home/ atoms/ td-cook-on-demand-efficiency

Turning off cooking for invisible TOPs eliminates wasted GPU time in multi-layer VJ rigs

In TouchDesigner every active TOP consumes GPU resources each frame (‘cooks’) whether or not its output is visible. A performance rig that loads many scenes simultaneously wastes GPU on scenes not currently on output. The discipline of ensuring inactive TOP branches do not cook — either through Switch TOPs, Display flags, or explicit cook flag management — is the primary efficiency lever in a large TD rig. Simple Mixer follows this rule: whenever TOPs are not visible, they are all turned off. Avoiding VRAM flush operations (which cause stutters) makes cook-on-demand the preferred strategy over flush-and-reload.

Examples

A rig with 20 scenes: at any moment, 1 scene is on air. 19 scenes have cooking disabled. GPU cost is approximately 1 scene worth of processing rather than 20.

Assessment

A TD project shows dropped frames during scene transitions. Identify two causes related to cooking policy and propose a fix for each without using VRAM flush operations.

“whenever tops are not visible, they are all turned off. Um and then there's no VRAM flushing because VRAM flushing will it it like it might cause a stutter. I don't want that either. I want smooth and steady”
corpus · touchdesigner-meetup-vj-tools-and-live-performance-rigs-free · chunk 6