SanDisk Extreme 512GB Review: It Beats Its Read Claim, and Writes Small Files Slowest in Its Class

What the package promises, and what the card delivered

Everything below was measured on a ProGrade PGM0.5 reader, microSD slot, with cardcheck v0.8.2 under the CKBench protocol, except the read row, which needed a second host and was re-measured on a SanDisk QuickFlow with cardcheck v0.8.3 on 25 August 2026. Every write figure is a committed write, flushed to the card before the timer stops, which is why these numbers read lower than the ones on the box and lower than most consumer benchmarks.

Why the read claim needed a second reader

On my main reader the card measured 180.5 MB/s, which is 95.0% of its printed 190. Ordinarily that is a miss. It was not, and the tell was that the number refused to move.

That is 180.4 give or take 0.6 MB/s across four orders of magnitude of transfer size, and across every physical region of the card. A card limited by its own flash varies with position or size. A ceiling this flat, that never shifts no matter what you ask of it, is the transport, not the media.

It was also the highest read figure this bench had produced at the time, beating the previous best of 174.94 MB/s, and it sat 6.1% above the UHS-I DDR200 window the tool watches for, which is why no interface-ceiling flag fired. SanDisk's own fine print says the 190 MB/s figure needs a compatible device capable of reaching those speeds and is based on internal testing. That is a disclosure, not a dodge, and reaching 190 was going to take a faster host than any reader I owned. So I bought one.

On a SanDisk QuickFlow the same card reads 193.3 MB/s, across three passes of 193.39, 193.26 and 193.19, which is 101.7% of the printed claim. The 180.5 was never this card: two unrelated cards land within 0.24% of it in the same ProGrade slot and nothing has ever gone faster there, so about 181 MB/s is that reader's own microSD ceiling. The row above now reads BEATEN. The reasoning behind the original NOT SETTLED verdict was right, and that is the part worth keeping: a figure that flat, across four orders of magnitude of transfer size, was the transport rather than the media, and calling it a fail would have blamed the card for my hardware. The rule it produced is that an instrument has to be able to exceed a claim before it can test it. The same question is still open on the Samsung PRO Ultimate, which is a full-size SD card the QuickFlow cannot take, and it is why every read figure on this site is attributed to a reader.

Five minutes of committed writing, and then the whole card

The five-minute sustained leg settled at 129.0 MB/s with a worst ten-second window of 127.6 MB/s. Against the V90 floor of 90 MB/s that is 42% of headroom on a card whose package prints V30. The trace never dropped below 114.3 MB/s across all 300 samples.

The five-minute number is the one that sets the class, but the capacity leg is the more interesting test, because it wrote all 511.8 GB in one pass. Across 3,980 one-second samples the fill averaged 128.6 MB/s, the slowest single sample was 111.2 MB/s, and not one sample fell below 100. Split into quarters, the card wrote at 128.78, 128.67, 128.64 and 128.18 MB/s. It gives up half a percent between the first quarter of the card and the last.

One caveat worth stating plainly, because it is easy to over-read: the capacity fill runs after the sustained test and the settle pre-pass, so the card arrives at it already churned and already at its media rate. A flat fill curve there is the normal result on this bench, not a special property of this card. What is not normal is the level it holds that curve at, and the fact that its floor never drops into double digits.

The A2 badge and the number behind it

This is where the card is weakest, and it is worth being precise about what was and was not measured. The A2 specification requires 4,000 read and 2,000 write IOPS, but it reaches those figures using command queuing, which USB card readers do not implement. No USB bench can test A2, including this one. The tool marked the row not testable on its own.

What can be measured is random 4K at queue depth 1, and there the Extreme returned 1,796 read IOPS and 745 write IOPS. It clears the A1 write floor of 500, but only by 1.49 times. Set against the other premium 512 GB cards on the same reader with the same test:

The Extreme wins sequential read by roughly 11 MB/s and loses random write by a factor of two and a half. That is not a defect, it is a design: this card is tuned for streaming large files, which is exactly what SanDisk sells the Extreme line for. It does mean the A2 badge sets an expectation at queue depth 1 that the card does not meet relative to its peers, and if small-file work is your use case, one of the Samsungs is the better buy.

The good news is that it does not degrade under churn. Across three consecutive 20-second random write passes it went 747, 743 and 744 IOPS, a fade of 0.3%. The SanDisk High Endurance control fades 2.1% on the same test, and the unbranded reference card in the panel above collapses 71%.

511.80 GB of a 512 GB label, verified byte for byte

Every CKBench review fills the card completely with seeded, dedup-proof data and reads all of it back. This one wrote 477 files totalling 511.80 GB, then verified every one of them: zero mismatched bytes, zero read errors. At 99.96% of label, the 0.2 GB shortfall is the normal decimal-gigabyte gap that SanDisk states on the package itself.

A note on the read-back timing, because two different numbers are defensible and only one of them describes the card. The verify pass took 52 minutes 23 seconds of wall time but only 47 minutes 15 seconds of read I/O. The five-minute difference is the verifier generating and comparing its expected data, which is work the harness does, not the card. The card's own read rate through that pass was 180.5 MB/s. Any figure derived from the wall clock, 162.8 MB/s, understates it by ten percent, so this review quotes the I/O figure and says why.

The hottest card we have ever measured, and it did not care

At the end of the five-minute sustained write, with the fan off as the protocol requires, the card surface read 172.3 °F against a 76 °F ambient. That is a 96.3 °F rise, and it makes this the hottest conventional card in the corpus, taking the record from the Samsung PRO Plus at 169.6 °F. Only the Samsung P9 Express is hotter, and that card runs on a PCIe bus in a different reader.

The second reading in that frame matters. A card that disappears into a reader can only be read through the slot aperture, which under-reports by 10 to 14 °F, so those figures are published as floors. A microSD in this reader's dedicated slot leaves real card body exposed, so this is a genuine card-surface measurement and it is directly comparable to every other microSD figure on the site.

None of that heat cost anything. The sustained curve is flat across all four quarters, the worst ten-second window lands at 2:02 rather than late in the run, and the full-card fill afterwards held its level to the last byte. The card gets hot and keeps working. Worth knowing if you plan to run it hard inside a sealed enclosure with no airflow, where the surrounding air will be a good deal warmer than 76 °F.

Every headline number from the test

How it scores against the rubric

SanDisk Extreme 512GB: common questions