The session is booked for 2pm. Drummer’s already set up, the bass player is tuning, and you’ve got eight inputs armed. You open your interface control panel, look at the buffer setting, and think: 128 or 256? Pick wrong and either the drummer’s headphones are a quarter-second behind his hands, or your CPU starts spitting dropouts on the first take. Neither is a conversation you want to have twice.
Here’s how I set buffers across a typical day in the studio. No universal numbers, just a working method and the reasoning behind it.
What buffer size actually does
The buffer is a chunk of samples your audio interface processes at once. Small buffer, low latency, high CPU load. Big buffer, more latency, CPU has breathing room. That trade-off is the entire calculation. The trick is that the correct number changes depending on whether you’re tracking, editing, or mixing, and on how many plugins are sitting on your mix bus.
Latency in milliseconds is roughly (buffer size ÷ sample rate) × 1000. At 44.1kHz, a 128-sample buffer is about 3ms per direction. At 256 it’s roughly 6ms. Add interface round-trip conversion and driver overhead and you’re looking at more than the raw math suggests. I don’t chase the theoretical number. I listen to the player and watch the CPU meter.
Tracking: small buffers, and don’t argue with the drummer
For live tracking I run 64 or 128 samples at 44.1 or 48kHz. Drummers notice latency before anyone else in the room. Once you’re past about 10ms of total round-trip, they start pulling their hits and the groove gets stiff. Guitarists with amp sims are the next pickiest. A DI through a plugin chain at 256 feels like playing through a wet sponge.
The catch is CPU. Small buffers mean less time per block, and every plugin on an armed track eats into that. If your drive is a VST3 plugin chain with saturation, an amp sim, and a reverb on the foldback, 64 samples might crackle. Drop to 128 and free up the CPU by printing the effects or using direct monitoring. Most interfaces let you monitor the dry input straight through the hardware, which takes the DAW out of the round-trip entirely. Use it during tracking. It’s the cheapest latency fix you’ll ever get.
Disable heavy plugins on tracks that aren’t armed. A convolution reverb sitting idle on a vocal bus still gets processed unless you freeze or bypass it. I keep one freeze-and-unfreeze ritual before every tracking session now. It’s not glamorous. It saves takes.
Mixing: big buffers, and yes, size matters here
Once tracking is done, crank the buffer to 512 or 1024. You are not playing to a click anymore. The extra latency does nothing while you’re nudging clips and automating a compressor. What you get in return is CPU headroom that lets the mix bus actually do its job.
That matters more than people admit. A mastering-grade chain locked to your mix bus (a gentle bus compressor gluing things together at 1-2dB of gain reduction, an EQ making narrow cuts at Q3 or higher, maybe a saturation stage for weight) is heavy. Add a limiter and a de-esser on the top end and you’re processing the whole stereo file continuously. At 128 samples, that chain can trip a dropout on a busy pass. At 1024 it doesn’t even breathe hard.
I keep the mix bus loaded while I’m mixing, not just while I’m mastering. It’s the only way to hear how the compression behaves on the full arrangement. If the CPU can’t hold it at a small buffer, that’s your signal that 1024 is the right number for this session, not a problem to fix.
The cheat sheet, in one place
- Tracking, drummer or guitarist in the room: 64 to 128 samples. Direct monitoring on. Freeze anything not armed.
- Tracking, keys or vocals with effects: 128. If crackles, print the effects to the track.
- Overdubs, editing, comping: 256. You’ll never feel it.
- Mixing with a loaded bus compressor and mix-bus chain: 512 to 1024. Let the plugins stretch their legs.
- Final bounce / stem export: 1024, and turn off any monitoring plugins you don’t need. Fewer moving parts, fewer surprises.
Notice there’s no “best” number. The cheat sheet is really a map of when to change it. That’s the actual skill.
When to break the rules
One exception worth knowing: if you’re using a plugin that relies on lookahead or convolution, it may already impose its own latency the buffer can’t fix. Slapping a big-buffer chain on while tracking won’t make a player feel better. It’ll just add another 20ms or so on top. So keep heavy plugins off while recording and save them for the mix bus pass. This is exactly why the mix bus is a mixing tool, not a tracking one.
Second exception: high sample rates. At 96kHz, 256 samples gives you roughly the same latency as 128 at 48kHz, but the CPU cost per sample is much higher. If you’re tracking at 96, you might accept 256 as your “low latency” number and adjust expectations accordingly. Don’t fight it. Just know why the meters look angrier.
Getting the habit in place
The biggest improvement to my sessions in the last year wasn’t a new plugin. It was a two-second buffer change at the top of every workflow phase. Track small, mix big. Say it out loud when you switch. After a week it’s automatic.
If a session is failing, ask the boring question first: is your buffer size wrong for what you’re actually doing, or is something else broken? A driver that’s genuinely misbehaving and a buffer that’s too small can both cause dropouts, and the fix for one won’t save you from the other. Read the interface control panel before you blame the plugin. It’s the least exciting part of the job, and it’s frequently the answer.