Which SD card do I need?
Three questions, and you get one or two cards that fit. Not a list of everything on Amazon: only cards we have actually put on the bench, filled to the last byte and read back. If a card is not here, it is because we have not tested it yet.
CKBench · card finderBuilt on 14 cards we bench-tested, 2026-07-21 to 2026-08-17
What is the card going in?
Pick what the card is going in and the answer appears here.
Every card here was bought or supplied, then run through the same bench: filled to the last free byte and read back, written to for five straight minutes to find its worst stretch, and photographed with a thermal camera. The method is on how we test and score memory cards. Some links are affiliate links, and they never change an answer.
Speed is the wrong thing to worry about
Almost everyone shopping for a dash cam card is trying to buy enough speed. It is the wrong worry. We measured what four dash cams actually write, and the busiest single second any of them produced was 6.09 megabytes per second. The slowest card on our whole bench holds 14.9, and that is its bad-day figure. A V30 stamp promises 30. Everything in the shop is fast enough, several times over.
The two things that do go wrong are less obvious. The first is capacity, because a card can pass every speed test on earth and still be a quarter of the size on its label, which is exactly what one of ours turned out to be. The second is endurance, because loop recording rewrites the same card thousands of times and a cheap controller gives up long before a good one does. Neither is printed on the package. Both are what the finder above weighs, after it checks that the speed is there.
If you want the whole picture rather than a recommendation, the full card picker takes a specific camera and shows every measurement for every card, with the arithmetic in the open.
Two sets of measurements, and nothing else
The card side is our own bench. Each card gets written to until it is completely full, then every byte is read back and checked against what was written, which is the only way to catch a card that is lying about its size. Then it writes flat out for five minutes so we can find its worst ten seconds rather than its best burst, and a thermal camera watches it while it does.
The camera side is footage the cameras wrote themselves. When you tell the finder your dash cam records in 4K, it is not looking up a specification: it is reaching for the real bitrates of three 4K dash cams whose footage we pulled off the card and measured. Same for 2K, and for the second camera in a two-camera setup, which is measured off the rear channel of the one dual-camera car we have recorded with.
That is also why the list of options is short. There is no entry for a resolution we have never recorded at, and no card in the results we have not run. It grows as the bench grows.
For a device we have never recorded with, an action camera or a phone or a drone, the finder asks you for the bitrate instead of guessing one. That is the honest version of covering everything: your number for the recording, our measurements for the cards, and the page says which half is which.
Questions people ask before buying a card
What size SD card do I need for a dash cam?
Work backward from how long you want to keep. A 4K dash cam we measured writes between 4.83 and 11.15 GB per hour depending on the camera, and a second camera adds about 3.81 GB per hour on top. So a full day of driving on a two-camera 4K setup is roughly 150 GB, and a 128GB card gets you most of a day. The finder above does that arithmetic with the measured numbers rather than the ones on the box.
Do I need a special card for a Nintendo Switch 2?
Yes. The Switch 2 needs microSD Express for game storage, and the marking to look for on the card is EX. An ordinary microSD will not run games on it, and neither will a UHS-II card: both have a second row of pins, but they carry completely different signals. That is a platform requirement rather than a measurement of ours, and there is exactly one microSD Express card on our bench so far.
Can I use a microSD Express card in a dash cam?
It works, and it is a waste of money. The Express half of the card is PCIe on a second row of pins, and a dash cam has no way to reach it, so the card falls back to ordinary speeds. We measured the same card both ways on the same day: 212.87 MB/s committed on an Express reader and 87.74 on an ordinary one. In a dash cam you would be paying for the first number and getting the second.
Is a more expensive memory card actually better?
Not reliably, and our bench keeps proving it in both directions. The best-scoring card we have tested undersells itself badly, committing writes at 89.7 MB/s against a printed 40. Meanwhile a full-size card that costs several times more missed its own printed write speed by 17%. What separates them is whether the printed numbers are honest and whether the card holds up under churn, and neither of those tracks the price.
How do I know a memory card is not fake?
Not with a benchmark, which is the trap. The counterfeit we tested passes Class 10, U3, A1 and V30 and posts perfectly respectable speeds, while holding 34.34 GB behind a 128 GB label. The only test that catches it is filling the whole card with data you can verify and reading every byte back, which is what we do to each card here before anything else.
Why are there only a few cards in the results?
Because those are the ones we have tested. Every recommendation here is backed by a full run on our own bench, published with its numbers and its date, and we would rather show you a short honest list than a long guessed one. The corpus grows every time a card comes through, and the finder picks it up automatically.
Want the full measurements?
The technical picker takes a specific camera and shows every card with its measured write floor, verified capacity and endurance, with the arithmetic in the open.
Open the full pickerEvery card we have tested
The corpus table and the full review of each card, including the counterfeit and the two that share one name.
Open the CKBench corpus