Technical
Buffer Size to Latency
Your buffer size sets the trade-off between latency and CPU load. Pick a buffer and sample rate to see the delay it introduces, so you can record tight and mix without crackles.
How the number is worked out
latency (ms) = (buffer size ÷ sample rate) × 1000
A 256-sample buffer at 48 kHz holds 256 ÷ 48,000 = 0.00533 seconds, or 5.33 ms of audio. That's the one-way delay for the buffer to fill. Round-trip (input through the buffer and back out) is roughly double, plus a little converter overhead, so about 10–12 ms here. Higher sample rates shorten the delay for a given buffer, because each sample represents less time.
Small buffer to track, big buffer to mix
The rule of thumb: small buffer for recording, large buffer for mixing. When you're tracking or playing a soft synth live, you want round-trip latency under about 10 ms so it feels immediate, which means a 64 or 128 buffer, at the cost of higher CPU load and possible crackles on a heavy project. When you're mixing and not playing anything in real time, crank the buffer to 512, 1024, or higher: the latency doesn't matter and your CPU gets the headroom to run more plugins.
Anything above ~20–25 ms of round-trip latency starts to feel laggy when playing, throwing off your timing. If you can't get low enough without crackling, freeze or bounce heavy tracks to free up CPU rather than fighting the buffer.
Your interface's driver matters more than the buffer number. On Windows, an ASIO driver will get you far lower stable latency than the generic one; on Mac, Core Audio handles this natively. A cheap interface on a bad driver can be laggier at a 64 buffer than a good one at 256.