I make an intensive use of cloners in all its modalities. Instance, render instance and multi instance. In case you have any idea.
 
User  Posts: 9
Oct 8, 2024 00:49
Hi Kirgman.

I have already adapted the scene by removing my tool and eliminating everything that can reduce the size of the scene but it doesn't eliminate the problem.

I have removed all unused materials, alembics, Xpresso, user data, etc.

Still the loss of performance is still occurring. The time is 1.25 in Centile 0.680 and 2.26 in Centile 0.697.

Although the images show a different time (1.45/2.48) the difference is still 1 minute.

As soon as you provide me with a contact email I will send you a link to download the scene.

I am using cinema4D 2024.5. If you use version 2025.0.1 you will get an error because one of the components of version 2024.5 is no longer in 2025.0.1. You must delete it or else it will not let you save any changes you make in the scene.
0.697.png (2.11 Mb)
0.680.png (2.16 Mb)
Edited: jiv21 - Oct 8, 2024 00:55
 
Administrator  Posts: 933
Oct 8, 2024 16:21
I have all Cinema versions including those you have mention.
The good email is info@centileo.com I will benchmark it. It's good the difference is 2x in this case
CentiLeo Chat: https://t.me/centileochat
 
User  Posts: 9
Oct 8, 2024 17:55
Mail sent. Good luck with the scene and I hope it doesn't take too much of your time to find the problem.

On another hand, I would appreciate it if you do not share or make the scene public.

Apologies for the wrong file name. :cry:
Edited: jiv21 - Oct 8, 2024 17:57
 
Administrator  Posts: 933
Oct 9, 2024 18:45
jiv21, thanks for scene! I see that in this scene the new version is slower than in older version. Already investigating for the reasons! Bug testing is very important!
I guarantee you that scene will not be shared for public.
CentiLeo Chat: https://t.me/centileochat
 
Administrator  Posts: 933
Oct 13, 2024 11:23
jiv21, hi! I have fixed the slowdown, now the new version is even slightly faster on your scene.

The issue was in the fact that you configured the stadium scenes with many collections of Mograph multi-instances. In 0.68 they were built as a single union collection. In earlier 0.69 they were built separately because of new shader flexibility. Now the flexibility and performance considerations are combined efficiently.
Thanks to your bugreport the renderer has become even better! Thank you very much!

Btw, this is Juventus stadium. But I want to say Forza Inter! 8) I am fan of Inter for 26 years already.
CentiLeo Chat: https://t.me/centileochat
 
Quote
Kirgman wrote:
jiv21 , hi! I have fixed the slowdown, now the new version is even slightly faster on your scene.

The issue was in the fact that you configured the stadium scenes with many collections of Mograph multi-instances. In 0.68 they were built as a single union collection. In earlier 0.69 they were built separately because of new shader flexibility. Now the flexibility and performance considerations are combined efficiently.
Thanks to your bugreport the renderer has become even better! Thank you very much!

Btw, this is Juventus stadium. But I want to say Forza Inter! I am fan of Inter for 26 years already.
What good news.

Thanks for the explanation. I'm not a programmer, but I remember commenting that maybe that was the problem.

I work with several 3d software packages and render engines and in the end one encounters many issues that tends to narrow down where the problem could be ;)

You are right. It is the juventus stadium “Allianz Stadium”.

I hope that sooner rather than later juventus will come back to delight us with the successes of the past.

Thanks again.
 
Administrator  Posts: 933
Oct 13, 2024 17:42
It was very interesting to investigate for the issue in the engine.

And I also hope the old good competition between Juve and Inter will rise again! Competition is always more interesting than one beating another one with a huge advantage.
CentiLeo Chat: https://t.me/centileochat
Pages: 1
Users browsing this topic