Why the storage in your device is slower than the number on the box

This is less a set of instructions about flash storage performance than an argument, and it is worth saying so at the start.

Flash cells store voltage levels, not bits

A flash cell holds a charge, and the controller distinguishes between voltage ranges to decide what was stored. Storing one bit needs two distinguishable levels; storing three bits needs eight, and storing four needs sixteen within the same physical range.

More levels means finer distinctions, slower and more careful writing, and less tolerance of the charge drifting over time. This is the entire reason cheaper high-capacity storage writes slowly once it is working hard.

The cache is what you measured

Drives reserve a portion of their cells operating in fast single-bit mode and write incoming data there first. Advertised sequential speeds are measured while that cache is empty, which is why a short benchmark looks excellent.

Once the cache fills during a large sustained write, speed drops to the native rate of the denser cells, sometimes by a large factor. Copying a very large file and watching the transfer rate partway through reveals this within seconds.

Erasing is the expensive operation

Flash can be written in small pages but only erased in much larger blocks, so modifying data means writing elsewhere and reclaiming the old block later. That reclamation, called garbage collection, requires somewhere to put the still-valid data from a block being erased.

A nearly full drive has little room to work with, so it performs more internal copying for every write you request. This is why performance falls sharply near capacity and recovers when space is freed.

Wear is real and usually not your problem

Each erase cycle damages the insulating layer slightly, and cells have a finite endurance measured in cycles. Controllers spread writes across the whole device so no block wears out early, which is wear levelling. For ordinary use the drive is very likely to be replaced for capacity reasons long before endurance becomes relevant.

Sustained heavy write workloads such as video capture or database logging are the exception where endurance ratings genuinely matter.

Interface, protocol and memory are three limits

The connection, the command protocol and the flash itself each impose a ceiling, and the lowest one governs. A modern protocol with queueing improves how many operations can be in flight, which matters far more for everyday responsiveness than sequential speed. Random access of small pieces of data is what an operating system actually does most of the time.

Mechanically, that is why a drive with modest sequential figures and good random performance feels faster in use than the reverse.

Buying and using sensibly

Look for sustained write figures and random operation figures rather than the headline sequential number. Larger capacities of the same model are usually faster, because more chips can be written in parallel and the cache is larger.

In practice, keeping meaningful free space is the single most effective thing a user can do for storage performance. Removable cards have their own speed classes describing minimum sustained rates, which is the figure that matters for video recording.

The takeaway

You benchmarked the cache. The real speed is whatever happens after it fills.

The constraint is almost always physical, and marketing rarely mentions which one.