A B2B guide to SD Application Performance Class A1 and A2: what the random-read/write ratings mean, how they differ from Video Speed Class, which devices benefit, and why A2 is wasted on sequential video.
TL;DR
The A1 and A2 marks on a microSD card are an Application Performance Class, and they measure random read and write performance rather than sequential video speed. A1 guarantees at least 1,500 random read and 500 random write IOPS; A2 guarantees at least 4,000 read and 2,000 write IOPS. This matters for devices that boot an operating system or run apps from the card — single-board computers, set-top boxes, handheld consoles — and it is irrelevant for sequential video recording. WINTEL specifies TF cards from 4GB to 256GB with A1 or A2 where your host needs it. Tell us how the board uses the card and we will match the class rather than oversell it.
Sequential speed and random speed are different problems
A video camera writes one long stream. A computer running from a card fires thousands of tiny reads and writes, scattered across the storage. These are different workloads, and they are measured by different numbers.
Video Speed Class (V30, V60, V90) measures sustained sequential write in MB/s. Application Performance Class (A1, A2) measures random input/output operations per second, or IOPS. A card can be excellent at one and only average at the other, because the controller optimises for the pattern it was graded against.
That split is the single most common source of a mis-specified card.
What A1 and A2 actually guarantee
The SD Association defines both classes as a floor on random performance:
- A1 — at least 1,500 random read IOPS and 500 random write IOPS.
- A2 — at least 4,000 random read IOPS and 2,000 random write IOPS.
A2 also assumes command queueing that lets the host schedule more operations at once. The headline is the random-IOPS floor, not a sequential speed.
One detail buyers miss: the floor is a minimum, not a typical. A card graded A2 may perform well above 4,000 read IOPS, but the class mark only promises the worst case you will accept. Specifying the class sets the floor your design can rely on, which is the number that matters when you lock a bill of materials. If a supplier cannot tell you the floor they guarantee, they are not really selling you a class — they are selling you a hope.
What random IOPS feels like in practice
Random IOPS is the difference between a board that feels responsive and one that feels broken. A single-board computer booting Linux scatters reads across thousands of small system files before the desktop appears. A set-top box launching an app does the same with its own bundle of small objects.
At 500 write IOPS the system still works, but app installs stutter and every log write adds latency the user can feel. At 2,000 write IOPS those same operations complete in a fraction of the time, and the device stops fighting its own storage. The user never sees the IOPS number on a spec sheet; they feel the stall, and the stall becomes a return.
This is why A2 is a user-experience specification, not a benchmark trophy for the procurement file.
Where A2 earns its price
A2 pays off on a device that treats the card as a system disk rather than a data bucket. A single-board computer booting Linux, a set-top box running apps, or a handheld that loads software from the card all scatter small files and small blocks across the media. That is the random-read/write pattern A2 is built to serve.
For those devices, a slow random-IOPS card shows up as a sluggish interface, long app launches and occasional freezes — symptoms a sequential speed test never reveals. If the board runs anything beyond a bare boot loader, A2 is worth specifying, and A1 is the minimum you should accept rather than the ceiling you should celebrate.
Where A2 is wasted
A2 is wasted on sequential video. A dashcam, an action camera or a drone records one continuous stream, and what it needs is a Video Speed Class such as V30 or V60. The random-IOPS headroom of A2 is never exercised, so the extra cost buys nothing the device can use.
Do not let a supplier upsell A2 into a video recorder. The right question for video is the sustained sequential write speed against your bitrate, not the random-IOPS class. If a quote leads with A2 on a dashcam, that is a sign the supplier is matching a label instead of your workload.
A card graded for random IOPS is not a faster video card. It is a different tool.
Host support decides whether A2 delivers
An A2 card works as storage in any microSD host. But the random-I/O gains depend on the host supporting the command queueing that A2 relies on, and not every board does. An older reader or a simple recorder treats an A2 card as a fast A1-class card and never reaches the higher floor.
So A2 is a system specification, not a card-only property. Confirm your host supports it before you pay the premium, or the class mark is just printing on the label. A card that cannot reach its floor is no better than the class below it, and you have paid for capacity the hardware cannot use.
A worked comparison: A1, A2 and V30 side by side
| Class | What it measures | Floor it guarantees | The workload it fits |
|---|---|---|---|
| A1 | Random IOPS | 1,500 read / 500 write | Light OS boot, casual app use |
| A2 | Random IOPS | 4,000 read / 2,000 write | Heavy OS, app stores, frequent writes |
| V30 | Sequential video write | 30 MB/s sustained | Dashcams, action cameras, drones |
Notice that the three numbers describe three different things. A V30 card and an A2 card can sit in the same product and both be correct, because one serves the video stream and the other serves the system. The mistake is reaching for one where the other is needed, then wondering why the device behaves differently than the benchmark implied.
Sizing the card to the OS image
A board that boots from the card needs capacity for the OS image plus the working space the system uses after boot. A lightweight Linux image fits on 16GB or 32GB; a desktop-class image with updates and caches wants 64GB or more. Confirm the image size first, then add headroom, rather than rounding to a number and discovering the card fills on first boot.
The Application Performance Class and the capacity are independent choices. A 256GB card with no A-rating performs poorly as a system disk, and a 16GB A2 card may be too small to hold the image. Specify both against the board, not against a catalogue.
Leave room for writes the board will make on its own. Logs, caches and over-the-air updates all consume space a bare image never shows, and a card that runs out of room mid-update is a field failure, not a spec-line footnote.
Qualifying A1 or A2 in your own board
Do not trust the class mark on its own. Put the candidate card in the actual board, boot the actual image, and measure boot time and app launch under load, not on a bench reader. A reader reports a sequential number that has nothing to do with how the board will feel in use.
Two checks catch most mistakes. First, confirm the board enumerates the card on cold power every time, because an A2 card that fails to initialise on first boot is worse than an A1 card that never fails. Second, confirm the board actually issues the queueing commands A2 needs; without them you have paid for a floor the hardware cannot reach, and the class mark on the label is the only place the performance exists.
Specifying the class to your supplier
Tell your supplier how the device uses the card: boot only, boot plus apps, or pure sequential recording. WINTEL supplies TF cards from 4GB to 256GB and specifies A1 or A2 per project against your host and OS image size. For a maker or OEM board, we also supply semi-finished boards and UDP modules if you build your own enclosure.
Send the host interface and the OS image size, and we will propose the class that fits the workload rather than the most expensive one.
If you are specifying microSD for a board that boots or runs apps, send us the host interface and the OS image size. We will tell you whether A1 or A2 is the right class — and whether you are paying for performance the host cannot use.
Reach the team at cathy@go-wintel.com or open the request a quote form.
Sources
Frequently asked questions
What does A1 versus A2 actually measure?
The Application Performance Class measures random read and write performance, expressed in IOPS. A1 requires at least 1,500 random read and 500 random write IOPS; A2 requires at least 4,000 random read and 2,000 random write IOPS. These describe a mixed, small-block workload, which a video speed class does not.
Do I need A2 for video recording?
No. Video is a sequential write workload, and a Video Speed Class such as V30 or V60 describes that directly. A2 adds random-I/O performance that a dashcam or action camera never exercises, so paying for A2 there is wasted budget.
Which devices actually benefit from A2?
Single-board computers and set-top boxes that boot an OS or run apps from the card. Those workloads scatter thousands of small reads and writes, which is exactly what A2 targets. Tell us how your board uses the card and we will specify the class.
Can any host use an A2 card?
The card works as storage in any microSD host, but the A2 random-I/O gains depend on host support for the command queueing A2 uses. An old reader treats an A2 card as a fast A1-class card. Confirm host support before you pay for A2.
What capacities does WINTEL offer for A1 or A2 cards?
Our TF cards run from 4GB to 256GB. Application Performance Class is specified per project against your host and workload, so tell us the board and the OS image size and we will propose the right class.