Scaling TouchDesigner for large multi-machine shows

One laptop running a VJ patch tops out fast: a festival main stage, a museum atrium, or an arena kinetic-light rig demands tens of thousands of DMX channels, multiple projectors on an as-built set, and render loads no single machine survives. This module is where TouchDesigner stops being a patch and becomes a show system — the same territory as disguise or Resolume Arena, but fully custom.

The arc starts supported and narrow. First, get two machines talking: the master-client media-server pattern (one TD master sending cues over the LAN to clients playing local media) is your first exercise, including a basic failover swap. Then decouple authoring from playback using the JSON-defined playback engine pattern, so zones can be reassigned across engines for load balancing without touching patches. In parallel, scale output: the GLSL-computed 40k-channel DMX technique moves channel math onto the GPU, and projection is locked in either by hand-drawing a mapping mesh on a camera view or via OpenCV camera/projector auto-calibration with structured light. Finally, feed the beast: pull a Houdini FBX/Alembic scene or bridge a Python-driven genetic-algorithm simulation into the real-time render.

The capstone assembles all of it unsupported, and every required atom gates a piece of it: no network control and JSON engine, no distributed topology; no GPU DMX path, no 40k channels in real time; no mesh-drawing or OpenCV calibration, no mapped projection on the day; no Houdini or simulation bridge, no data source. Cue-sending across the master/client link is drilled to automaticity inside rehearsal runs, because mid-show is no time to think about OSC routing. The supporting atoms enrich rather than gate — framing how autonomous versus operator-guided your system should be, and how real-time rasterized renderers trade accuracy for interactivity, context that sharpens your previz and architecture choices.