Compute and merge node behaviour in animation

Started by Kadri, February 10, 2019, 02:20:50 am

Previous topic - Next topic

Kadri

February 10, 2019, 02:20:50 am Last Edit: February 10, 2019, 06:28:24 am by Kadri

Say we have an animated node setup like the one below where the merge node combines the groups together in time.
Is the effect of the compute node restricted too in time where the merge node restricts-acts on the compute node?
Will we see gradually slower render times from group 1 to 2 and then gradually faster from 2 to 3 ?
If yes we should not see slower render times in the last group for example?
Or is it more complicated? Like the compute node slowing down all first 3 groups the same way for example?


Dune

Interesting headbreaker. I wouldn't know, actually. I'd think they all take around equal time, as the compute is part of the whole scene; computed every frame, but the computed area not always used. Why not test it with a simple setup and render just one frame from each group and compare?
Easiest would be if you don't need the compute, but I guess you're having laterals.

Kadri


Yes testing might be good Ulco. But i have 2 computers full rendering and i am even using Pixelplow :)


Kadri

February 10, 2019, 08:58:43 am #3 Last Edit: February 10, 2019, 12:40:34 pm by Kadri

Ok i got curious until the answer and sacrificed half a second of video time :)

I used a template that i found from the web (look at the attachment below).
Not quite fun to count and put those numbers manually. But there were only 40.

I made the same animation example above as a real animated project
and rendered every 10'th frame and counted how much second every one of them took to render.

There is no incremental increase between groups and makes kinda sense if this test is anywhere representative of what is going on really.
There are some fluctuations ( i used the computer for web browsing at the same time).
But three of them are where the merge nodes are overlapping in time so it might be related.

Interestingly the first group before the compute node is much less affected by the merge then the group after the compute node.
I had expected that it should be the same for the group before the compute node (hopefully not an error in the project on my side).
If this is really related to the compute node, in some situations there could be a different look then you normally expect maybe.

The first powerfractal without any other group renders in 68 seconds by the way.
Actually when you disable the other 3 groups it is 68 second. But when you delete those groups it is 67 seconds.
I tried this 3-4 times and it was every time the same.

I am testing this right now on a longer render. But not 3-4 times. Only 1 time. I am curious but not so much :)


Kadri

February 10, 2019, 09:26:27 am #4 Last Edit: February 10, 2019, 12:36:28 pm by Kadri

Render with 3 disabled groups (see the attachment) took 28 minutes.
Render with those 3 groups deleted took 27 minutes.

That 1 minute looked a little more then i expected so i am rendering ones more the first test.
For a still this is nothing i would think about. But for an animation it could add quite much.

The problem here could be that disabling and deleting nodes don't give the same look every time.
Not sure if the same situation applies here too.

Kadri

February 10, 2019, 11:01:25 am #5 Last Edit: February 10, 2019, 12:37:39 pm by Kadri

Turns out i was curious enough to test it 3 times  ::)

There were different results-variations in render times.
But the difference with disabling and deleting the 3 groups is around 2 minutes in average.
Around 27 minutes deleted and 29 minutes disabled.