FitcamX 64GB microSD: the card that looked broken

Why test a card nobody chose to buy?

Almost every dash cam arrives with a memory card already in the box, and almost nobody knows whether that card is any good. It has no packaging, no spec sheet, and usually no brand that means anything. The advice you find online is to throw it away and buy a real one, which is easy to say and costs you money that may not need spending. This is the third bundled card I have run through the full bench, so it is becoming possible to answer the question properly rather than guess. It joins CKBench, the running index of every card on the bench, alongside the Botslab 128GB that shipped with a four channel camera, the Kioxia Exceria G2, and the counterfeit ImageMate that shows what a genuinely bad card looks like.

This one came inside the FitcamX, the tidy little screen-free camera I reviewed here. One unit, tested on 2 August 2026. It was not bought separately, because it cannot be.

What the card claims, and what the card did

There is no box, so every claim below was read off the face of the card itself. That is a shorter list than usual, and the gaps are as informative as the entries: this card prints no speed figure of any kind, no V-class mark, and no A-class mark. The only sustained throughput promise it makes anywhere is U3, which asserts a 30 MB/s floor.

Two things follow from that table. There is no headline speed claim to check, which is unusual here and means the card cannot be caught over-promising: it never promised. And the absent V and A marks are not failures. A card is only judged against what it claims, and this one claims C10, U3 and UHS-I. It meets all three.

One caution on the U3 row, because it is easy to misread. Clearing a 30 MB/s floor is exactly what a V30 rating would also require, but this card carries no V30 mark and I am not awarding it one. The measurement is real; the certification is not something a bench can grant.

Does the 64GB card actually hold 64GB?

Yes, and this was the first thing I checked rather than the last. The card face does not carry the SD Association's microSDXC logo. In the exact spot where that mark belongs, set in the same layout and typeface, is a substitute that reads "Memory card" over "XC", with the SD swoosh replaced by the word card. That is a transcription of what is printed, not an accusation. Bundled cards are commonly generic, and the SD logo is licensed, so a supplier filling a camera box has little reason to pay for it.

But it does remove the one cheap assurance a retail package normally gives you, so the card earned a full destructive capacity fill before anything else was measured. I wrote every usable byte, read all of it back, and compared. The result: 62.3 GB written and verified, zero mismatches, zero errors. The card holds what it says it holds.

If you plug this card in and Windows says 58 GB, nothing is wrong. Card labels count a gigabyte as a billion bytes, and Windows counts it as 1,073,741,824. The same 62.3 GB of real space gets printed as 64GB on the label and displayed as 58.0 GB by your computer. It is two ways of writing one number. A fake card looks completely different: it reports a size it does not have and then silently destroys anything written past the real end, which is what happened with the counterfeit ImageMate I tested. This card does the honest thing.

Why did the speed chart look broken?

Here is where this card got interesting. Watching the sustained write test run live, the graph did not look like a card slowing down or speeding up. It looked like it was tearing itself apart: a fast zigzag between roughly 8 and 83 MB/s, with the size of that zigzag itself swelling and collapsing every half minute. A sawtooth inside a sawtooth. Nothing else on this bench has ever drawn that.

Taken at face value, the numbers are alarming. Across 300 one-second samples, 84 of them, 28 percent, fall below 30 MB/s. The slowest reads 8.4 MB/s. If that were the whole story this card would be unusable.

It is not the whole story, and the giveaway is that averaging just two neighboring seconds together removes about nine tenths of the variation, while averaging three seconds together makes it worse. Only one thing behaves like that: a signal that flips back and forth roughly every two seconds, recorded once per second.

My bench records at one-second resolution on purpose. The verdict on any card here rests on its worst continuous ten seconds, and one-second samples are what that measurement is built from. What no sampling rate can do, though, is draw a rhythm that lands close to its own interval. It is the same reason a spoked wheel appears to stand still or spin backwards on film: when the thing you are watching repeats at nearly the rate you are watching it, you see the interference between the two rather than the motion itself. At 2.13 seconds per cycle there are barely two readings per cycle, so the raw trace shows that interference, not the card's actual output.

Because my sample clock and the card's rhythm drift slowly against each other, I could fold all 300 samples back onto a single cycle and recover what the card is really doing between my readings. The lower panel below is that reconstruction.

The card runs on a cycle of 2.13 seconds. Within each cycle its speed rises and falls once, smoothly, and then it does it again. What makes this more than a curiosity is what stays constant: every cycle commits very close to 96 MiB, regardless of how fast the card happens to be running. I split the full-card fill into six separate stretches and measured each one. The average speed moved by 1.4 percent between them and the cycle length moved by 1.2 percent, but they moved in opposite directions, and the amount written per cycle held to within half a percent: 96.05 MiB, give or take 0.19. When the card slows down, the cycle stretches to match. It is counting bytes, not seconds.

That number is a fixed chunk of the card's internal housekeeping, the block size it works in when it tidies up and moves data around. I will not pretend to know which mechanism it is from the outside. What I can say is that the bench measured a repeating 96 MiB unit of work, and that no memory card I have tested has ever exposed one this clearly.

One last point that matters more than any of it: the card never actually stalls. Across 13,489 individual one-megabyte writes, the slowest single write took 154 milliseconds, and not one exceeded half a second. There is no freeze, no dropout, no moment where the card stops accepting data. It is smoothly modulating how fast it accepts writes, not pausing.

Does any of that matter in a dash cam?

No, and it is worth being concrete about why. The slow part of each cycle lasts under a second. Every dash cam buffers video in RAM before writing it, precisely so that a brief dip cannot drop a frame. A dip that short, on a card averaging 47.1 MB/s, is invisible to the camera.

The measurement that decides the verdict is the worst ten second stretch of the whole test, because ten seconds is long enough that no buffer saves you. That figure is 43.8 MB/s. A 4K30 dash cam needs about 6 MB/s. Even the most demanding format I test against, high bitrate 4K120, needs 35. The card clears all of them.

I will add one honest caveat that applies only to this card. Because its cycle is 2.13 seconds long, a ten second window does not contain a whole number of cycles, so exactly where that window lands changes the answer a little: a nine second window gives 41.1 MB/s and an eleven second window gives 41.9. The verdict does not move, since all of those clear 30 MB/s comfortably, but the honest way to state the result is "roughly 44 MB/s" rather than a figure carrying decimal places it has not earned.

Does the bundled card overheat?

It runs cooler than anything else I have measured. Under a continuous full-card write with the ambient at 76 °F, it peaked at 131 °F. For comparison, the two VIOFO High Endurance cards are the hottest on file at 160 °F and 153 °F, the Kioxia reached 142 °F, and the Botslab 135 °F. That is a 55 °F rise over ambient here, and a dash cam on a sunlit windscreen is a hot place to live, so a card that does not add much of its own heat is doing you a favor.

Consistency across the physical card was also excellent. Writing to the first, middle and last tenth of the card gave 47.3, 47.0 and 46.7 MB/s, a spread of 1.2 percent, and reads varied by 0.8 percent. Speed did not fall off as the card filled. There was also no cache collapse: many cheap cards write quickly into a small fast buffer and then fall off a cliff once it is full, and this one simply did not, holding its rate all the way through a full 62.3 GB fill. After the random-access tests had churned it, it returned to full speed within 90 seconds.

Who actually made this card?

Every SD card carries a small identity register that names its manufacturer, and reading it needs a real SD host rather than a USB adapter, so I read this one on a Raspberry Pi wired directly to the card. The answer is not a household name. The manufacturer ID comes back as 0xFE, which my system's lookup table does not recognize, and the product name is the four character string SZYL. Combined with the unlicensed logo, the picture is a white-label supplier rather than a flash vendor you have heard of.

The more interesting result came from comparing it with the Botslab 128GB I tested the same day, a card from a completely different camera brand. Both report the same manufacturer ID of 0xFE and the same product name, SZYL. They differ in their OEM code and were made a month apart, and they deliver almost identically honest capacity: 97.33 percent of label here against 97.37 percent there. Two unrelated dash cam brands, buying bundled cards from the same supplier.

None of that is evidence of anything dishonest, and it is worth saying plainly because unfamiliar identity fields set off alarm bells they should not. Both cards passed a full capacity verification. The counterfeit in this corpus went the other way entirely: it wore a flawless SanDisk identity and was lying about its size. A scruffy label and an honest card are perfectly compatible, and that is exactly what this is.

How the FitcamX 64GB scores

The bottom line on the FitcamX 64GB

I went into this expecting to tell you to throw the bundled card away. Instead it holds every byte it claims, writes faster and more evenly than the card in a rival camera, runs the coolest of anything on my bench, and never once stalled across a full-card write. The thing that looked like a fatal flaw, that violently swinging speed graph, turned out to be the most interesting engineering detail I have found on a memory card: a fixed 96 MiB unit of internal work, repeating every 2.13 seconds, visible from the outside only because my sampling rate happened to collide with it.

What it lacks is paperwork. No warranty, no endurance rating, no licensed SD logo, no manufacturer you can hold to anything, and no way to buy a replacement that is guaranteed to be the same card. If the footage on it might one day matter to an insurer, that absence is a real reason to buy something with a name on it. If you just want to know whether the card in the box is quietly letting you down, this one is not.

Disclosure: The card was not bought or sent to me on its own. It arrived inside a FitcamX dash cam that I purchased myself, and it was tested destructively, which is why there is no unit to return. This post contains affiliate links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you. Affiliate links like these and my YouTube channel are what keep the site running.