How do I test the real write speed of a USB drive without lab equipment?
Write a single file larger than the drive's cache, and time it. That is the whole method. Advertised speeds are usually burst figures measured over a few hundred megabytes, which a cache absorbs; a file of 4GB or more forces the drive to write to the actual memory, and the rate you see after the first 30 seconds is the number that predicts real use. Use a large video file or a zip archive rather than thousands of small files, because small-file transfers measure random access, not sequential speed. On Windows, copy the file and watch the transfer dialog settle; on macOS or Linux, use dd and read the reported rate.
Why the number on the box does not match what you see
Flash drives contain a small cache. When you start a write, data lands in the cache first, and the cache is much faster than the memory behind it. For the first few hundred megabytes the transfer looks excellent.
Once the cache fills, the drive has to write through to the memory, and the rate drops — often by half or more. If your test writes less data than the cache can hold, you measured the cache, not the drive.
This is why a datasheet can be truthful and still useless: it reports a real measurement of the wrong thing.
A method that actually measures the drive
The procedure is deliberately boring, and it works without any special software.
- Empty the drive, or at least ensure it has more free space than your test file.
- Create or obtain one file of 4GB or larger. A video file works well; thousands of photos do not.
- Copy it to the drive and watch the transfer rate. Ignore the first 20–30 seconds.
- Record the rate once it has been steady for at least a minute. That is your sustained write speed.
- Delete and repeat once. A large discrepancy between two runs usually means a thermal or cache effect worth understanding.
Reading the result honestly
There is no single "correct" sustained speed, because it depends on the controller, the memory, the capacity and the host port. What you are looking for is whether the drive is in the range you were promised, and whether it degrades badly as it fills.
Two patterns are worth acting on. If the rate collapses to a small fraction of the burst figure after the cache fills, the memory is much slower than the marketing implies. If the rate falls steadily over several minutes, the drive is heating up and throttling — see the heat question for what that means.
| What you observe | What it usually means |
|---|---|
| Fast start, sharp drop after 30s | Cache exhausted — normal, but tells you the real rate |
| Rate keeps falling over minutes | Thermal throttling in a small body |
| Rate far below any advertised figure | Memory grade lower than the quote implied |
| Transfer stalls entirely part-way | Possible capacity fraud — test capacity next |
Sources
- USB.org — USB performance and interface specifications
- Microsoft — storage and file system performance guidance
Figures quoted on this page are from the published reporting listed above. Where a number is an estimate rather than a published figure, it is described as approximate.
Frequently asked questions
Does the port matter when I test speed?
Yes, substantially. A USB 3.0 drive in a USB 2.0 port will be limited by the port, not the drive, and the ceiling is around 30–40MB/s regardless of how good the drive is. Confirm the port and the cable support the speed you are trying to measure, and avoid hubs during a test — they are a common source of artificial slowdowns.
Why do small files copy so slowly?
Because each small file is a separate operation with its own overhead, and the drive has to look up locations rather than stream data. This measures random access performance, which matters for some workloads but is not the same as the sequential write speed a datasheet quotes. Test with one large file for sequential speed, and with a folder of many small files only if that is your actual workload.
Should I test every drive in a batch, or a sample?
Test a sample of each batch for speed, because speed is a property of the design and the memory grade, which should be constant within a batch. A speed outlier in a sample is a signal to look at the whole carton. Full write-and-read-back capacity testing, by contrast, is worth doing more broadly, because capacity fraud is per-unit and silent.
Working on a specification?
Send us the product type, capacity, quantity and target date. We reply to business enquiries within 1 business day.
Request a quote