A fresh SSD that “feels slow,” or a drive you suspect is throttling, deserves a proper test rather than a guess. But naive benchmarking produces misleading numbers all the time—testing into the SLC cache, on a nearly full drive, or with the wrong queue depth can make a great SSD look bad or a mediocre one look heroic. Here is how to measure real performance you can trust.
What the numbers actually mean
Two figures matter for different reasons. Sequential speed (large contiguous transfers) is what the box advertises—think 3,500 MB/s for PCIe 3.0, 7,000+ MB/s for PCIe 4.0. Random 4K speed at low queue depth (QD1) is what you feel day to day: booting, launching games, opening apps. A drive with blistering sequential numbers but weak QD1 random reads will still feel sluggish in normal use.
| Metric | PCIe 3.0 NVMe | PCIe 4.0 NVMe | SATA SSD |
|---|---|---|---|
| Sequential read | ~3,000–3,500 MB/s | ~5,000–7,400 MB/s | ~550 MB/s |
| Sequential write | ~2,000–3,000 MB/s | ~4,000–6,500 MB/s | ~500 MB/s |
| Random 4K read (QD1) | ~40–60 MB/s | ~60–90 MB/s | ~25–40 MB/s |
| Random 4K write (QD1) | ~100–180 MB/s | ~150–250 MB/s | ~60–90 MB/s |
Set up a fair test
Use a standard tool such as CrystalDiskMark (Windows) or a cross-platform equivalent. Before you run anything: close background apps, disable any active virus scan on the target drive, and make sure the drive is not mid-firmware-update. Test the drive that hosts the OS only when idle, or you will measure Windows activity, not the SSD.
Pick a test file size of at least 8–16 GB. Small 1 GB tests fit entirely inside the drive’s fast SLC cache and report inflated write speeds a full drive can never sustain. A larger sample forces the controller past the cache so you see steady-state behavior.
Watch for the throttling trap
Two things distort write numbers most. First, SLC cache exhaustion: many drives write at full speed for the first 50–200 GB, then drop to 800–1,500 MB/s (or lower on QLC) once the cache fills. Second, thermal throttling: a bare Gen4/Gen5 drive can hit 70–80 C and cut speed within seconds. Log the drive temperature during a sustained write—if speeds collapse as temperature climbs, you need a heatsink, not a new SSD.
Interpret results honestly
Compare against the drive’s rated spec and its interface ceiling, not against a different class of drive. A SATA SSD will never beat 560 MB/s—that is the bus limit, not a fault. A Gen4 drive stuck near 3,500 MB/s in a Gen3 slot is running exactly as expected. And a fresh, mostly empty drive should hit rated sequential numbers; if it is 30–50% low, check the slot generation, the cable (for SATA), and firmware.
Frequently asked questions
Why does my write speed drop halfway through a big copy?
That is the SLC cache filling. Once the fast pseudo-cache is used up, writes fall back to the slower native NAND speed. It is normal; the fix is a drive with a larger cache or higher native write speed if you routinely move 100+ GB at once.
Should I benchmark a full drive or an empty one?
Both tell you something. An empty drive shows peak potential; a 70–80% full drive shows realistic everyday speed, since a nearly full SSD has a smaller usable cache and less room for wear leveling. Test near your normal fill level for the most useful number.
Bottom line
Trustworthy benchmarking means a large test file, a cool and idle drive, and comparing results to the right interface class. Separate the marketing sequential figure from the QD1 random speed that governs real responsiveness, and keep an eye on temperature and cache behavior. Do that and you will know whether a drive is genuinely underperforming or simply doing exactly what its class allows.
