How to Fix Latency in Your DAW: The Order That Actually Solves It
Fix latency in this order: turn on direct monitoring in the interface so the signal never makes the round trip, then drop the buffer to 128 samples for tracking, then freeze or bounce heavy tracks, then take look-ahead plugins off the record-armed channel, then check you are on a native ASIO or Core Audio driver. Use two buffer settings, not one: 128 samples for tracking and 512 to 1024 for mixing.
Definition: Latency is the delay between a sound entering an audio interface and the processed version of it coming back out, and it is set mostly by the buffer size, which is the number of samples the computer collects before it processes anything.
Latency is not one problem, it is four problems that produce the same symptom, and the reason people struggle with it is that they reach for the buffer size first when the buffer size is usually the second or third thing to fix. The order below is deliberate: each step either removes the delay entirely or reduces the load that forces you to accept it. Work down the list and stop when the problem is gone.
How do you turn a buffer size into milliseconds?
A buffer is the block of samples your computer collects before it processes anything. Bigger blocks are more efficient and more forgiving of a busy CPU, and they take longer to fill. The arithmetic is:
milliseconds = buffer samples / sample rate x 1000
So 128 samples at 48 kHz is 2.67 ms, and the same 128 samples at 44.1 kHz is 2.90 ms. That is one direction only. What you feel while monitoring through the computer is the round trip: in and out again, roughly double, plus another 2 to 6 ms of converter and driver overhead that no setting exposes. A 128 sample buffer at 48 kHz therefore lands somewhere around 7.3 to 11.3 ms in practice, which is why the number your DAW reports never matches the number you feel.
| Buffer | 44.1 kHz | 48 kHz | 96 kHz | Use it for |
|---|---|---|---|---|
| 32 | 0.73 ms | 0.67 ms | 0.33 ms | Rarely stable on a full session |
| 64 | 1.45 ms | 1.33 ms | 0.67 ms | Overdubbing on a light session |
| 128 | 2.90 ms | 2.67 ms | 1.33 ms | The tracking default |
| 256 | 5.80 ms | 5.33 ms | 2.67 ms | Tracking on a busier session |
| 512 | 11.61 ms | 10.67 ms | 5.33 ms | Mixing, or MIDI playback only |
| 1024 | 23.22 ms | 21.33 ms | 10.67 ms | Mixing a heavy session |
| 2048 | 46.44 ms | 42.67 ms | 21.33 ms | Final mix with everything running |
A useful physical reference: sound travels about 1.13 ft per millisecond, so 10 ms of round trip delay is the same as standing 11.3 ft from the source. Guitarists do that on stage every night without noticing. Forty milliseconds is the far side of a large room, and nobody can play in time with that. The latency buffer calculator runs these figures for any combination, and the sample rate latency chart is the printable version.
Almost every current interface has hardware direct monitoring, which is the fix that removes the problem instead of shrinking it. On the Focusrite Scarlett 2i2 (4th Gen) it is a front panel button. On the MOTU M2 and most others it is a mix knob that blends input against playback. On the RME Babyface Pro FS it is a full routing matrix. Compare what each unit offers in the interface spec database.
Step 1: Turn on direct monitoring
Direct monitoring sends the input signal to your headphone output inside the interface, before it ever crosses the USB cable. The delay is not reduced, it is removed, because the signal never goes to the computer and back. You hear yourself instantly and you hear the DAW playback mixed in alongside.
The cost is that you do not hear plugins on the input. For a singer who wants reverb in the headphones, that is a real limitation, and there are three ways around it. Some interfaces provide reverb in their own hardware mixer, such as the loopback and effects features in Steinberg's dspMixFx on the Steinberg UR22C studio bundle . Some provide DSP plugins that run on the interface itself, which is the entire argument for the Universal Audio Apollo Twin X DUO USB Heritage Edition . Or you accept a low buffer and monitor through the computer.
One thing to check when you turn direct monitoring on: mute or disable input monitoring in the DAW at the same time. If both are active you hear yourself twice, once instantly and once delayed, and the resulting flam is far more distracting than plain latency ever was. This is the single most common support question about direct monitoring and it has that one cause.
Step 2: Set the buffer to 128 for tracking
If you must monitor through the computer, because you want plugins on the input or because your interface has no direct path, drop the buffer. At 48 kHz, 128 samples is the practical target: low enough that most players cannot feel it, high enough to stay stable on a session that is not enormous.
Do not chase 32 or 64 samples. The gain is a millisecond or two and the cost is crackles and dropouts, which ruin takes far more reliably than 3 ms of delay does. If 128 is not stable, the answer is not to try 64, it is step three.
And use two settings, not one. This is the habit that separates people who fight latency from people who do not think about it. Track at 128. When you stop tracking and start mixing, raise the buffer to 512 or 1024 so every plugin has room to run. Both are one dropdown apart in every DAW, and both take three seconds to change.
The habit I would give anyone is to treat buffer size as part of switching modes, the same way you would change from a tracking headphone mix to a monitor mix. My session template opens at 128 because the first thing I usually do is play something in. The moment I stop recording and start arranging, the buffer goes to 1024 and stays there until I need to record again. I used to run everything at 256 as a compromise and it was the worst of both: slightly annoying to play against and not enough headroom for a heavy mix. There is no single number that serves both jobs, and looking for one is the actual mistake.
Step 3: Freeze or bounce the heavy tracks
The reason you cannot run a low buffer on a large session is that every plugin instance has to complete its work inside the time the buffer represents. At 128 samples and 48 kHz that is 2.67 ms of wall clock time per block for the entire session. Add enough convolution reverbs and orchestral sample instruments and the computer misses the deadline, which is what a dropout is.
Freezing renders a track and its plugins to audio and disables the processing while keeping the ability to unfreeze later. Every major DAW has it under some name: Freeze in Logic and Cubase, Freeze or Commit in Pro Tools, Freeze in Studio One, Bounce in Place in most of them. Virtual instruments are the biggest win because they are usually the heaviest single thing in a session.
A related move that costs nothing: disable, do not just bypass. A bypassed plugin in many hosts still occupies its processing slot and, more importantly, still reports its latency to the delay compensation system. Taking it offline entirely is what actually gives the resources back.
Step 4: Take look-ahead plugins off the input chain
Some plugins cannot work without seeing the future. A look-ahead limiter needs to know a peak is coming before it arrives. A linear phase EQ has to buffer enough samples to perform its filtering without phase shift. A convolution reverb has to gather a block before it can convolve it. Each of these reports a fixed processing delay to the host, and the host compensates by delaying every other track to match, so playback stays in sync.
That compensation is correct and it is also the problem. If you have a linear phase EQ on the master bus reporting 40 ms, then the signal path from your record-armed input, through its own plugins, to your headphones now includes that delay too. The buffer setting is not what is hurting you. A single plugin is.
| Plugin type | Typical added delay | Safe on an input chain? |
|---|---|---|
| Minimum phase EQ | 0 samples | Yes |
| Standard compressor, no look-ahead | 0 samples | Yes |
| Compressor with look-ahead enabled | 1 to 10 ms | Turn look-ahead off while tracking |
| Linear phase EQ | 10 to 90 ms | No, use the minimum phase mode |
| Look-ahead limiter | 1 to 5 ms | Not on the master while tracking |
| Convolution reverb | 5 to 50 ms | Use an algorithmic reverb for cue mixes |
| Pitch correction, graphical mode | 20 to 100 ms | No, use the low latency mode |
| Multiband dynamics | 2 to 40 ms | Check the plugin's own latency readout |
Most DAWs have a low latency monitoring mode that automatically bypasses any plugin reporting above a threshold while a track is record-armed. Logic calls it Low Latency Mode, Cubase calls it Constrain Delay Compensation, Studio One has a dropout protection setting. Turning that on is usually faster than auditing every plugin by hand, and it does exactly this job.
Step 5: Check the driver
On macOS, Core Audio is built into the system and most current interfaces are class compliant, so there is little to get wrong. On Windows the situation is different and this step matters. You want the manufacturer's own ASIO driver, downloaded from the manufacturer, not a generic wrapper. ASIO4ALL is a wrapper around the Windows audio system rather than a real hardware driver, and it exists to make devices work that have no ASIO driver of their own. If your interface has a real driver and you are using the wrapper, you are giving away performance for nothing.
Also check that your DAW is set to the ASIO driver rather than the Windows WDM or DirectSound path, which adds tens of milliseconds. And check the interface control panel for a separate "safety buffer" or "USB streaming" setting, which some manufacturers expose in addition to the buffer size and which can quietly add more delay than the buffer does.
Connection type matters less than driver quality. Thunderbolt does have a lower floor than USB because of how the bus handles transfers, and it shows up as a couple of milliseconds of difference at the same buffer, but a good USB driver beats a mediocre Thunderbolt one. The full comparison is in USB vs Thunderbolt audio interfaces, and the sample rate side of the tradeoff is in 44.1 vs 48 vs 96 kHz.
The failure mode that costs the most time is the one where nothing is technically wrong. The session is at 128 samples, direct monitoring is on, and the singer still says it feels weird. Nine times out of ten it is not latency at all, it is the headphone mix: too much of their own voice, not enough of the track, or a closed back pair clamped hard enough that they are hearing themselves through bone conduction louder than through the cans. I keep a pair of closed back headphones set up specifically for tracking and I fix the balance before I touch a buffer setting, because a performer saying "something feels off" is almost never diagnosing the actual cause.
What if none of it works?
Then work the way the delay allows. Record the take with direct monitoring and no plugins at all, get the performance, and add everything afterward. This is what people did for the first fifty years of multitrack recording and it produced most of the records you know. The idea that a singer needs to hear the finished vocal chain in their headphones to perform is recent, and it is a preference rather than a requirement.
If your machine genuinely cannot hold 128 samples on a small session, the bottleneck is usually background processes, a power saving mode, or a shared USB bus rather than raw CPU speed. Plug the interface into its own port, set the power profile to high performance, and close anything doing network sync in the background. The signal flow guide maps out where each delay in the chain is introduced, which is what lets you bisect the problem rather than guess at it.
Related tools
- Latency buffer calculator: milliseconds for any buffer and sample rate
- Sample rate latency chart: the full grid in one page
- How to set up a home recording studio: the wider setup this fits into
Frequently asked questions
What buffer size should I use for recording?
Use 128 samples for tracking and 512 or 1024 for mixing, and switch between the two rather than looking for one setting that does both. At 48 kHz, 128 samples is 2.67 ms one way and lands somewhere near 7 to 11 ms round trip once converter and driver overhead are counted, which most players cannot feel. At 1024 samples you get 21.33 ms one way, which is unplayable but lets your plugins run without dropouts.
How do I calculate latency from buffer size?
Divide the buffer size in samples by the sample rate in hertz and multiply by 1000 to get milliseconds. So 128 samples at 48000 Hz is 2.67 ms. That is one direction only. Round trip latency, which is what you actually feel when monitoring through the computer, is roughly double that plus another 2 to 6 ms of converter and driver overhead that no setting in your DAW exposes.
What is direct monitoring and why does it fix latency?
Direct monitoring routes the input signal to your headphones inside the interface hardware, before it ever reaches the computer. The delay is effectively zero because the signal never makes the round trip. Almost every interface has either a hardware mix knob or a small software mixer that does this. It is the first thing to try because it removes the problem entirely rather than reducing it, at the cost of not hearing plugins on the input.
How much latency can a musician actually feel?
Most players stop noticing below about 10 ms round trip, start to feel something is wrong around 15 ms, and find 20 ms or more genuinely disruptive. Drummers and percussionists are the most sensitive, singers are next because they also hear themselves through bone conduction with no delay at all, and keyboard players are usually the most tolerant. Ten milliseconds is equivalent to standing about 11 feet from an amplifier.
Does a higher sample rate reduce latency?
Yes, proportionally, because the same buffer of samples represents less time. A 128 sample buffer is 2.90 ms at 44.1 kHz, 2.67 ms at 48 kHz and 1.33 ms at 96 kHz. The catch is that 96 kHz roughly doubles the processing load, so you often have to raise the buffer to keep the session stable, which gives back what you gained. Lowering the buffer at 48 kHz is usually the cheaper route.
Why did my latency get worse after adding a plugin?
Some plugins introduce their own processing delay, and the DAW compensates by delaying every other track to match. Linear phase EQs, look-ahead limiters, convolution reverbs, pitch correction and many mastering plugins are the usual culprits, and a single one can add 30 ms or more. Delay compensation keeps playback in sync, but it adds directly to what you feel when monitoring a record-armed track through the computer.
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.