To test SD card health you need three tools, not one, because they answer three different questions and only one of them can be answered by a single number. A full-surface test proves the card holds data across its entire capacity today. A speed benchmark shows whether it can still sustain the write rate your camera demands. Neither tells you how long the card will keep holding charge in a drawer.
For anyone rotating a stack of cards through paid work, the goal is not a pass or fail verdict. It is a baseline and a trend, so a card that is quietly degrading gets retired on your schedule rather than failing in the middle of a shoot.
Three Tools, Three Different Questions
The three utilities most often recommended for card testing measure different things, and using the wrong one produces a false sense of security.
| Tool | Platform | What it actually measures | What it cannot tell you | Typical runtime |
|---|---|---|---|---|
| H2testw | Windows | Whether the card stores and returns every byte across its full capacity | Speed under sustained load, future reliability | 20 min for 32 GB, 60-90 min for 128 GB |
| F3 (Fight Flash Fraud) | macOS, Linux | The same capacity-and-integrity check, with a faster write-only mode | The same limits as H2testw | Comparable to H2testw |
| Blackmagic Disk Speed Test | macOS, Windows | Sustained read and write throughput on large blocks | Whether the card is structurally healthy | Under a minute for the default run |
H2testw and F3 are integrity tests. They write numbered files until the card is full, read each one back, and compare. A card that cannot store what it claims, or that silently fails to return a block it accepted, cannot pass that comparison. Blackmagic Disk Speed Test is a performance test: it reports megabytes per second and, in video mode, which frame rates and codecs the measured throughput would support.
Only the combination is useful. A card can pass a full-surface test and still be too slow to record 4K. A card can hit its rated speed and still be ten percent short of its label capacity.
What a Full-Surface Test Actually Proves
The premise is simple and brutal. The test fills the card with verifiable data and then reads it back, so any block that accepts a write but returns something different is exposed. That covers the two failures that matter most in practice: a controller that is lying about capacity, and NAND blocks that have stopped holding data reliably.
Two details about the test matter more than its verdict.
It is destructive and it is not free. A full-surface test writes the entire card, consuming one program/erase cycle across the whole device. On a card rated for a few thousand cycles that is a rounding error, but it is a reason not to run the test weekly. Test empty cards, before they go into service, not cards in the middle of a job.
It tests retention at reading time, not over time. This is the counterintuitive part, and it is the reason a passing result should never be treated as a warranty. Flash cells hold their charge less securely as they wear, so a cell can read back perfectly this afternoon and lose that charge three weeks later without any warning in between. Nothing about a consumer write-and-verify test measures how well the card will hold data in a drawer. Because wear drives that retention window down, the more important signal is often the trend rather than the result — the mechanism is covered in the notes on NAND wear and card lifespan.
A card that fails a full-surface test outright is done. A card that passes has shown that it works today, which is a weaker statement than it sounds.
What Blackmagic Disk Speed Test Adds
Blackmagic Disk Speed Test answers a question integrity tools ignore: can this card still keep up with the camera?
That question matters because a card’s advertised rating describes its best case. Real sustained write throughput depends on the controller, the flash, the reader, and the bus, and it falls as the card wears, since the controller has to spend more time on error correction and on finding healthy blocks. A card that recorded 4K without dropping frames last year can become marginal this year without producing a single error message.
Two habits make the number meaningful:
- Compare against the card’s own history, not against the label. Write down the result when the card is new. A steady 80 MB/s is fine for that card; a fall to 30 MB/s on the same reader is a signal.
- Run it long enough to leave the cache. This is the second counterintuitive point on this page. Many cards write fast for the first gigabyte or two and then settle into a much slower steady state once their internal cache is consumed. A short burst benchmark measures the cache, not the card. For anything above 100 MB/s, run a sustained write or copy a file larger than 8 GB and time it.
Also keep the reader constant. A reader with a slow bridge can cap throughput well below what the card can do, which makes an unrelated bottleneck look like card degradation. The gap between ratings and real-world throughput is covered in the notes on memory card speed classes.
How Often to Actually Test
Cadence should follow exposure, not the calendar.
| Situation | What to run | Why |
|---|---|---|
| A card arrives new or used | Full-surface test, then a speed baseline | Catches a counterfeit or a worn used card before it goes into service |
| Before a paid shoot, cards in heavy rotation | Speed test on each card | A five-minute check beats discovering a marginal card on location |
| Quarterly, for cards in light use | Full-surface test on empty cards | Catches blocks that have failed since the last check |
| Any card kept after an unexplained fault | Full-surface test, then decide | A card with one unexplained unreadable file has already given a warning |
| A card of unknown origin, or from a bundle deal | Full-surface test before anything else | No service history means no assumptions are safe |
The capacity test comes first for a new card because a counterfeit cannot be fixed later. A card that reports a capacity it does not have will destroy files the first time it is filled, and the failure looks like corruption rather than fraud. The full procedure for that specific case is in the counterfeit card guide.
The Signals That Mean Retire the Card
Age is a poor criterion on its own. A lightly used card can be healthy at eight years; a card that lived in a dashcam can be finished at eighteen months. Retire on the first signal that applies, and copy everything off the card before running any test that writes to it.
| Signal | What it means | Action |
|---|---|---|
| Two unexplained unreadable files | Wear has passed the point where ECC covers it | Copy everything off, test, then retire |
| Sustained write speed far below baseline | The controller is spending time on error correction or bad-block management | Retire from main-camera duty |
| Read-back mismatches during a full-surface test | Cells that no longer hold data reliably | Retire the card |
| Reported capacity lower than the label | Counterfeit card, or a controller remapping around dead flash | Retire; check the counterfeit symptoms |
| The card drops out mid-transfer and returns | Controller instability, often heat-related | Retire; see the notes on card error messages |
| The card becomes read-only or write-protected | The controller’s last-resort protection state | Never carry primary footage on it again |
Retiring a card is not the same as discarding data. A card that has started throwing errors is often still readable, and its existing files may be perfectly intact even though it can no longer be trusted to keep new ones. Copy the contents, verify the copies, then decide whether the card is worth a recovery pass before it becomes a spare for non-critical work. The line between recoverable and genuinely gone is drawn in when data is unrecoverable.
Handling and Storage Affect Every Result
A test result describes a card in the state it was tested, and the variables are worth controlling.
Keep cards in a case rather than loose in a bag, since dust and grit in the contact recesses produce exactly the intermittent contact problems that look like controller failure. Store them cool and dry, because elevated temperature accelerates charge loss and shortens retention. Eject properly from computers rather than pulling the reader, and use a reader rather than the camera for transfers, since camera firmware can write to the volume. This is routine care rather than technique, and it is collected in the memory card care guide and the notes on safely ejecting a card.
One practical caution for anyone swapping cards between bodies all day: label them. A test result you cannot attribute to a specific card is not a baseline, and a rack of identical black cards makes attribution impossible.
If you are on Windows and a failing card is still detected by a computer, a purpose-built recovery tool such as XRecovery (xappset.com) can scan SD, microSD, and other memory cards for deleted or formatted photos and videos. As with any recovery tool, stop writing to the card first and always save recovered files to a different drive. Run the recovery pass before the card is reformatted or reused, because data that has been overwritten cannot be recovered by any tool.
Frequently Asked Questions
What is the best free tool to test SD card health?
H2testw on Windows and F3 on macOS and Linux are the standard choices for integrity, because both write the full capacity and verify every byte on the way back. Add a throughput benchmark such as Blackmagic Disk Speed Test to catch cards that are healthy but too slow. Together the three cover capacity, integrity, and sustained speed.
How do I know when to replace an SD card?
Replace it on the first clear signal: two unexplained unreadable files, sustained write speed well below the card’s own baseline, read-back errors in a full-surface test, a capacity below the label, or a card that drops out and returns mid-transfer. None of these improve with further use.
Does testing a card wear it out?
A full-surface test writes the entire card once, which uses one write cycle across the device. That is negligible against a rating of a thousand or more cycles, but it is a reason to test empty cards periodically rather than run integrity tests every week. Speed benchmarks are effectively free.
My card passes every test. Is it safe for a paid shoot?
It is safe today, in the configuration you tested it in. A passing result cannot measure how well the card will hold data three weeks from now, and it cannot reproduce the thermal or power conditions of a long recording session. Test, baseline the speed, and carry redundancy for work that cannot be reshot.