XDA spent the week walking through the regrets of stuffing every M.2 slot on a consumer motherboard, and one number kept showing up in the comments: the second NVMe drive ran at half the speed of the first. Shared PCIe lanes, chipset bottlenecks, and a heatsink that turned out to be plastic were the usual culprits. Marketing specs assume an unloaded primary slot with a copper cold plate. Real builds rarely look like that. SSD benchmarking is how you find out which slot is actually fast, whether a drive is throttling after thirty seconds under load, and how random 4K reads compare to the sequential numbers on the box.
We tested 7 of the best apps for SSD benchmarking across Windows, macOS, and Linux. The yardstick was practical: do the defaults match what reviewers publish, can you push queue depths high enough to saturate a Gen 5 drive, and does the tool tell you anything useful once the cache is exhausted. Most of the strong options are free, and the few that cost money are inexpensive.
What to look for in an SSD benchmark app
A few criteria separate a tool that answers real questions from one that prints a pretty number:
- Sequential and random workloads. Sequential reads tell you whether a slot is wired for the full PCIe generation. Random 4K reads at low queue depth predict how the drive feels in daily use. You need both.
- Queue depth and thread control. Gen 4 and Gen 5 NVMe drives only hit their headline numbers at high queue depth across multiple threads. A tool that hides these knobs cannot saturate a modern drive.
- Thermal soak test. A short burst always looks great. The interesting question is what the drive does at three minutes of sustained writes, when the SLC cache is gone and the controller is warm.
- Comparable result database. CrystalDiskMark numbers are the lingua franca on tech forums. Tools that match those defaults make it easy to compare your drive to other people’s.
- Drive-specific feature awareness. Vendor utilities can flip on TurboWrite, magician modes, or rapid caches that change the numbers. Knowing what a tool measures matters.
- OS coverage. NVMe lives on Windows, macOS, and Linux. The best app for your build is the one that runs on the OS you boot.
Quick comparison
| App | Best for | Platforms | Free plan | Standout feature |
|---|---|---|---|---|
| CrystalDiskMark | Numbers everyone else publishes | Windows | Yes (open source) | Default profiles match how reviewers test |
| AS SSD Benchmark | A second opinion using incompressible data | Windows | Yes | Access-time test and copy-pattern checks |
| ATTO Disk Benchmark | Spotting transfer-size cliffs | Windows | Yes | Sweep across block sizes from 512 B to 64 MB |
| AmorphousDiskMark | A CrystalDiskMark workalike on Mac | macOS | Yes (Mac App Store) | Familiar UI tuned for APFS and external Thunderbolt |
| Blackmagic Disk Speed Test | Quick read on whether a drive can record video | macOS | Yes | Looping sustained test that surfaces thermal drops |
| KDiskMark | A CrystalDiskMark-style GUI on Linux | Linux | Yes (open source) | fio under the hood with a sane default UI |
| fio | The control everyone else’s tool is hiding | Windows, macOS, Linux | Yes (open source) | Scriptable workloads and reproducible test files |
The 7 best apps for SSD benchmarking on desktop
1. CrystalDiskMark — best for numbers everyone else publishes
CrystalDiskMark is the reference. The default profile runs sequential and random reads and writes at the queue depths most SSD reviews use, which means the numbers you see on screen line up with what tech outlets and forum posts report for the same drive. Version 9 keeps the familiar table view, adds finer control over test size and pass count, and exposes mix workloads for sites that test read/write blends. It is the tool to run first when you want a sanity check on a new build or a freshly mounted drive.
Where it falls short: Windows only. The Real World profile mimics typical desktop use but does not stress-test a drive long enough to surface thermal throttling on its own.
Pricing:
- Free: open source, no licence fee
- Paid: none
Platforms: Windows
Download: crystalmark.info
Bottom line: Pick CrystalDiskMark for SSD benchmarking if you want a number you can compare to the rest of the internet without footnotes.
2. AS SSD Benchmark — best for a second opinion using incompressible data
AS SSD Benchmark writes incompressible data, which is the point. Some controllers (notably older SandForce designs and a handful of DRAM-less drives) post inflated sequential numbers when the test pattern is easy to compress. AS SSD strips that advantage and gives you a truer floor. The access-time test and the copy-pattern benchmarks add useful signal beyond raw throughput, and the score at the end is a quick way to rank drives across a fleet.
Where it falls short: The project has not seen frequent updates, and the maximum test size is on the smaller side. Modern Gen 4 and Gen 5 drives can blow past the cache before AS SSD finishes its longest run.
Pricing:
- Free: no licence fee
- Paid: none
Platforms: Windows
Download: alex-is.de
Bottom line: Pick AS SSD Benchmark for SSD benchmarking when you want a check on whether a drive’s headline numbers came from compressible test patterns.
3. ATTO Disk Benchmark — best for spotting transfer-size cliffs
ATTO Disk Benchmark sweeps a read and a write across block sizes from 512 bytes up to 64 megabytes and graphs throughput at each step. The result tells you exactly where a drive starts hitting its stride and where it falls off. That shape matters: a drive that peaks at 1 MB blocks but lags at 4 KB will feel slow opening apps, even if its top-line sequential number is excellent. It is also the benchmark a lot of vendor marketing slides are built on, which makes it useful for cross-checking claims.
Where it falls short: No random workload, no queue-depth control to speak of, and Windows only. ATTO is a one-dimensional view by design.
Pricing:
- Free: no licence fee
- Paid: none
Platforms: Windows
Download: atto.com
Bottom line: Pick ATTO Disk Benchmark for SSD benchmarking when you want to see the throughput curve across block sizes rather than a single peak number.
4. AmorphousDiskMark — best CrystalDiskMark workalike on Mac
AmorphousDiskMark is the macOS answer to CrystalDiskMark. The layout, the default profiles, and the metrics line up close enough that you can hand the result to a Windows reviewer and they will know what to do with it. It runs cleanly on APFS, handles external Thunderbolt enclosures without surprises, and reports both MB/s and IOPS for each test row. For a Mac user comparing a new internal NVMe to an external T7 over USB-C, this is the fastest path to apples-to-apples numbers.
Where it falls short: Distribution is now Mac App Store only, which means a one-time charge in some regions and no portable build. The UI follows CrystalDiskMark closely, which is a strength when you want familiarity and a limitation when you want something more visual.
Pricing:
- Free: yes, on the Mac App Store
- Paid: optional one-time in-app upgrade in some regions
Platforms: macOS
Download: katsurashareware.com
Bottom line: Pick AmorphousDiskMark for SSD benchmarking on a Mac when you want results that translate directly to the Windows numbers you see in reviews.
5. Blackmagic Disk Speed Test — best quick read on video-capable speed
Blackmagic Disk Speed Test was built so editors could check whether a drive could keep up with a given video format, and that framing makes it unusually honest about sustained performance. The test loops, the dial keeps moving, and any thermal drop or cache exhaustion shows up as the numbers visibly fall over a few minutes. A grid at the bottom lights up the video resolutions and codecs the drive can handle, which is a useful translation layer if you do not want to read a number in megabytes per second.
Where it falls short: No random 4K workload, no queue-depth control, and limited reporting. It answers one question well rather than every question shallowly.
Pricing:
- Free: yes, on the Mac App Store
- Paid: none
Platforms: macOS (Windows version exists but is older and less maintained)
Download: blackmagicdesign.com
Bottom line: Pick Blackmagic Disk Speed Test for SSD benchmarking on a Mac when you want to watch what happens after the cache runs out, not just the first ten seconds.
6. KDiskMark — best CrystalDiskMark-style GUI on Linux
KDiskMark wraps fio in a Qt interface that mirrors CrystalDiskMark almost cell for cell. You pick a profile, point it at a mount, and the results table looks identical to what a Windows tester would produce. Under the hood it is running a respected benchmark engine with sensible defaults, which means the numbers are trustworthy and reproducible. AppImage builds run on most distributions without needing a package manager dance.
Where it falls short: Linux only. Some package repositories ship older versions, so prefer the AppImage or the project’s GitHub releases for current builds.
Pricing:
- Free: open source, no licence fee
- Paid: none
Platforms: Linux
Download: github.com/JonMagon/KDiskMark
Bottom line: Pick KDiskMark for SSD benchmarking on Linux when you want the comfort of the CrystalDiskMark layout without leaving the desktop you boot.
7. fio — best for the control everyone else’s tool is hiding
fio is the benchmark engine most of the other tools wrap. It runs on Windows, macOS, and Linux, takes a plain-text job file, and lets you describe a workload precisely: block size, queue depth, thread count, read/write mix, alignment, direct I/O, runtime, ramp time. The result is reproducible across machines and easy to script into a CI run that catches a drive that has quietly degraded. The same job file that runs on a Linux home server runs on a Windows desktop the next day.
Where it falls short: No GUI. The first job file you write involves reading documentation. Output is a wall of text by default, and turning it into a chart is on you.
Pricing:
- Free: open source, no licence fee
- Paid: none
Platforms: Windows, macOS, Linux
Download: github.com/axboe/fio
Bottom line: Pick fio for SSD benchmarking when you have a specific question and want a tool that will answer it the same way next month and on a different machine.
How to pick the right one
If you want a number you can compare to a reviewer’s chart, run CrystalDiskMark on Windows or AmorphousDiskMark on a Mac.
If you suspect a drive’s marketing numbers came from compressible data, cross-check with AS SSD Benchmark.
If you want to see how performance changes across transfer sizes, run ATTO Disk Benchmark.
If a drive feels fine in short bursts but seems to stall during long copies, leave Blackmagic Disk Speed Test running on a Mac and watch the dial drop.
If you boot Linux, KDiskMark is the closest thing to a familiar GUI, and it produces trustworthy numbers because it leans on fio.
If you need reproducibility, scripting, or workload control beyond what a GUI exposes, learn fio. Every other tool on this list is, on some level, a friendlier face on top of it.
Most builders end up running two: CrystalDiskMark or its platform equivalent for the quick read, and fio when a number looks wrong and they want to understand why.
FAQ
Why are my sequential numbers lower than the box says?
Two common reasons. The slot may share PCIe lanes with another device (a second M.2 slot, a SATA controller, or a graphics card slot that drops from x16 to x8 when populated). Or the drive is running at a lower PCIe generation than expected because of motherboard or CPU limits. Check the slot’s spec in the motherboard manual and confirm the link speed in a tool like CrystalDiskInfo or HWiNFO.
Random 4K reads are way below the advertised number. Is the drive broken?
Probably not. Headline random IOPS numbers are quoted at very high queue depth across many threads, which is closer to a server workload than a desktop one. The number that matches daily use is random 4K at queue depth 1, and it is usually a small fraction of the marketing figure. That is normal.
How do I tell if my SSD is thermally throttling?
Run a sustained write workload for three to five minutes and watch the throughput line. If it starts strong and drops sharply, the drive is hitting its thermal limit and the controller is slowing writes to cool down. Blackmagic Disk Speed Test on Mac and fio with a long runtime on any OS will both expose this. A heatsink or better airflow usually fixes it.
Does benchmarking wear out my SSD?
Each test writes a measured amount of data, which counts against the drive’s total bytes written rating. A few benchmark runs are negligible compared to normal use. Repeatedly running long sustained-write tests every day is not, so save the heavy fio scripts for when you have a real question to answer.
