Why Disk Speed Test Results Vary Between Runs

A benchmark is a health reading, not a spec sheet

A speed test reports what the drive does right now, in the state the system happens to be in. It is closer to a blood-pressure reading than to the number printed on the box. Two runs on the same drive, minutes apart, can differ by 10 to 15% without anything being wrong.

So do not chase one perfect figure. Watch the range and the direction: where results usually land, and whether they keep slipping over weeks.

Five things that move the result

Nearly every "why is it slower this time" traces back to one of these:

• The SLC cache. Many SSDs only hit their advertised write speed while the cache has room. Run a write test right after copying a large file and the cache has not been reclaimed yet, so the score drops. A few idle minutes restore it.
• Heat. An SSD controller nearing its limit throttles itself, and a hot Mac downclocks everything. DiskBee runs as a macOS user-space file system, so its throughput follows system-wide downclocking fairly closely, and numbers taken on a hot machine look worse.
• Background activity. Spotlight indexing through mds_stores reads the disk directly, and Time Machine plus iCloud sync write to it. All of them compete with the test for bandwidth.
• Power and cabling. macOS limits performance on battery, so plugged-in is the honest state. A worn cable or an extra hub can also negotiate the link down a mode.
• Platter geometry. On a mechanical drive the outer tracks move data nearly twice as fast as the inner ones, so where the test file lands changes the score. That is physics, not a fault.

Getting a usable number

• Connect the power adapter first. Battery mode caps performance.
• Leave the drive and the Mac idle for a few minutes so caches clear and temperatures settle.
• Keep backups, large copies and sync jobs away while the test runs.
• Run it two or three times and read the cluster. One outlier means little.

When a change is worth worrying about

Normal fluctuation needs no action. Three patterns do:

• Results keep falling across days or weeks, clearly outside the usual 10 to 15% band;
• write speed stays below a third of read speed, which points to cache exhaustion, throttling or a drive in decline;
• slow numbers arrive together with disconnects, clicking or I/O errors, in which case stop trusting the drive with important data and run a health check.

How the built-in benchmark measures

The test follows the CrystalDiskMark 8 pattern: four profiles covering sequential and random 4K, read and write measured separately. Each run opens its test file with F_NOCACHE so the macOS cache stays out of the way, and discards a warmup pass before counting. The result describes the drive rather than the memory cache in front of it.

Traffic also goes through the app's own NTFS driver channel, which means the figure matches the speed you see when copying real files instead of a synthetic path that nothing else uses.

The questions that come up most

Some drift between runs is expected, up to roughly 10 to 15% on unchanged hardware, and only a sustained shift outside that band is worth chasing. A second run that comes in faster or slower usually reflects cache and temperature state: the SLC cache may have just been reclaimed, or may still be recovering from an earlier copy, which is why a few idle minutes before testing change the answer.

A single low number is not evidence of a failing drive. Cable, power and background load can each produce one. What matters is direction across repeated runs.

Reading the trend rather than the run

A benchmark is a health reading of the current moment, not a label for the drive. A 10 to 15% swing is normal. Test with the power adapter connected, after a short idle, two or three times, then judge the cluster. Two situations deserve action: a persistent decline, or write speeds stuck below a third of read speed.