SD Card Speed Classes, Explained: What Actually Affects Load Times

Specs are marketing, and nowhere is the gap between the number on the package and the number that matters wider than on a microSD card. "Up to 170 MB/s" sells cards. It also describes the one metric emulation barely uses. This guide is the translation layer between the logos printed on a card and the load times you will actually feel.

The four rating systems, untangled

A microSD card can carry up to four different speed marks, and they measure different things:

• Speed Class (C2/C4/C6/C10): the old guard — a minimum sustained sequential write floor (Class 10 = 10 MB/s). Effectively a baseline every modern card clears.
• UHS Speed Class (U1/U3): a minimum sustained sequential write of 10 or 30 MB/s. U3 is the one to want.
• Video Speed Class (V6/V10/V30/V60/V90): also a sustained sequential write floor, aimed at video capture. V30 = 30 MB/s and overlaps U3 in practice.
• Application Performance Class (A1/A2): the mark that matters most for emulation, because it is the only one that specifies random read/write IOPS rather than sequential throughput.

Write those out and a pattern appears: three of the four systems describe sequential writing, which a handheld does rarely. The workload that defines emulation — opening a frontend, loading save states, streaming assets from many small files — is random reads. That is the A-class column.

Why A2 (random IOPS), not the big MB/s number

The A1 spec guarantees a minimum of 1,500 random read IOPS and 500 random write IOPS. A2 raises that to 4,000 read and 2,000 write IOPS. Those govern how fast a save state resumes or a large library's menu populates — small, scattered reads, not one long sequential slurp.

The "170 MB/s" on the front is peak sequential read, and a SNES or PS1 game never asks for 170 MB/s of continuous data; it asks for thousands of tiny reads. A card that wins the sequential beauty contest but has weak random performance can feel slower in a menu than a "slower" card with a strong A2 rating. This is the most common spec-sheet trap in handheld storage, and it is exactly why marketing leads with the number that flatters.

Where it actually shows up — and where it does not

Be honest about the size of the effect, because that is the hype-immune part:

• Menu and library scrolling, box-art loading: A2 helps, sometimes visibly, on large libraries.
• Save-state load and resume: random reads; a good A2 card shaves the wait.
• First-time launch and shader compilation on PC handhelds: partly storage-bound; faster random reads help.
• In-game performance once a title is loaded: for retro systems, almost entirely RAM- and CPU-bound. The card is not your framerate.

The card affects waiting, not the chipset's ceiling. Do not expect an A2 card to fix stutter that is really a silicon limit — that is what the Emulation Ceiling Finder is for. Both matter; they are different problems.

Quick reference

The honest buying rule

Buy a reputable brand's A2, U3/V30 card in the capacity you need, from a seller you trust, and stop optimizing. Counterfeits are the real hazard in this category — a fake "512GB" card that is actually 64GB will corrupt a library the day it fills. The premium for a genuine name-brand card over a suspicious bargain is a few dollars; the cost of the bargain is your entire save history. Prices move week to week, so treat any figure as quoted at analysis (August 2026) and re-check before buying.

Capacity, endurance, and formatting

Capacity is a comfort decision, not a performance one — a 256GB and a 512GB card of the same model read at the same speed. Buy for the library you actually own plus headroom; you can always add a second card later. Endurance ratings (TBW) matter far more for dashcams than for a handheld that mostly reads, so do not overpay for high-endurance lines unless you also record a lot of gameplay. Format in the device when possible; most retro handhelds expect exFAT, and formatting on the handheld itself avoids allocation-size mismatches. The first-setup guide covers the format-in-device step in sequence.

For specific models and cost-per-gigabyte, see Best microSD Cards for Handhelds; for how to split a library between the card and internal storage, see Handheld Storage Management. The verdict is boring and correct: A2 over big-MB/s, name brand over bargain, right capacity over biggest number.

Test a new card before you trust it

The single habit that saves libraries: when a card arrives, verify its real capacity before you fill it. Counterfeit cards report a false size to the operating system, then silently discard or overwrite data once you pass the true limit — and you find out only when a save state you needed is gone. A full-capacity write-and-verify pass (write known data across the whole card, then read it back and confirm it matches) is the standard check; free capacity-verification utilities exist for every desktop OS. Run it once on arrival, before you have anything to lose. A genuine name-brand A2 card passes; a relabeled fake fails at the point where its real silicon ran out.

Pair that with a boring backup routine — copy your saves and states off the card on a schedule — and the speed class becomes what it should be: a performance decision, not a data-loss risk. Get the card right once and you stop thinking about storage, which is the whole goal. Faster menus are nice; a library you never lose is the point.