How We Test Hard Drives and SSDs

Each year at PCMag, our experts review dozens of hard drives and solid-state drives (SSDs). We put each drive we review through our rigorous testing regimen at PC Labs, tailored to the specific type of drive being tested. This gives us the data necessary to compare its performance with similar drives and evaluate it fairly. Here, we’ll detail the process, as well as the tests and equipment we use.

---

The Testbeds We Use to Benchmark Storage Drives

Speed is one of the key metrics we use to judge SSDs and hard drives, so part of our testing process involves connecting the reviewed drive to a testbed computer and observing its performance on benchmark tests.

We’ll get into that process shortly, but before we do, we should note that speed isn’t everything when it comes to storage. We also evaluate drives on the basis of value for money and additional features, such as warranty, durability ratings, security features, and supplementary software. Storage drives have become so good these days that sometimes, it’s subtle things that separate an average drive from a winner.

We maintain three dedicated storage testbed systems. Depending on the bus architecture (PCI Express or SATA) and connection protocol (M.2 PCI Express or 2.5-inch SATA for internal SSDs; USB or Thunderbolt for external hard drives and external SSDs), we test any given drive that comes through the labs on one of these testbeds, or on a pair of them. We’ll explain why below.

Internal M.2 SSDs (PCI Express 3.0 and 4.0) and 2.5-Inch Serial ATA Drives

M.2 drives are internal drives in a “gumstick” form factor. PCI Express is the most common bus type now in play with M.2 SSDs. Early implementations of M.2 used the slower Serial ATA (SATA) bus, and some SATA-based M.2 drives and M.2 slots are still out there. But they are yesterday’s news, and we haven’t tested a new SATA M.2 drive in years.

The M.2 internal SSDs we test these days push data over different flavors of the PCI Express (PCIe) bus, which dictate potential top speeds. PCIe 3.0 support is close to ubiquitous among desktop PCs sold in the past five years or so that have M.2 slots. (Most desktops now have at least one M.2 slot supporting PCIe of some flavor; some laptops do, too.) PCIe M.2 support on drives and slots is backward-compatible, and a drive will work at the peak speed of the lesser version.

While PCI Express 4.0 M.2 SSDs offer higher potential sequential-throughput speeds than PCI Express 3.0 ones, PCI Express 4.0 support is generally available only on later-model systems. With desktops, that means AMD-based systems from the X570/B550/TRX40 chipset period onward (using Ryzen 3000 series CPUs or later), and for Intel systems, desktops with Z490 or more recent motherboards (using 11th Generation CPUs or later). You can use a PCI Express 4.0 SSD in a 3.0-only motherboard, but it will bounce down to 3.0 speeds.

(Credit: Joseph Maldonado)

M.2 PCIe SSDs support version 3.0, 4.0, or 5.0 of the spec. To test the speed potential of 3.0 and 4.0 drives, we use a testbed built on an MSI Godlike X570 motherboard with an AMD Ryzen 9 3950X CPU installed. The system has 16GB of DDR4 Corsair Dominator RAM clocked to 3,600MHz, and it employs an Nvidia GeForce RTX discrete graphics card.

PCI Express 4.0 M.2 drives tend to generate a lot of heat under sustained loads. If a drive has its own heatsink, we test it with the sink in place. If it lacks a heatsink or just has a basic label-thickness heat spreader, we test it using the MSI motherboard’s own heat sink.

The rare traditional 2.5-inch SSDs that we still get in for review are also tested on this testbed. They are installed on the first SATA port powered by the motherboard’s main SATA controller.

Internal PCIe 5.0 SSDs

Our latest testbed PC, designed specifically for benchmarking leading-edge PCI Express 5.0 M.2 SSDs, consists of an ASRock X670E Taichi motherboard with an AMD X670 chipset, 32GB of DDR5 memory, one PCIe 5.0 x4 M.2 slot (with lanes that have direct access to the CPU), and three PCIe 4.0 slots. It sports an AMD Ryzen 9 7900 CPU using an AMD stock cooler; a GeForce RTX 2070 Super graphics card with 8GB of GDDR6 SDRAM; and a Thermaltake Toughpower GF1 Snow 750-watt power supply. The boot drive is an ADATA Legend 850 PCIe 4.0 SSD.

PCI Express 5.0 SSDs, which offer maximum theoretical sequential read speeds in excess of 10,000MBps, generate lots of heat under sustained loads and require an effective heatsink. We test them either with the heatsink provided by the manufacturer or with the ASRock board’s actively air-cooled heatsink.

We also test USB4 external SSDs on this testbed, as described below.

External SSDs and Hard Drives (USB or Thunderbolt)

USB 3.2 Gen 2 and Gen 2×2 external drives (which comprise most of the external drives we review these days) are tested on a Windows-based storage testbed housed in a SilverStone tower case. It is equipped with an Asus Prime X299 Deluxe motherboard with an Intel Core i9-10980XE processor. On the board is 16GB of DDR4 Corsair Dominator RAM clocked to 3,600MHz, and the system uses an Nvidia GeForce RTX discrete graphics card to power video. The testbed’s own PCI Express SSD, the boot drive, serves as the primary drive, and the drive being tested is configured as supplemental storage.

Depending on whether an external drive supports the commonplace USB 3.2 Gen 2 standard or the higher-speed USB 3.2 Gen 2×2, it is tested attached to the appropriate port on this PC. In the former case, it is attached to this motherboard’s native USB 3.2 Gen 2 USB Type-C port (a 10Gbps port). In the latter case, the drive is attached to a 20Gbps USB-C port on a USB 3.2 Gen 2×2 PCIe expansion card made by Orico. We installed this card on this testbed to facilitate the testing of these speedier drives.

We don’t use this testbed for USB4-based external SSDs. Instead, we use the same testbed system that we use for PCI Express 5.0 drives. Its motherboard has two native 40Gbps USB4 Type-C ports.

(Credit: Joseph Maldonado)

After we’ve run the tests defined below for external drives with the drive formatted as NTFS, we then format the drive to exFAT and run a couple of supplemental tests on an M3-generation Apple MacBook Pro, testing over Thunderbolt or USB Type-C, as applicable to the drive. If the drive is a Thunderbolt-only drive, we run just the MacBook-based tests.

We test external hard drives the same way we do external SSDs, using the same testbeds and benchmark tests, which we’ll get into next.

---

The Benchmarks We Use: Internal SSDs

Here is a breakdown of the benchmark set we run on internal SSDs, whether M.2 “gumstick” drives or 2.5-inch SATA internals. The drives are secure-erased between each run of the different tests. All drives are set up as secondary data drives on their respective testbeds.

PCMark 10 Storage

The main PCMark 10 Storage test from UL is an invaluable measure, providing a high-level view of how the drive will function under various everyday workloads, such as word processing and videoconferencing.

(Credit: UL)

For internal SSDs, we first run the drives through the PCMark 10 Full System Drive benchmark, which simulates 23 different “traces” (simulated tasks) in the course of the run. The traces flex the drive in ways that approximate launching Adobe-based creative programs, booting up Windows, copying files, launching popular games, and more.

(Credit: UL)

The overall score that PCMark 10 reports back represents how well a drive does throughout the entire PCMark 10 run. This score is the sanctioned score presented by UL’s software at the end of each run. It reflects a weighted average of the various activities that the PCMark 10 storage test simulates, a general indicator of how consistently a drive can perform through the 23 different usage scenarios.

It’s a proprietary number, though, and is meaningful only when compared with scores of other competing drives. That is where our reviews come in.

Getting Granular: Booting Windows (PCMark 10 Trace)

We also dig into the more granular trace data that PCMark 10 generates. The first part of it we report is culled from the Windows boot trace, which simulates a full operating-system startup procedure. The throughput number we report reflects how quickly the drive is able to feed the data required for that task set.

This and the following three PCMark 10-derived, trace-based tests represent a simulation of how quickly a drive is capable of feeding data when launching a particular program, copying files, or, in this case, booting Windows. PCMark 10 records how many megabytes per second the drive is reading what are known as “shallow-queue 4K random” blocks of data (i.e., of the kind in which most applications, games, or operating systems are stored). While UL recommends using the overall “read/write MBps bandwidth” metric in these tests, we dug a bit deeper to include only random 4K bandwidth in order to paint what we believe is a more specific picture of how well a drive can perform in these tasks.

Game Launching Tests (PCMark 10 Trace)

Next, we report data from PCMark 10’s traces around game launching. This again reflects how quickly a drive can read shallow-depth small random 4K packages. Note that the “4K” we’re talking about here is file-block size, not file size; 4K is one of the more commonly used file-block sizes for game installations, though that composition does depend on the title you’re playing.

(Credit: Activision Publishing)

While the three games tested in PCMark 10 are stored primarily in small random 4K, tests from around the web have shown that MMORPGs can more often use the 16K block size, and some games in other genres may tend to employ larger block sizes, from 32K up to 128K. However, for the sake of these tests, 4K small random read is the most accurate block-size metric relevant to these three popular FPS titles: Battlefield 5, Overwatch, and Call of Duty: Black Ops 4. We again report the read throughput for this kind of file.

Adobe Application Launch Tests (PCMark 10 Trace)

Next is the set of results based on traces simulating Adobe application launches. As anyone who regularly works in programs like Adobe Premiere or Photoshop can tell you, the time it takes for these programs to launch is a constant pinch point.

(Credit: Adobe)

Mind you, our results don’t tell the whole story of how a drive will perform for all creative applications. Depending on the complexity of your work and the number of elements in a scene, your software may have to load 3D models, sound files, physics elements, and more; in other words, more than just the program. Still, this is interesting fodder for folks who live and breathe these Adobe apps.

Copy Tests (PCMark 10 Trace)

Finally, we report on PCMark 10’s traces that simulate file-copy actions. While at first these numbers might look low compared with the straight sequential-throughput numbers achieved in benchmarks like Crystal DiskMark, that’s due to the way this score is calculated and the nature of and differences between the source data. If you’re regularly moving files around on your drive from one folder to another, this test is a handy relative throughput measure.

Crystal DiskMark 6

Beyond PCMark 10, we also use the venerable Crystal DiskMark utility for a second opinion on throughput. Crystal DiskMark’s sequential-read tests measure read/write activity with data written in a large contiguous block on the drive, similar to how manufacturers test drives to advertise their peak performance. These tests represent a “best case,” straight-line scenario for file transfers.

(Credit: Crystal Dew World)

We also use Crystal DiskMark’s 4K tests to measure random reads/writes, which reflect data activity in which the drive fetches and writes scattered files and pieces of files across the drive. This is mostly used as a reality check against the wealth of 4K read data culled from PCMark 10’s traces.

3DMark Storage Gaming Benchmark

Gamers have long relied on 3DMark testing to benchmark their CPUs and GPUs. This venerable benchmark suite is also useful for storage evaluation: 3DMark Storage takes SSDs through their paces in performing a variety of gaming-related functions.

(Credit: UL)

It produces an aggregate score, combining traces of tasks from some popular AAA games. These include loading Battlefield V, Call of Duty: Black Ops 4, and Overwatch; recording a 1080p gameplay video at 60fps while playing Overwatch; installing The Outer Worlds from the Epic Games Launcher; saving game progress in The Outer Worlds; and copying the Steam folder for Counter-Strike: Global Offensive. The score is mainly useful compared with other 3DMark Storage scores.

---

The Benchmarks We Use: External SSDs and Hard Drives

In testing, we attach external drives to a native USB 3.2 Gen 2 port on our external storage testbed and afterward (if relevant) to a Thunderbolt/USB Type-C port on our test MacBook Pro testbed laptop. With the Windows machine, we’ll cite if a drive supports Gen 2×2 speeds and is attached instead to the testbed’s expansion-card USB 3.2 Gen 2×2 port.

PCMark 10 Data Drive Benchmark

We’re not done with PCMark 10 quite yet! The Data Drive Benchmark is a solid test to run on any drive you intend to use as a data archive or a backup drive, and it typically takes between 10 and 30 minutes to run, depending on the drive type and its connection standard.

(Credit: UL)

Like the PCMark 10 Storage test, it runs through a host of trace-based activities to simulate typical daily drive activities for a secondary drive. The score it reports is useful compared against other drives’ PCMark 10 results.

Crystal DiskMark 6

For external SSDs, we run the Crystal DiskMark 6 test under the same parameters as for internal drives above (sequential read/write, and 4K read/write).

Blackmagic Disk Speed Test

With this and our next test, we move the drive, if compatible, over to our Apple MacBook Pro test laptop and reformat it into exFAT. We use the macOS-only version of the Blackmagic Disk Speed Test app from the professional media software maker Blackmagic Design (best known for DaVinci Resolve) to perform this test. It reports back a drive’s throughput in bits per second.

(Credit: Blackmagic Design)

This utility is typically used to discern whether a given drive has enough throughput to play back specific video formats smoothly. But it also returns some useful throughput measurements. Blackmagic offers both a read score and a write score, which we compare with those of similar drives. These scores are useful in discovering a drive’s theoretical maximum speed.

Folder Transfer Test

Our final test for external drives is a drag-and-drop test, also performed on our MacBook Pro. It uses the macOS Finder to copy a 1.23GB test folder (full of several different file types) from the testbed’s internal drive to the external SSD being tested. We hand-time the scores, in seconds.

---

How We Test Network Attached Storage

Because a wide range of network conditions outside of our control can affect the performance of network attached storage (NAS) devices, we only perform a folder transfer test on these devices. The specific hard drives installed in a NAS can be a big variable if you’re looking at a NAS drive that doesn’t come pre-populated with drives. (These days, most don’t.)

(Credit: Joseph Maldonado)

This test uses a different sample transfer folder than the one we use on our internal and external drive tests. The NAS test folder comprises 4.9GB worth of music, video, photo, and office document files. We hand-time how long it takes to transfer the folder to the NAS via a wired connection, as well as how long it takes to transfer the file from the NAS back to our testbed.

About Tony Hoffman

Senior Analyst, Hardware

Since 2004, I have worked on PCMag’s hardware team, covering at various times printers, scanners, projectors, storage, and monitors. I currently focus my testing efforts on 3D printers, pro and productivity displays, and drives and SSDs of all sorts.

Over the years, I have reviewed iPad and iPhone science apps, plus the occasional camera, laptop, keyboard, and mouse. I’ve also written a host of articles about astronomy, space science, travel photography, and astrophotography for PCMag and its past and present sibling publications (among them, Mashable and ExtremeTech), as well as for the PCMag Digital Edition.

Read Tony’s full bio

Read the latest from Tony Hoffman

About John Burek

Executive Editor and PC Labs Director

I have been a technology journalist for 30-plus years and have covered just about every kind of computer gear—from the 386SX to 64-core processors—in my long tenure as an editor, a writer, and an advice columnist. For almost a quarter-century, I worked on the seminal, gigantic Computer Shopper magazine (and later, its digital counterpart), aka the phone book for PC buyers, and the nemesis of every postal delivery person. I was Computer Shopper’s editor in chief for its final nine years, after which much of its digital content was folded into PCMag.com. I also served, briefly, as the editor in chief of the well-known hard-core tech site Tom’s Hardware.

During that time, I’ve built and torn down enough desktop PCs to equip a city block’s worth of internet cafes. Under race conditions, I’ve built PCs from bare-board to bootup in under 5 minutes.

In my early career, I worked as an editor of scholarly science books, and as an editor of “Dummies”-style computer guidebooks for Brady Books (now, BradyGames). I’m a lifetime New Yorker, a graduate of New York University’s journalism program, and a member of Phi Beta Kappa.

Read John’s full bio

Read the latest from John Burek