Latency accumulates in small purchases: the converter, the buffer, the processing, the output device, and for visuals, the render and the display. Each stage is a few milliseconds, and the sum decides whether a system feels connected to the music or trailing it.
Where the milliseconds come from
A round trip through a computer spends its budget in predictable places. Analog-to-digital conversion costs a fraction of a millisecond. The audio buffer is the big spender: at a 48 kHz sample rate, a 256-sample buffer holds about 5.3 ms, and systems usually buffer on both input and output. Plugins add processing and lookahead of their own. Bluetooth output adds 100 to 300 ms, which is why wireless headphones are unusable for tracking and why a visuals demo over a Bluetooth speaker feels mysteriously drunk. Displays add a frame or two on top.
How much is too much
Musicians tracking themselves feel discomfort somewhere past 10 ms of round trip; drummers complain first. Audiences watching visuals are more forgiving, but the audiovisual binding window is finite, and reactions arriving tens of milliseconds late read as soggy even when nobody can articulate why. A working rule: under 20 ms feels attached, under 50 ms feels responsive, past 100 ms feels like watching a cover band.
Reducing it
Smaller buffers, a real audio interface instead of the built-in device, wired output, and closing what you are not using. Past that, the craft is spending the budget where it buys the most: tight buffers for detection, prediction (like beat tracking) to hide what cannot be eliminated. Photism analyzes audio inside Live’s own device chain precisely to avoid the routing cost a virtual audio cable would charge.