Sample Rate and Latency Chart: Buffer Size to Milliseconds
One way latency in milliseconds is buffer size divided by sample rate, times 1000. A 128 sample buffer at 44.1 kHz gives 2.90 ms, and at 48 kHz gives 2.67 ms. A 256 sample buffer at 48 kHz gives 5.33 ms. Round trip is roughly double that plus 2 to 6 ms of converter and driver overhead, so a 128 sample buffer at 48 kHz lands near 9 to 11 ms in practice. Under 5 ms round trip is unnoticeable for most players, 10 ms is where singers start to feel it, and 20 ms or more is unplayable for percussive instruments.
Definition: Latency is the delay between a signal entering a recording system and leaving it again, and its dominant cause is the audio buffer: a block of samples the computer must fill before it can process anything.
Latency is arithmetic, not luck. The formula is ms = buffer samples / sample rate x 1000, and everything else in the discussion is either a second application of that formula or a fixed overhead added on top. This chart gives you every combination you are likely to set, calculated to two decimal places, plus the perceptual thresholds that decide whether the number is a problem.
What is the one way latency for every buffer and sample rate?
Find your buffer size in the left column and your sample rate across the top. This is the time the system spends filling one buffer, which is the irreducible core of the delay. It happens once on the way in and once on the way out.
| Buffer | 44.1 kHz | 48 kHz | 88.2 kHz | 96 kHz | 176.4 kHz | 192 kHz |
|---|---|---|---|---|---|---|
| 16 | 0.36 | 0.33 | 0.18 | 0.17 | 0.09 | 0.08 |
| 32 | 0.73 | 0.67 | 0.36 | 0.33 | 0.18 | 0.17 |
| 64 | 1.45 | 1.33 | 0.73 | 0.67 | 0.36 | 0.33 |
| 128 | 2.90 | 2.67 | 1.45 | 1.33 | 0.73 | 0.67 |
| 256 | 5.80 | 5.33 | 2.90 | 2.67 | 1.45 | 1.33 |
| 512 | 11.61 | 10.67 | 5.80 | 5.33 | 2.90 | 2.67 |
| 1024 | 23.22 | 21.33 | 11.61 | 10.67 | 5.80 | 5.33 |
| 2048 | 46.44 | 42.67 | 23.22 | 21.33 | 11.61 | 10.67 |
Two patterns fall out of the table immediately. Doubling the buffer doubles the latency exactly. Doubling the sample rate halves it exactly. That means a 256 sample buffer at 96 kHz produces the identical 2.67 ms as a 128 sample buffer at 48 kHz, but gives the CPU twice as many samples per interrupt to work with. That trade is the whole reason people track at high sample rates, and it is a real engineering argument rather than an audio quality one.
What is the round trip latency you actually feel?
The buffer figure is one way. A monitored signal passes through the buffer twice, once entering and once leaving, so start by doubling. Then add the fixed overhead: analogue to digital conversion, the driver layer, and digital to analogue conversion on the way out. That overhead is 2 to 6 ms on typical class compliant hardware and is lower on Thunderbolt and higher on cheap USB interfaces.
| Buffer | 44.1 kHz round trip | 48 kHz round trip | 96 kHz round trip | What it feels like |
|---|---|---|---|---|
| 16 | 4.73 | 4.67 | 4.33 | Indistinguishable from direct monitoring. |
| 32 | 5.45 | 5.33 | 4.67 | Fine for everything, if your CPU can hold it. |
| 64 | 6.90 | 6.67 | 5.33 | Comfortable for drums and percussion. |
| 128 | 9.80 | 9.33 | 6.67 | The standard tracking setting. Singers may just notice it. |
| 256 | 15.61 | 14.67 | 9.33 | Guitar and keys are fine. Vocals feel detached. |
| 512 | 27.22 | 25.33 | 14.67 | Mixing only. Unplayable for anything rhythmic. |
| 1024 | 50.44 | 46.67 | 25.33 | Mixing with a heavy plugin load. Do not track here. |
| 2048 | 96.88 | 89.33 | 46.67 | Rendering and final mix passes only. |
At what point do players actually notice latency?
The thresholds below are the ones that matter, and they vary by instrument far more than most latency discussions admit. A guitarist and a drummer tolerate wildly different numbers, and a singer is a separate case again because of a physical mechanism the others do not share.
| Round trip | Distance equivalent | Who notices | Effect on the performance |
|---|---|---|---|
| Under 5 ms | Under 5.7 ft | Nobody | Feels like playing an acoustic instrument in a room. No adaptation needed. |
| 5 to 10 ms | 5.7 to 11.3 ft | Drummers, some singers | Playable for everything. Percussionists may push slightly ahead of the beat. |
| 10 to 15 ms | 11.3 to 17 ft | Singers, drummers, fast players | Singers hear a disturbing doubling of their own voice. Timing drifts late. |
| 15 to 25 ms | 17 to 28 ft | Almost everyone | Sustained parts still work. Anything with attack falls apart. |
| 25 to 40 ms | 28 to 45 ft | Everyone | Reads as a slapback echo rather than as delay. Unusable for tracking. |
| Over 40 ms | Over 45 ft | Everyone | Two distinct events. The performer stops and asks what is wrong. |
The distance column is the most useful way I know to explain latency to a player who is not interested in buffers. Sound travels at 1130 ft per second, so one millisecond is about 1.13 ft. Ten milliseconds of round trip latency is therefore acoustically identical to standing 11 ft back from your amp, which every guitarist has done without noticing. Forty milliseconds is standing 45 ft away, which nobody has ever done and enjoyed.
Singers are the case that surprised me. I had a session where the guitarist was perfectly happy at 256 samples and the vocalist could not get a take, and I spent twenty minutes checking the headphone mix before I understood why. A singer hears their own voice twice: once through bone conduction, arriving instantly, and once through the headphones, arriving late. At 15 ms round trip those two signals are far enough apart to read as a flanged double and close enough to be confusing rather than obviously wrong. The singer's response is always to push harder, which changes the tone, which makes the take worse. Dropping to 64 samples fixed it in one pass. Now I set the buffer to what the vocalist needs, not what the machine prefers.
Which sample rate should you actually record at?
For most home recording, 48 kHz is the right default, and the reasons are practical rather than sonic. It is the standard for video, so anything you write to picture matches without conversion. Its buffer arithmetic is slightly friendlier than 44.1. And the Nyquist limit at 48 kHz is 24 kHz, comfortably above the top of human hearing, so there is no content being lost.
| Rate | Nyquist limit | 24-bit stereo, per minute | Use it for |
|---|---|---|---|
| 44.1 kHz | 22.05 kHz | 15.1 MB | CD delivery and anything destined only for streaming. |
| 48 kHz | 24 kHz | 16.5 MB | The sensible default. Video standard, and universally supported. |
| 88.2 kHz | 44.1 kHz | 30.3 MB | Halving buffer latency when the destination is 44.1. |
| 96 kHz | 48 kHz | 33.0 MB | Low latency tracking, and heavy pitch or time manipulation. |
| 176.4 kHz | 88.2 kHz | 60.6 MB | Specialist work. Halves your plugin count for no audible gain. |
| 192 kHz | 96 kHz | 66.0 MB | Sound design where extreme pitch shifting is planned. |
The honest case for 96 kHz is latency and processing headroom, not fidelity. If you track at 96 kHz with a 256 sample buffer you get 2.67 ms one way, the same as 48 kHz at 128 samples, but the CPU gets a block twice as large to work on, which usually means fewer dropouts. The cost is double the file size and roughly double the plugin load. Above 96 kHz the returns disappear and the costs do not, which is why 176.4 and 192 kHz stay in the table as reference rather than as a recommendation.
What else adds latency besides the buffer?
Plenty, and some of it is invisible in the DAW's reported figure. This is why the number you feel rarely matches the number on screen.
- Conversion. Analogue to digital and digital to analogue each take time, and the sigma-delta converters in modern interfaces use digital filters that add roughly 0.5 to 1.5 ms per direction. Some units offer a low latency filter mode that trades a little filter steepness for less delay.
- The driver. A native ASIO or Core Audio driver adds very little. A generic driver, a class compliant fallback, or anything running through an OS mixer adds a lot. Interfaces from RME are bought specifically for their driver behaviour rather than their converters.
- Plugin latency. Any plugin with lookahead reports its delay to the host, which then compensates by delaying everything else. Linear phase EQs, lookahead limiters and most mastering compressors are the usual culprits, and a single one on the monitoring path can add 20 ms or more.
- Wireless anything. Bluetooth headphones add 100 to 300 ms. There is no setting that fixes this and no reason to track through them.
- The USB path. A hub, a long cable or a shared bus with a drive adds jitter and can force the driver to use a larger safety buffer than you selected. The cable chart covers the run length limits.
How do you get low latency without dropouts?
Dropouts happen when the CPU cannot fill a buffer before the interface needs it. Smaller buffers give the CPU less time, so low latency and heavy processing pull against each other. The way out is to stop treating the buffer as a single setting for the whole project.
- Track at 128 samples, mix at 512 or higher. These are different jobs with different requirements. Nobody needs low latency while mixing, and nobody needs sixty plugins while tracking.
- Use direct monitoring while tracking. Almost every interface routes the input to the headphone output in hardware, which sidesteps the computer entirely. Round trip becomes effectively zero regardless of buffer size.
- Freeze or bounce heavy tracks before overdubbing. A frozen track costs almost nothing to play back, which frees the CPU for a small buffer.
- Bypass lookahead plugins on the monitor path. A linear phase EQ on the master bus while tracking is a self-inflicted 20 ms.
- Give the interface its own USB port. Not a hub, not shared with a bus powered drive, and using the cable that came in the box.
If none of that gets you there, the limit is the hardware. Interfaces vary widely in round trip performance at the same buffer size, and the difference comes from the driver rather than the converters. A MOTU M4 and a Focusrite Scarlett 2i2 (4th Gen) at the same 128 sample setting can land several milliseconds apart. The latency and buffer calculator lets you plug in your own measured overhead, and the latency troubleshooting guide walks the full diagnostic order.
The setting I actually keep is 128 samples at 48 kHz for tracking and 1024 for mixing, and I change it with a keyboard shortcut rather than a menu because the friction of the menu meant I used to just leave it at whichever setting I last needed. The other habit worth stealing: before a vocal session, I record five seconds of the singer talking, play it back, and ask whether the headphone mix feels immediate. That is a better test than any number, because the only latency figure that matters is the one the performer can feel.
Does buffer size affect sound quality?
No. This needs stating plainly because it comes up constantly. The buffer is a container, not a process. A sample recorded with a 32 sample buffer is bit for bit identical to the same sample recorded with a 2048 sample buffer. What changes is only when it arrives.
What can change sound quality is a buffer set too small for the machine, because the resulting dropouts are real audio faults: clicks, gaps and glitches printed into the file. That is not a subtle degradation, it is an obvious defect, and it is the only quality argument the buffer setting has any part in. If your recordings have clicks, raise the buffer and rely on direct monitoring for feel.
The same applies to headphones. A pair of closed back tracking headphones does not change the latency, but good isolation lets you run the headphone mix quieter, which reduces bleed into the microphone and gives the performer a cleaner reference to play against. Latency and monitoring comfort are separate problems that get confused with each other constantly.
Related tools and charts
- Latency and buffer calculator: enter your own buffer, rate and overhead
- How to fix latency in your DAW: the diagnostic order that finds it fastest
- Cable and connector chart: run length limits for USB and Thunderbolt
- Best audio interface for a home studio: driver quality is the thing to buy for
- Intermediate home studio setup: a build sized for low latency tracking
Frequently asked questions
How do you calculate latency from buffer size?
Divide the buffer size in samples by the sample rate in hertz, then multiply by 1000 to get milliseconds. A 128 sample buffer at 44,100 Hz gives 128 divided by 44100, multiplied by 1000, which is 2.90 ms. The same buffer at 48,000 Hz gives 2.67 ms. That figure is one way only. Round trip latency, which is what you actually feel, is roughly double it plus another 2 to 6 ms of converter and driver overhead.
What buffer size should I use for recording?
Use 128 samples for tracking anything performed in time, which gives 2.90 ms one way at 44.1 kHz and lands around 10 ms round trip on typical hardware. Drop to 64 if your machine handles it and you are recording percussive parts. Raise the buffer to 512 or 1024 once you are mixing, because latency no longer matters and the extra headroom lets you run more plugins without dropouts. Changing the buffer between the two stages is normal practice, not a workaround.
Is 10 ms of latency noticeable?
It depends entirely on the instrument. Ten milliseconds is roughly the delay of standing eleven feet from a guitar amp, which is unremarkable for a guitarist and genuinely disruptive for a drummer hitting a pad. Singers notice it too, because bone conduction delivers their own voice instantly while the headphone signal arrives late, and that mismatch reads as an unsettling doubling rather than as delay.
Does a higher sample rate reduce latency?
Yes, and this is the one genuine performance argument for recording at 96 kHz. A 128 sample buffer takes 2.90 ms to fill at 44.1 kHz and only 1.33 ms at 96 kHz, because the samples arrive more than twice as fast. The cost is roughly double the CPU load and double the disk usage. Running 96 kHz with a 256 sample buffer gives you the same 2.67 ms as 48 kHz at 128 samples, with more processing headroom.
Why does my DAW report less latency than I feel?
Because the reported number usually accounts only for the buffer, and often only in one direction. The real path adds analogue to digital conversion, the driver layer, the buffer on the way in, the buffer on the way out, digital to analogue conversion, and any plugin with lookahead. Those additions total 2 to 6 ms on typical hardware and considerably more if you have a linear phase EQ or a lookahead limiter on the monitoring path.
What is direct monitoring and should I use it?
Direct monitoring routes the input signal straight to your headphones inside the interface, bypassing the computer entirely, so the round trip latency for what you hear is effectively zero. Use it whenever you are tracking without needing a plugin on the input. The tradeoff is that you hear the raw signal, so a vocalist who performs better with reverb needs either an interface that adds reverb in hardware or a low enough buffer to monitor through the DAW.
Working out your own room and signal chain? The Home Studio Build Planner is the paid version of these pages: 8 printable worksheets you fill in with your own numbers, plus the full PDF, $29.