Why File Copies Run at Half Your SSD's Speed (and When That's Normal)

The spec sheet measures a different job

The number on the box is sequential throughput at high queue depth: one large file, read and written back-to-back, with many commands in flight at once. Real copies are rarely that job. Moving a folder of documents, source code or photos means thousands of independent files, each one a separate read, a separate write, and a pile of bookkeeping in between — and that work is bound by latency, not by throughput. A drive that sustains 3,500 MB/s sequentially can still crawl at effective speeds of a few MB/s when the workload is many tiny files.

None of that means your drive is slow; it means the units are different. Latency, queue depth, file size distribution and metadata overhead decide how fast a copy feels, and the marketing number only covers one corner of that picture. Once you separate those factors, the mystery mostly dissolves.

• Sequential throughput — the box number — applies only to large single files moved back-to-back
• 4K random latency, not throughput, decides how fast thousands of small files move
• Every file pays a metadata tax: MFT record, permissions, timestamps, optional streams
• Real-time antivirus reads each file on the way out and often again on the way in
• A copy onto the same drive competes with itself for read and write headroom
• Budget QLC drives write at full speed only until their SLC cache runs dry

Where the seconds really go

Small files are the classic killer: 10,000 files of 50 KB each pay the per-file cost ten thousand times, and the drive spends more time on bookkeeping than on moving bytes. Explorer's copy dialog is also conservative — it walks the source tree, computes the destination, and reports progress per file, which adds visible pauses that a raw block copy would never have. Real-time antivirus doubles the work by scanning data as it leaves the source and again as it lands. Copy to the same physical drive and the headroom is spent twice, since every byte is a read plus a write on one medium.

The pattern holds outside the SSD too. A gigabit link tops out near 110–118 MB/s no matter how fast both ends are, and Wi-Fi delivers a fraction of that under real conditions. Cheap USB sticks pair slow flash with a filesystem hostile to small files, so a stick rated 'up to 150 MB/s' can average single digits on a photo folder. Sustained writes can also trip thermal throttling on compact M.2 drives, which slows the second half of a big copy.

What genuinely helps

For many small files, robocopy is the honest answer: its /MT switch copies several files in parallel and routinely turns a crawling Explorer copy into a brisk one — try /MT:16 to start. Keep about a tenth of the SSD free, because a nearly full drive loses write cache headroom and collapses first on exactly these workloads. Avoid stacking parallel copies onto the same destination, especially a USB stick: one deep queue beats three competing ones. And check the Performance tab in Task Manager during a slow copy — active time pinned at 100 percent with a short queue means you are latency-bound, not throughput-bound.

Before blaming hardware, measure it once with a benchmark tool: if sequential speed lands near the rated number and the slow copies involve small files or many recipients, the drive is fine and the workload is simply hard. Copy accelerators that stage everything through RAM promise more than they deliver and risk data on a crash. In most cases the fix is not a faster drive but a better-shaped copy command.

Questions and Answers

Read next