Will this SD card keep up with your dash cam?
Pick a dash cam we have measured, or type your own camera's bitrate, and every SD card on the CKBench corpus is judged against it. The answer is built from two sets of measurements and nothing else: what the camera actually wrote, and what the card actually did on the bench. No spec sheets, no manufacturer numbers, no guesses about hardware we have not run.
CKBench · card pickerBuilt on 14 cards we bench-tested, 2026-07-21 to 2026-08-17
Every camera listed here recorded the footage its numbers come from. If yours is not on the list, pick the last option and type its bitrate instead.
In megabits per second, and add the channels together if it records front and rear. Your manual usually prints it. If it does not, the file size of one clip divided by its length gets you there: a 3 minute clip of 420 MB is about 19 Mbps.
2. What that asks of a card
Choose a camera and the measured recording load appears here.
The verdicts land here, one per card, each with the measurements behind it.
Everything above is measured. Camera figures come from footage the camera wrote, read back with dashcheck. Card figures come from one CKBench run per card, and the full method is on how we test and score memory cards. Some links here are affiliate links, and they never change a verdict.
Write speed is almost never what decides it
This is the finding that surprised us, and the picker will keep showing it to you. Across the four dash cams we have measured, the busiest single second of recording asked for between 2.12 and 6.09 MB/s, and the sustained rate over a whole drive ran between 1.34 and 3.38 MB/s. The slowest microSD in our corpus holds 14.9 MB/s in its worst ten seconds, and that is its bad-day figure, measured right after a burst of random writes. It clears the hungriest of those cameras by a factor of 2.4.
Put the same numbers against the mark on the package and the gap is wider still. A V30 stamp is a promise of 30 MB/s. The most demanding dash cam on this bench, a 4K front plus a 1080p rear both writing to one card, peaked at 6.09 MB/s. That is a fifth of the floor the cheapest V30 card is rated for.
So the picker checks the write floor first, because it is the question everyone asks, and then spends most of its answer on the two things that actually separate these cards: whether the capacity on the label is really there, and how the card behaves under the churn that wears one out. One card in our corpus fails outright, and it fails on capacity while passing every speed test we can throw at it.
Three measurements, and where each one comes from
1. The write floor, against the camera's busiest measured second. The card figure is its worst ten-second stretch out of five minutes of continuous committed writing, which is a much harsher number than any burst benchmark: every byte had to land on the card before the clock stopped. The camera figure is the highest one-second window in the footage it actually recorded. Comparing the two is deliberately pessimistic on both sides, because a dash cam that drops frames on the one busy second is the failure that matters.
2. Capacity, verified byte by byte. Every card in the corpus was filled to the last free byte with seeded data and read back, so the figure is what it handed over and proved, not what it reports as its size. Anything under 99% of the label is called out with the gigabytes it is missing. Anything under 90% is not a rounding story, and the picker rules that card out no matter how fast it is.
3. Endurance and claim context. The endurance figure is how much random-write speed the card gave up across three churn passes. It is an indicator, not an endurance test, and nothing done in one session can be one. Alongside it sit the two claims a shopper can read off the package, the printed write speed and the printed video speed class, each next to what the card measured.
Two smaller rules run underneath. A full-size SD card cannot fit a microSD slot, so the picker moves it to the ruled-out list rather than ranking it. And a card whose second row of pins carries PCIe needs a slot that speaks PCIe: our microSD Express card was run twice, once on an Express reader and once on an ordinary one, and a dash cam gets the ordinary result. That is the row the picker shows it.
What this cannot tell you
It cannot tell you about a card we have not tested, or a camera we have not recorded with. There is no database of manufacturer specifications behind this page and there deliberately is not going to be one: every number here exists because something was measured, and a spec sheet is a claim rather than a measurement. That is why the fallback is a box you type your own bitrate into rather than a list of cameras we have only read about.
It cannot promise your card will still be working in three years. Endurance is a property that only time settles, and the fade figure is the best directional evidence a single session can produce. It cannot tell you whether your specific camera has a firmware quirk with a specific card, which happens and which no bench can predict. And it cannot tell you anything about a counterfeit you have not tested, only that our one counterfeit passed every speed test while holding barely a quarter of its label. If that worries you, the fix is not a benchmark. It is filling the whole card and reading it back.
Where a number is soft, the picker says so on the card itself rather than hiding it in a footnote.
Common questions about dash cam SD cards
What write speed does a dash cam actually need?
Less than almost anyone expects. Across the four dash cams we have measured, the busiest single second of recording asked for between 2.12 and 6.09 MB/s, and the average over a whole drive ran between 1.34 and 3.38 MB/s. The most demanding of them records 4K at the front and 1080p at the rear onto one card. A V30 rating is a promise of 30 MB/s, roughly five times what that camera peaked at.
Do I need a V30 or V60 card for a 4K dash cam?
On the cameras we have measured, a card holding V30 has several times the headroom the footage asks for, and every microSD in our corpus holds V30 or better in its worst ten seconds. The class mark is still worth reading, because it is the one printed number with a defined floor behind it, but treat it as a lower bound rather than a target to beat. Capacity and endurance separate these cards far more than the class mark does.
Can a memory card be too slow for a dash cam?
Yes, and the picker will show it the moment your camera's demand crosses a card's measured floor. It is a real risk on a high-bitrate multi-channel camera, which is exactly what the bitrate box is for. It just is not the risk it is usually sold as, because no card in our corpus falls under any dash cam we have recorded with.
Why does my 128GB card only show about 119 GB?
That is normal and nothing was taken. Your computer counts in binary, so 128,000,000,000 bytes displays as about 119 GB. The picker sidesteps the confusion by reporting what each card actually wrote and verified in decimal gigabytes, against the decimal number on its own label. A real shortfall looks completely different: our counterfeit delivered 34.34 GB against a 128 GB label.
Should I replace the card that came with my dash cam?
Not automatically. Three bundled cards have been through this bench and all three delivered what they promised, with the Kioxia beating its own published write spec by 31%. Run your bundled card through the picker if it is in the corpus. If it is not, the thing worth checking is its capacity, not its speed.
How many hours of footage will my card hold?
The picker prints an estimate on every card once you have chosen a camera, using the gigabytes per hour that camera actually recorded and the capacity the card actually delivered. For a card or a camera outside the corpus, the dash cam storage calculator does the same arithmetic from numbers you supply.
Where these measurements come from
The card side is CKBench, our storage test bench. Every figure comes from one results file per card: writes are committed and confirmed before the timer stops, a cache-leak canary aborts the run if the operating system serves a cached read, and every test byte is seeded so it can be verified on the way back. The camera side is dashcheck, which reads footage the camera already wrote and never asks the manufacturer anything.
Every card we have tested
The full corpus table: committed write, sustained floor, verified capacity, peak temperature and a published score for each one.
Open the CKBench corpusHow we test and score
The bench, the five-dimension rubric, the unit conventions, and an honest list of what we deliberately do not test.
Read the method