Stack Composite Mode: Over vs TopMost

We've noticed a difference in load time when using OVER and TOPMOST composite mode. There is also a weird..bug..maybe...when using TOPMOST.

I have python creating multiple stacks in a session where in each stack, there is the plate and the latest version of the render. 

When OVER is set in composite mode the loading times are pretty slow. It is clocked at about 4-6 times slower then when setting the composite mode to TOPMOST.  So I modified our scripts to use TOPMOST. 

There is a side effect to this. Using topmost is super fast, but it will only cache the topmost layer in the stack. It doesn't cache the plate. Once we switch to the plate using Shift+), RV will hang for about 2-4 seconds, then begin caching the plate. Using OVER will cache both the plate and the render at the same time but is super slow. 

The artists and sups have found this to be annoying :D

Is there a way to keep the extremely fast caching with TOPMOST and have the loading features of OVER? (have RV cache both/all souces while in TOPMOST mode)



1 comment

  • Avatar
    (Michael) Kessler Official comment

    Hi Nick,

    Thanks for posting!

    Unfortunately, the speedup you are seeing is specifically because the topmost mode doesn't evaluate the other paths in the graph.

    I think my best suggestion is to optimize your reading performance which may include adjusting your reader threads.  As a general rule, I start with setting the reader threads to the number of your physical cores minus one.  For instance, if you have a machine that says it has 12 cores but has hyperthreading, then your machine likely has 6 physical cores, therefore I'd start with the reader threads set to 5.  More details on optimizing playback can be found here:

    Also, try out Look Ahead caching and Region caching; they have their different place depending on what sort of playback you are doing, and one may work better for your playback needs over the other.

Please sign in to leave a comment.