V90 in a reader, V60 in a camera

Why a 512 GB UHS-I card was worth a full afternoon

Every other card on this bench so far has been 128 GB or smaller, which makes a complete capacity fill a coffee break. This one took an hour and five minutes to fill and another fifty five to read back, and it was worth it, because it is the first card here whose printed speed cannot be reached over the standard interface at all. Samsung prints 200 MB/s read on the face. The UHS-I bus tops out near 104. Something has to give, and working out exactly what turned into the most interesting result of the day. It joins CKBench, the running index of every card on the bench, alongside the ProGrade and Lexar UHS-II V60 pair and the Kioxia Exceria G2 128GB.

One unit, bought at retail, tested on 2026-08-04. Every figure below comes from a run whose file name is printed under the chart it produced.

What the card claims, and what it did

This card face is unusually complete. Most cards print one speed and leave you to guess its direction; this one prints both, each labelled, along with SDXC, U3, V30 and the UHS-I mark. Everything below is sourced per row, because a number stamped on the card and a number on a web page are two different levels of evidence.

That warranty carries a condition worth reading before you buy. Samsung excludes write-intensive use, naming dashcam, CCTV and surveillance recording, from the PRO Ultimate warranty. It is a warranty term rather than a measurement, and this bench cannot test it either way, but it tells you what the card was built for.

The video speed class changes with what you plug it into

A video speed class is a floor, not an average. V30 promises that the worst ten seconds of a sustained write will never drop below 30 MB/s. That is the number a camera cares about, because one dip under the floor is a dropped frame or a stopped clip.

I ran the same five minute sustained write on this one card through two readers, back to back on the same evening. Through the ProGrade PGM0.5, which speaks the fast mode this card supports, the worst ten second window was 133.1 MB/s. That is a V90 result. Through a plain UHS-I reader, the worst ten second window was 88.7 MB/s. That is a V60 result, and it misses V90 by 1.29 MB/s.

Both numbers came from the same card, an hour apart, running the same test. Neither is wrong. But a camera does not have a ProGrade reader inside it: a camera gives the card a plain UHS-I bus, so 88.7 is the figure that predicts what happens while you are shooting, and 133.1 is the figure that predicts what happens while you are offloading. Quoting the 133.1 without naming the reader would tell a camera buyer something false, and you will see that number quoted.

The honest headline is the boring one: whichever host it gets, this card sustains at least double the V30 it is sold as. On the camera bus it is a V60 card wearing a V30 badge. On a fast reader it is a V90 card wearing a V30 badge.

What Samsung actually promised, and why 200 MB/s is not a simple number

Samsung's own product page qualifies the headline figures: up to 200 and 130 MB/s when using Samsung USB readers. I do not own one, so I could not put the claim in the conditions it was made under, and this review does not pretend otherwise.

There is a deeper reason the read figure resists a simple pass or fail. The UHS-I interface, in its fastest standard mode, tops out around 104 MB/s. A 200 MB/s read is therefore not something UHS-I can do at all. It requires a proprietary extension, generally called DDR200, that runs the bus faster than the specification allows, and the reader has to speak the same extension or none of it happens. That is not fine print. It is the architecture.

So when this card reads 169.8 MB/s on my reader, there are three genuinely different explanations and they must not be collapsed into one: the card cannot do 200, or my reader cannot go faster, or my reader speaks a different extension than Samsung's. My own analyzer picked the second one without being asked, flagging 169.8 as within 5 percent of the DDR200 ceiling of roughly 170 MB/s. That makes 169.8 a floor for the card and a ceiling for the reader, and it is why the claim row in the table above says not settled on this hardware rather than passed or failed.

What the two readers do settle is how much of this card's speed belongs to the extension rather than the card. Same card, same evening: the fast reader is worth 76 percent more read speed and 54 percent more write speed than the plain one. That is the largest host effect I have measured on any single card, and I went into what the reader alone does in the ProGrade PGM0.5 reader review.

The write claim needs none of this. Samsung says 130 MB/s. On non-Samsung hardware, with every byte committed and confirmed, the card averaged 137.3 across three passes and never dipped below 137.2. It beat a claim it made about someone else's reader while running on mine.

Five minutes flat out, then 512 GB without a knee

Plenty of cards open fast and collapse once a small quick buffer fills. That collapse, the knee, is the most common way a card that benchmarks well fails in a camera, and it only appears if you keep writing.

This one has no knee anywhere. The five minute sustained write started at 134.4 MB/s and finished at 134.3, and the analyzer classed the whole curve as stable with no cache cliff to find. Then the capacity test wrote to the card until it was full, an hour and five minutes of continuous writing, and speed measured at the first, middle and last tenth of the card came out at 130.2, 131.5 and 130.2 MB/s. That is a spread of 1.0 percent across 512 GB. Reading back, the same three positions gave 153.0, 154.8 and 153.9, a spread of 1.17 percent.

Nor did it ever pause. Across 38,490 individual one megabyte writes on the fast reader, not a single one took longer than half a second, and the longest of all took 38 milliseconds. On the plain UHS-I reader, 25,457 blocks, longest 41 milliseconds, again nothing over half a second. For recording, that steadiness is worth more than any peak number on a box.

One caveat on recovery. The card returned to full speed within 60 seconds of being handed a churned workload, but 60 seconds is the earliest my analyzer is allowed to declare stability, so the honest form of that result is 60 seconds or less, not 60 seconds. A dedicated follow-up run would pin it down.

Is it really 512 GB? Every byte, written and read back

The capacity test fills the card with uniquely seeded data so a controller cannot fake it by recognizing a repeated pattern, then reads every byte back and compares. It is slow, and it is the only test that catches a fake.

478 test files, 512.64 GB written, 512.64 GB verified, zero mismatched bytes, zero read errors. Against a 512 GB label that is 100.13 percent, which ties the SanDisk High Endurance for the highest figure on this bench. For contrast, the same test has returned 96.7 percent on a legitimate card and 26.8 percent on a counterfeit I took apart earlier this year.

If your computer shows about 477 GB rather than 512, nothing is missing. Card capacity is counted in decimal, where a gigabyte is a billion bytes, and Windows displays binary, where a gigabyte is 1,073,741,824 bytes. The same storage, two different rulers.

Two honest footnotes. The card was not blank when the fill started, so verification covered its free space: two pre-existing files totalling 88 bytes were left alone. And the read-back is where this card quietly does its best work for a photographer. Reading 512.64 GB back took 55 minutes 20 seconds, an average of 154.4 MB/s across the whole card. That is the number you feel offloading a full card, and it is a different measurement from the 169.8 MB/s short-file figure further up. Both are real. I am not going to invent a mechanism to reconcile them.

How it behaves under scattered writes

Sequential writing is the easy case. The hard case is random 4K: thousands of tiny writes landing all over the card, which is what filesystem updates, event saves and overwrites actually look like. Cards that are going to misbehave tend to misbehave here first, and the tell is that each pass comes out slower than the last as the card's internal bookkeeping falls behind.

Three 20 second passes, back to back, no rest: 1,826, 1,819 and 1,821 write IOPS. IOPS is just operations per second, so that is a card doing the same amount of work in pass three as in pass one. The drop from first to last is 0.25 percent, which is the steadiest result of any card on this bench, ahead of the 0.50 percent from the ProGrade V60 and a long way ahead of the 5.23 percent from the Lexar.

Three passes is an indicator, not an endurance test, and I want to be clear that it is not one. Nobody publishes an endurance rating for this card, Samsung included, and I have no way to measure one. What three passes can tell you is whether a card falls over the moment it is asked to work untidily, and this one does not notice.

How hot it gets, and which number I trust

Ejected the moment a sustained write ended, with the room at 76 °F, the card body measured 144.4 °F. That places it fourth of six cards measured this way, between the ProGrade V60 at 145.1 °F and the Kioxia at 143.3 °F, and a long way under the VIOFO High Endurance at 160.8 °F.

Getting a card temperature honestly is harder than it sounds, and this card is the reason I now write the method down beside the number. A full-size SD card disappears almost entirely into the fast reader, so with the card in place the only hot spot a thermal camera can see is through the slot opening, and that read 130.0 °F. It is a real reading of a real thing, and it is not the card: it is the hottest surface visible through a gap, and it runs 10 to 14 °F low. A third reading, taken with the card sitting proud of a shallower reader, came in at 150.0 °F while the card was doing a third less work. Three numbers, one card, all correct, none of them interchangeable. Only the ejected card-body reading goes on the chart below.

What matters more than any of them is that the heat cost nothing. Through the five minute fan-off run the card started at 134.4 MB/s and ended at 134.3, and its worst ten second window, which landed at second 286 near the very end of the run, was 133.1. A card that was throttling would show that window collapsing as it heated. This one gave up one percent, and one percent is not throttling.

How it scores on the five dimensions

The bottom line on the PRO Ultimate 512GB

Strip away the headline and this is simply a very good card. It beat the write speed on its own face, it delivered slightly more capacity than it promised across half a terabyte, it never developed a cache cliff, it never stalled, and under scattered writes it is the steadiest thing I have measured. Five hundred and twelve gigabytes that behave identically from the first block to the last is not a small achievement.

The headline is where it gets complicated, and the complication is not really the card's fault. Samsung printed 200 MB/s on a card whose interface cannot do 200 MB/s, then footnoted that the number needs Samsung's own reader. Both halves of that are true, and together they produce a specification that most buyers will never see met and cannot easily find out why. What I can tell you is that the card was never the bottleneck in any test I ran.

My advice is to ignore the 200 entirely and buy this for the two numbers that survive contact with real hardware: it writes faster than it says it does, and it holds a video class at least double its badge no matter what you plug it into. If you shoot on a UHS-I body, this is a great deal more card than the badge suggests. If you shoot on a UHS-II body, a real UHS-II card will beat it by a full video class, and that gap is the only reason to spend the extra money. I put that decision side by side in do you need a UHS-II SD card, using this card as the UHS-I control.

Disclosure: I bought this card myself, at retail, and Samsung had no involvement in the testing and did not see this review before it published. 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.