Looking for a cause of Interactive Rendering slowdown (+Poll)

[steyin]

Steyin, the difference between parsing times for IR and final renders should definitely not be so big - there is certainly some problem (maybe on our side). If you could provide us the scene so we could have a look at it, it would help us a lot with finding the potential issues.

I uploaded the scene in question. Initial parsing time was around 30 seconds, seems it is due to the Rhino geometry, but switching between viewports IR was ok. Every now and again it slows down a little though. Still, hitting render instead of IR parsing was 4-5 seconds.

[lupaz]

Just finished my first test between Corona 1.5 Hotfix 2 and Corona 4 HF1.

The difference is HUGE!
Edit: Huge difference meaning Corona 1.5's IR is WAY better.

My process was:

In a windows 10 workstation, with 3ds max 2016,

1-I uninstalled Corona 4
2- installed Corona 1.5
3- I opened an old-ish scene that I knew didn't have the latest developments.
4- Made it simple enough to not run our of memmory or to make IR decently fluid.
5- Tested IR. Works nice.
6- Installed Corona 4
7- Opened the same scene
7.5- Dismissed the warnings about the new displacement and refraction (I think it was).
8- Set IR subsampling to 0
9- Started IR
10- cried when I saw the difference.

One thing to note is that with Corona 1.5 I get the "Fault Tolerant Heap" warning. With Corona 4 I didn't.

[TomG]

Can you send us the scene via private uploader?

[lupaz]

Here's a video:

I can send you the scene tomorrow, but honestly I believe the issue is scene independent.

[selene]

Here's a video:

I can send you the scene tomorrow, but honestly I believe the issue is scene independent.
...

Hello lupaz, thank you very much for your feedback and for dedicating your time and energy into recording the video - the slowdown is indeed visible in this case. Since you mentioned that you get "Fault tolerant heap" warning, it is highly probable that this could be the cause of the slowdown. "Fault tolerant heap" is a feature of Microsoft Windows OS, which can get automatically turned on if OS detected multiple crashes of some specific application (in this case 3DS Max). This feature may have severe performance impact on rendering speed. It is not possible to turn off this feature from Corona Renderer, we can only warn our users that it is turned on (as you can see in 0:28), but you can turn it off it manually, as described on our Freshdesk:

https://coronarenderer.freshdesk.com/support/solutions/articles/5000678431-windows-enabled-fault-tolerant-heap-for-3dsmax-or-dependent-processes-this-may-have-severe-perform

I recommend you to turn it off and try the Interactive Rendering again, hopefully it will be much better now. Please let us know if turning off this feature helps.

Anyway, if you could send us the scene you used for this comparison, it would really help us a lot, since we don't have enough "real-world" examples (scenes) where we could reproduce these kind of issues. :)

[selene]

...
You can bet that I will do that. :) I would also give you the scene.
Until now I could only notice that IR sometimes crashes, if I made a lot of changes in the scene very quickly - like every few seconds a change.

Regarding parsing time, I noticed that some scenes are simply parsing quicker than others.
Tried erasing parts of the scene... until it remained literally almost empty. Still the same long parsing...

In other scenes that are much bigger, parsing is much quicker...
Strange phenomenon that I had noticed also in previous Corona versions.

I´ll continue testing.

Designerman77, if you had some empty scene which parses for a long time, please could you send it to us? This indeed should not happen and we would like to investigate what could be the cause of this issue.

[selene]

Forestpack and railclone (itoo software plugin for 3dsmax) slow down deeply the iR.

Hello Alexandre,  thank you very much for your feedback. We are aware of this problem and currently we are still investigating whether it is some fault on our side or if the slowdowns are caused by aforementioned plug-ins. Also if you could provide us some scene which manifests this issue, it would help us a lot. :)

[selene]

...

I uploaded the scene in question. Initial parsing time was around 30 seconds, seems it is due to the Rhino geometry, but switching between viewports IR was ok. Every now and again it slows down a little though. Still, hitting render instead of IR parsing was 4-5 seconds.

Steyin, thank you very much for the scene. We are going to have a look at it. :)

[lupaz]

Anyway, if you could send us the scene you used for this comparison, it would really help us a lot, since we don't have enough "real-world" examples (scenes) where we could reproduce these kind of issues. :)

I'm uploading the file to dropbox. The file name is 044.zip.

[lupaz]

I recommend you to turn it off and try the Interactive Rendering again, hopefully it will be much better now. Please let us know if turning off this feature helps.

I did.
There was no difference.
IR behaves like on the video.

[rowmanns]

I recommend you to turn it off and try the Interactive Rendering again, hopefully it will be much better now. Please let us know if turning off this feature helps.

I did.
There was no difference.
IR behaves like on the video.

Hi,

Did you also try with subsampling enabled and disabled to see if this has any effect?

Cheers,

Rowan

[lupaz]

I recommend you to turn it off and try the Interactive Rendering again, hopefully it will be much better now. Please let us know if turning off this feature helps.

I did.
There was no difference.
IR behaves like on the video.

Hi,

Did you also try with subsampling enabled and disabled to see if this has any effect?

Cheers,

Rowan

I didn't try that. Because with those big pixels it's very hard to compare.
I dislike that style of IR.

[lupaz]

Anyway, if you could send us the scene you used for this comparison, it would really help us a lot, since we don't have enough "real-world" examples (scenes) where we could reproduce these kind of issues. :)

I'm uploading the file to dropbox. The file name is 044.zip.

Hi.
Did you guys receive the scene? I was wondering if you got the same outcome.

[maru]

Here's a video:

I can send you the scene tomorrow, but honestly I believe the issue is scene independent.

I am not saying that this is not true, and actually in my simple tests I had the same impression, that in older versions IR was (or at least felt) faster, but here is what actually happens in the above video:

Older version: image is rendered.....camera is moved.....the previous rendering remains in the background and pixels from the new rendering start appearing immediately on top of it.....after some time the 1st pass is finished and next passes continue
Newer version: image is rendered.....camera is moved.....the previous rendering remains in the background, but pixels from the new rendering do not start appearing.....after some time the 1st pass is finished and is then placed on top of the previous rendering

So the way it works is simply different in 1.5 and newer versions, and it is hard to compare render/refresh times. Basically, in 1.5 IR creates this fade-in effect, which is not there in newer versions.
This "new image interactively blending on top of the old one" effect definitely adds to the feeling of better speed and responsiveness, and I described it in my internal report id=371632450

[lupaz]

Yes! Please!:

- Bring back "new image interactively blending on top of the old one"
- Remove the, in my opinion, unnecessary IR subsampling (at least by default)