Originally published on Maker.io (DigiKey).
Part 1 got the per-message time to 140 µs through hardware changes. Part 2 applies software techniques that are considerably harder to implement. Switching from blocking printf to snprintf with asynchronous DMA or interrupt-driven transmission frees the processor while bytes go out over the wire. Compiling at -O3 reclaims a few more microseconds. Switching the transport from UART to USB CDC with a transmit buffer delivers the biggest remaining gain: 88 µs per message, with the processor only busy for about 63 µs of that.
The cumulative result across Parts 1 and 2 is approximately 80× faster than the original 7,030 µs baseline. The caveat: each technique here is meaningfully harder to implement than those in Part 1, and faster-for-its-own-sake is rarely justified — the right time to apply these is when you’ve confirmed that printf throughput is the actual bottleneck.