How Garmin, WHOOP, Oura, Polar, Ultrahuman, Withings, Amazfit, and Eight Sleep measure HRV

2026-07-02

One body, one recovery state, and a pile of heart-rate-variability numbers that refuse to agree. On matched nights held at the same recovery state, one person’s own devices reported HRV anywhere from 38 to 95. Apple Health said 95, Ultrahuman said 75, Polar said 52, WHOOP said 46, Garmin said 45, Oura said 40, and Fitbit said 38. Everything there is RMSSD except Apple, which is SDNN. The strangest part sits inside the list, not next to it. The same Ultrahuman ring showed 75 in its own app and wrote 95 into Apple Health for the same night. So the real question is not which device is right. It is why the numbers are so different, and the answer is device by device.

First, why two numbers can describe the same heartbeats

HRV is not one measurement. It is a family of ways to summarize the tiny timing gaps between your heartbeats. Consumer devices use two of them, and they are just two different formulas run on the same beats. RMSSD looks at how much each beat differs from the one right before it, so it catches the fast, moment-to-moment wobble. SDNN looks at the total spread of all those gaps across the whole recording, so it soaks up slower rhythms like your breathing and body clock too. Because they measure different things, SDNN usually reads higher than RMSSD from the same person, which is most of why Apple’s 95 sits at the top. The two were never meant to match, so lining them up side by side is comparing apples to a slightly different fruit.

WHOOP

WHOOP measures from the optical sensor on your wrist and reports RMSSD in milliseconds, which its developer API confirms as the hrv_rmssd_milli field. The twist is timing. WHOOP does not average the whole night. Its own education material and patent describe a sleep-weighted figure that leans on your last big slow-wave-sleep episode before you wake, the theory being that deep sleep is the cleanest window on recovery. So a WHOOP HRV is really “recovery HRV from late in the night,” not a flat overnight mean.

Oura

Oura reads from an infrared sensor against your finger and reports RMSSD. Rather than take one reading, it samples in five-minute chunks across the whole night and reports the average of them all. That whole-night average is the raw nightly number. The friendlier “HRV Balance” layer on top compares a rolling two-week average to your three-month baseline, using nighttime HRV only. Oura is one of the more transparent devices here, and it spells all of this out in its own support pages.

Garmin

Garmin measures optically at the wrist, only while you sleep, and reports HRV Status as a seven-day rolling average compared against a personal baseline that takes about three weeks of nights to settle. Garmin’s own docs are clear on the what and the when but never actually print the metric name. It is almost certainly RMSSD, because Garmin bought Firstbeat in 2020 and Firstbeat’s engine is well documented as RMSSD-based, but that last step is industry consensus rather than a line in Garmin’s manual.

Polar

Polar is the most upfront of the bunch. It names RMSSD explicitly, measures optically at the wrist over roughly the first four hours of sleep as part of Nightly Recharge, and compares the result to a baseline built from your previous 28 nights. Polar even explains why it uses the early hours rather than the whole night, arguing those hours reflect recovery more sensitively than a full-night average. So a Polar number and an Oura number can disagree simply because one watched the front of the night and the other watched all of it.

Ultrahuman

Ultrahuman reads from a finger sensor and, by its own account, works with both RMSSD and SDNN. In its own app it shows an RMSSD-style figure, which is the 75 above. When it hands data to Apple Health, though, that data lands in Apple’s single HRV slot, which is SDNN, so the same night resurfaces as 95. That is the whole “one ring, two numbers” mystery. It is not a glitch, it is two different metrics wearing the same label. Ultrahuman does not publish its exact windowing, and its own study notes HRV windowing was not studied, so the finer details here are reasonable inference rather than documented fact.

Fitbit

Fitbit reads optically at the wrist and states plainly that it uses RMSSD, measured during sleep, and needs a sleep block longer than three hours to report anything. The precise construction, the median of five-minute RMSSD windows taken from your longest qualifying sleep period, comes from a peer-reviewed validation of the Versa 4 rather than Fitbit’s own help pages. The practical point is that a Fitbit number is a median of chunks, not a single flat average, which nudges it away from what a whole-night-averaging device would report.

Apple Watch and Apple Health

Apple is the deliberate outlier. Apple Health stores HRV under exactly one type, SDNN, in milliseconds, so every HRV reading in the Health app is an SDNN value no matter which app wrote it. The Apple Watch also does not compute one overnight average. It takes short reads, on the order of a minute, when you are still, including background samples overnight and during Breathe or Mindfulness sessions, so a single day can hold several SDNN samples. Between SDNN reading higher than RMSSD and those short sampling windows, Apple numbers routinely sit above the RMSSD crowd, which is exactly what the 95 at the top shows.

Withings

Withings reports RMSSD and takes an unusual approach, reporting separate HRV for the first and last stretches of sleep rather than folding the night into one figure. Its ScanWatch line also carries an FDA-cleared single-lead ECG you trigger yourself for a clinical-grade 30-second reading, which is a different animal from the passive optical measurement it does overnight. The RMSSD choice and the split-window approach are documented in reputable secondary sources, though Withings does not publish its exact baseline math.

Amazfit and Zepp

Amazfit, whose app is Zepp, reports RMSSD measured overnight through its BioTracker optical sensor at the wrist, and compares it against a baseline drawn from roughly the last seven days of sleep HRV. This one is straightforward and documented on Amazfit’s own pages. It sits in the same RMSSD, wrist-optical, whole-night family as Garmin and Fitbit, which is why devices from that group tend to cluster closer together than any of them do to Apple.

Eight Sleep

Eight Sleep never touches your skin. Its Pod sits under you and reads HRV contact-free through the mattress, detecting the tiny movements your body makes with each heartbeat, a technique called ballistocardiography. It reports RMSSD in milliseconds and says it has validated that against ECG across hundreds of nights. So a bed and a finger ring can both hand you an RMSSD number, arrived at through completely different physics, which is a good reminder that the sensor and its placement shape the result as much as the formula does.

What the number is actually worth

Line those up and the 38-to-95 spread stops looking like a fault. Different metric, different slice of the night, different sensor, different smoothing, and even the six RMSSD devices ran from 38 to 75. None of them is wrong. Each is internally consistent. They are simply not measuring the same thing. The useful takeaway is that the absolute number means very little on its own. HRV is a signal only as your own trend, on one device, against your own baseline. Switching devices resets that baseline, which is why a “drop” the week you change rings is usually telling you about the ring, not about you.

This is why freddy keeps each source in its own unit and its own baseline and never quietly blends them. When your AI reads your data, whether that is Claude, ChatGPT, or anything else, it sees “Oura RMSSD, whole-night average” and “Apple Health SDNN, periodic reads” as the distinct measurements they are, and compares your Oura trend to your Oura history and your Apple trend to your Apple history. That is the honest version of the question you actually care about, which was never “is 95 better than 46,” it was “is tonight better than my normal.”

Method details verified 2026-07-02 against provider documentation and peer-reviewed validation studies. Check vendor sites for current details, and treat HRV as trend information, not medical advice.

Sources

RMSSD vs SDNN fundamentals

Shaffer & Ginsberg 2017, An Overview of Heart Rate Variability Metrics and Norms: PMC full text
Task Force of the ESC/NASPE 1996, Heart rate variability standards: European Heart Journal