CrystalDiskMark can show whether an SSD reaches the performance expected from its slot and workload. The result is only as good as the setup. Background downloads, a small test file, an empty dynamic cache or the wrong target drive can make a healthy SSD look strange.

Quick Answer

Confirm the drive model and free space, close background work and let temperatures settle. Run CrystalDiskMark's standard profile first with its normal test count and a large enough test size to reduce cache noise. Read sequential and random results, then compare them with a review that used the same capacity, interface and settings.

Use Evetech's SSD catalogue and popular drive shortlist when the evidence supports an upgrade. Box speed is often a best-case sequential result, not the expected number in every row.

Prepare a Fair Test

Back up important files before diagnosing a drive with errors. Check SMART health, firmware, current temperature, free capacity and interface link. Verify that Windows is not downloading an update, indexing a new library or scanning the target.

Connect a laptop to mains power and select a normal performance mode. Keep the same power plan for comparisons. Let a recently copied SSD sit until its controller has finished obvious background work.

Choose the target by drive letter and model. A mistaken benchmark on the system disk, USB backup or RAM disk produces a convincing set of numbers for the wrong device.

Choose the Profile

Start with CrystalDiskMark's default profile. It gives a repeatable mix of sequential and random queues without inventing a custom workload. The NVMe SSD profile can expose deeper queue behaviour, though it can be less representative of a lightly loaded desktop.

Use the same version, profile, test size and run count when comparing changes. A 1GiB file may fit inside the drive's SLC cache. A larger size gives a better view of consistency but still may not exhaust a large dynamic cache.

Do not run the All test in a loop for entertainment. Benchmarks write data and heat the controller. A few controlled passes are enough for diagnosis.

Read Sequential Results

Sequential Q8T1 or a similar high-queue test measures large contiguous transfers with several requests in flight. It is the row most likely to approach the advertised peak on a fast NVMe drive. Large video copies and scratch work can care about it.

Sequential Q1T1 uses a lighter queue and may sit lower. That is expected. A single everyday copy cannot always feed the controller like a deep synthetic queue.

If every sequential result lands near half or a quarter of expectation, check PCIe generation and lane width under load. A Gen4 SSD in a Gen3 x4 slot is capped by the link. A drive negotiated at x2 loses more room.

Read Random Results

Random 4KiB tests access small blocks spread around the test file. Queue-one random latency and throughput often relate more closely to desktop launches and small-file work than a deep-queue headline.

Do not convert one random score into a promise of game FPS. Application caches, CPU work, file-system state and decompression share the result. Use it to compare the storage path, then measure the application.

IOPS, MB/s and latency are different views. Keep the display unit consistent before comparing screenshots. A result in IOPS can look much larger while representing the same work.

Control Heat and Cache

Watch the SSD sensor through a separate light monitor or the maker utility. If later passes slow as temperature reaches the documented throttle point, inspect the heatsink and airflow. Keep the monitor's polling modest so it does not interfere.

A fast first pass followed by a slower large-file result can show SLC-cache exhaustion. CrystalDiskMark's short default run is not a full-drive sustained test. Use a trusted review's long-write graph rather than filling the SSD yourself.

Free space changes dynamic cache size. Record how full the drive is and compare at a similar state. A nearly empty review sample and an 85 per cent full system disk are different conditions.

Diagnose a Low Result

Check these in order:

  • PCIe generation and lane width, or the negotiated USB rate for an external drive.
  • Temperature, free capacity and active background work.
  • Firmware, storage driver and motherboard lane-sharing rules.
  • Test profile, file size and whether the result came from the exact capacity.

Run a second pass after correcting one item. Avoid changing BIOS, driver, firmware and slot together because you will not know which change mattered.

If the drive reports media errors, disappears or corrupts data, stop benchmarking and back up. Speed diagnosis comes after data safety.

Compare With the Right Reference

Find an independent test of the same model and capacity. Match CrystalDiskMark version and profile where possible. NAND revisions can change during a product's life, so note the full model and firmware.

Expect normal variation from temperature, CPU and system activity. A result a few per cent below a published chart is different from a drive running at half its negotiated-link potential.

Record the final screenshot with drive fill, slot, firmware and temperature. Those details turn a benchmark image into useful evidence for future troubleshooting.

FAQ

Should every CrystalDiskMark row match the box speed?

No. The headline usually reflects a best-case sequential workload, while random and light-queue tests measure different access patterns.

Is the NVMe profile always the best test?

Use it for NVMe queue behaviour, but start with a standard repeatable profile and choose settings around the workload you need to diagnose.

Can CrystalDiskMark prove an SSD is healthy?

No. Pair performance with SMART data, error logs, file verification and stable detection.

Benchmark the exact SSD under a recorded profile, read sequential and random rows in context and correct one link, heat or background limit before retesting.