SanDisk High Endurance 512GB: the stall a speed class cannot see

Why a used card is the most interesting High Endurance card I have tested

Every other endurance card on this bench arrived fresh. This one arrived with a history: it spent months in my VIOFO A229 Plus, a camera that had been flagging a slow card intermittently, with this card and with others before it, and only then went on the bench. That makes it the closest this bench can get to the question the High Endurance line is actually sold on, and it comes with a confound that has to be said up front. A weak result on a used card can be use rather than design, so every number here belongs to this unit and none of them transfer to the model. It joins CKBench, the running index of every card on the bench, alongside the SanDisk High Endurance 128GB, three fresh units of which give this card its comparison set, the SanDisk MAX Endurance 128GB, the tier above it, and the Vantrue Endurance PRO 128GB, the other card here that needed two readers to settle.

One unit, used, tested on August 25 and 26, 2026, and published September 15. Six runs across two independent card readers, four of them complete, two of them whole-card fills. The card face is the only claim source I have, because a used card comes with no box; the 100 and 40 MB/s figures and the 40,000-hour rating come from SanDisk's own datasheet.

What the card claims, and what it did

The card face prints the capacity, C10, U3, the UHS-I mark and V30, and nothing else: no speed figure, no model code. The read and write speeds and the endurance rating are SanDisk's datasheet figures, and the table says so on each row. Every claim is met. Read the table knowing that none of its rows can see the stall, because a claim is a floor and a floor is a ten-second average.

The read figure is attributed to the SanDisk QuickFlow because that reader has produced 203 MB/s on another card, so 180.4 there is the card's own limit. The ProGrade's microSD slot tops out at about 181, so its 180.5 could have been the reader; the two agreeing settles it. The A-class row is not a failure. The High Endurance line prints no A mark, and A2 is untestable on any USB reader for any card in any case.

What is on the card, and what its registers add

A used card has no box, so everything printed is on the face: SanDisk, High Endurance, 512, the microSDXC logo, U3, the UHS-I mark, V30 and a circled 10. No read or write figure, no model code, no lot code. The registers fill in what the print leaves out. This card reports itself as product name SN512, revision 8.5, built August 2025, which is the same silicon generation as the third of my three High Endurance 128GB units, the one that reads 175 MB/s where the other two read 97. So a read near 180 is what this generation does, not a surprise and not a shortfall, and the full table is in the identity appendix at the end.

Why does a card that holds V60 throw a slow-card error?

Because it stops. The bench logs the longest single blocked write of every five-minute sustained test. On my primary reader, a ProGrade PGM0.5, one block of this card's stream took 3,636 milliseconds to land. I put the card in a SanDisk QuickFlow, a reader with a different controller chip on a different USB path, and ran the same test: 3,266 milliseconds. Both stalls happened at a chunk boundary, both were the only block over half a second in their run, both show up in the one-second samples as two consecutive seconds at exactly 0 MB/s, and both have the same shape. The write rate collapses over three or four seconds, the card sits dead for two, and then it is back at full speed inside a second.

Two readers matter more than any number on this page. A stall seen on one reader cannot be separated from that reader, and this bench has been burned by exactly that: one session where a sensor utility polling the reader produced five-second stalls on three healthy cards, all since voided. A fault caused by the host follows the session. This one followed the card to a second host and reproduced to within ten percent. It is a property of this card.

It is not a one-off, either. The two whole-card fills record only a throughput sample every second, so a dead stop shows up there as consecutive seconds at zero. The first fill has five of those clusters and the second has six, spread through the hour it takes to write 511.8 GB. Across everything this card wrote under the bench, about 1.4 TB over three hours, that is thirteen dead stops of a second or more: one every five to seven minutes of continuous writing. Underneath the dead stops there is a milder rhythm as well, slow patches of five to seven seconds at 15 to 25 MB/s that on the QuickFlow arrived about every 30 seconds. The worst ten seconds of every run is one of those patches, which is why the class figure moved between 52.0 and 68.1 MB/s across four complete runs and why one run fell to V30.

The V-class system cannot see any of this, and that is the point worth taking away. A two-second stop inside a ten-second window on a card averaging 140 MB/s still leaves the window comfortably above 60. The run containing the 3.6 second dead stop scored 68.1 through its worst ten seconds and holds V60, a full class above the V30 printed on it. Nothing in the rating measures the instantaneous floor. A dash cam does. It has a small buffer and a continuous stream, and when the card stops accepting data for long enough the camera reports a slow card. That is the warning my A229 Plus had been showing with this card in it, and this stall is a plausible cause of it. Plausible, not proven: no camera was on the bench. One more thing the runs say about where it lives: the 30-minute wear probe, which rewrote one region in place, produced no dead stop at all, while every leg that grew new files did. Within this card the trigger is allocating new space. Across the bench the discriminator is the card: every other card's fill grows files the same way, and none of them stopped.

Is it alone? Yes. I scanned every run file on this bench, 68 runs across 42 cards, for the longest blocked write. No other card has a single block over one second on any reader, and none over two. The next-worst genuine card that holds its label, a Newegg TeamGroup, tops out at 221 milliseconds, which makes this card's stall 16 times longer than anything a genuine card here has produced. The three fresh High Endurance 128GB units I own top out at 168 milliseconds between them with zero blocks over half a second.

How fast is it when the writing does not stop?

This is the number that decides a video speed class, and it is always the worst ten-second window rather than an average, because a camera does not get to record over the mean. Five minutes of continuous committed writing with the cooling fan off, sampled once a second. The panel below is the primary run, and you can watch the stall happen at 2:34.

V60 on three runs out of four, and V30 was never in danger

Away from the patches this card is fast and flat: 133.1 MB/s steady state, no cache cliff anywhere, and a write rate that ended the five minutes fractionally above where it started. The verdict window came in at 68.1 MB/s on the primary run, 66.0 on the second reader, 62.7 on an earlier run, and 52.0 on the one run where a slow patch stretched to seven seconds. Three of the four hold V60. The lowest worst-ten-seconds this card produced on any run is 52.0, which is 1.7 times the V30 floor it prints, so the printed class was never at risk. What was at risk was the recording.

One disclosure about the primary run. It was launched without the settle pre-pass, so its sustained test started on a controller still busy from the random-write test before it. That did not set the number: its worst window is the highest of the four complete runs and sits at 2:58, not in the first minute. The second reader's run was settled and stalled the same way.

Does a used endurance card fade under the writes loop recording makes?

Loop recording is not one long write. It is thousands of small scattered writes as the camera overwrites old clips, and a card that was not built for it gives up random-write speed pass after pass. Three 20-second passes of random 4 KB writes, compared to the reference card that collapsed 71% on the same test.

It does not fade, and its last pass was its fastest

728, 728 and 738 write IOPS across the three passes: no fade at all, on a card that has already spent months being written to in a dash cam. A 30-minute wear probe run afterward, rewriting one 1 GiB region in place 204 times on the full card, moved 0.41% between its first and last rewrite and verified clean. Nothing here says the card is wearing out. What the wear probe did show is the shape of the fault: seven blocks over half a second inside those 30 minutes, but no dead stop, where streaming across the card produced one every few minutes. The stall seems to belong to writing new ground, not to rewriting old.

Is all 512 GB really there?

The only test that catches a fake, and on a used card it is also the strongest statement the bench can make: whatever the card has been through, every cell still holds what it is told. Every free byte written with unpredictable data, every byte read back and compared.

511.8 GB, verified twice, zero errors

511,802,474,496 bytes written and the same number verified, zero mismatched bytes and zero read errors, and the fill the day before returned exactly the same byte count with the same clean result. That is 99.96% of the 512 GB label, the top of the range genuine cards occupy on this bench, and the 0.2 GB shortfall is what the exFAT volume itself costs. Your computer will display it as about 476 GB because it counts in binary. Write speed at the start, middle and end of the card sat within 3.8% of each other, and read within 1.0%, which is one consistent piece of silicon end to end. The dead stops live in this leg too: six clusters of zero-throughput seconds across the 66-minute fill.

How hot does it get, and does it slow down?

Peak temperature at the end of five minutes of uncooled writing, read at the reader slot, with the write speed drawn beside it so a throttle would show as a bend.

Hot, third-hottest on the bench, and it never slowed down

174 °F, which is 78.9 °C, at a 76 °F ambient, and all four complete runs returned that same reading independently. Only one card I have tested runs hotter, the Samsung P9 Express, and the SanDisk Extreme 512GB ties it. The write rate never bent: the five-minute trace ends marginally above where it started, and the whole-card fill held 135.9 MB/s for over an hour. So this is a headroom finding rather than a throttling one, and the headroom is 6 °C to the roughly 85 °C these parts are typically rated for, measured in an air-conditioned room. It is doing a lot of work for that heat, which is why it lands mid-pack on heat per unit of throughput, but a card sold to sit behind a windshield has less margin than almost anything else here.

How does a used High Endurance card compare to three fresh ones?

I own three fresh High Endurance 128GB units, all benched, so the comparison the used card was bought for is available. One caveat sits on top of it: a 512 GB card has more flash dies to write to at once than a 128 GB card, so its throughput is expected to be higher and says nothing about wear. What the fresh units can answer is narrower. Do fresh High Endurance cards stall? Across four runs between them, their longest blocked write is 168 milliseconds, with zero blocks over half a second.

The used 512 GB is not slower than any fresh 128 GB unit on any throughput metric, its churn is flat, its capacity is byte-perfect, and it is the only card of the four that stalls. Whether that stall is wear or design is the one question this table cannot answer, because the cards differ in capacity as well as in history. A fresh 512 GB unit is the experiment that settles it.

What does 40,000 hours actually mean on a 512GB card?

SanDisk's datasheet opens with "up to 40,000 hours" and puts the qualifier in a footnote: that is the 512GB figure, the top of a ladder that runs 2,500, 5,000, 10,000, 20,000 and 40,000 hours for 32, 64, 128, 256 and 512 GB. This card is the capacity the headline actually belongs to. Divide every rung by its capacity and the ladder is exactly linear at 78.125 hours per gigabyte, so the rating is the capacity times a line constant rather than a per-capacity measurement. The number that gives it meaning, the recording bitrate, is printed on the High Endurance retail box and appears nowhere in the datasheet. A used card has no box, so this one carries no bitrate at all. If the line's 26 Mbps condition applies, 40,000 hours works out to about 468 TB written, or roughly 914 complete fills of this card. The same datasheet warrants it for two years; at 24/7 the rating implies 4.6 years.

None of that is a measurement and I claim none of it. Nothing done in a session tests 40,000 hours. What the bench measures are proxies, sustained write, churn and heat, and on this card it found a behavior the rating does not describe at all: a card can be rated for 40,000 hours of continuous recording and stop accepting the recording for two seconds every few minutes.

Why this review has six runs instead of one

The first two runs stopped inside the sequential test and are excluded, though the first is worth a line: its warm-up write ran at 14 MB/s and its read at 16, and forty minutes later the same card in the same reader ran at 138 and 177. I cannot explain a tenfold cold start that never came back, so it is recorded and not published as a finding. Runs three and four completed the day before the two runs the figures come from. Run three held V60 with a 723-millisecond block; run four fell to V30 and verified the whole card. Run five, the primary, is the full suite with the fill, the verify and the wear probe, and it carries the 3.6-second stall. Run six is the same card in the second reader the same afternoon, and it carries the stall too. The scores use run five, with run six as the confirmation that the fault belongs to the card.

The card did not arrive at the bench empty; I emptied it. Before the first run I pulled every clip off it, from a card the camera had filled almost to the last clip pair, which is where the A229 Plus review gets its numbers on locked clips and on what a driving hour costs in storage. Then I formatted it, because a format is the first thing anyone tries against a slow-card warning, and the bench saw a clean exFAT volume. So the history was read before it was erased, and the runs measured the card in the state a troubleshooting owner leaves it in.

How the 4 breaks down, and the rule this card added

The bottom line

Bottom line: this is a genuine 512 GB card that beats every figure SanDisk publishes for it, holds a video class above the one it prints, does not fade under churn after months of dash cam service, and stops writing for about two seconds every few minutes. I reproduced the stall on a second card reader, which is what makes it the card's fault rather than my bench's, and I found nothing like it in 68 runs across 42 other cards. It is a plausible explanation for the slow-card warning my A229 Plus kept showing with this card in it. It is also one used unit, and I will not tell you the 512GB model does this until a fresh one either does or does not. If yours has thrown that error, the card can be the cause even when every speed test says it is fine, and a full write of the card is the only test that will show it.

Disclosure: I bought this card myself, some months before this test, and ran it in my own VIOFO A229 Plus for months before it went on the bench. It is a used unit and is described as one throughout. 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.