How Storage Works, Part 2: How a Hard Drive Works

A hard drive is a stack of spinning magnetic plattersA platter is a spinning disk inside a hard drive that holds the magnetic recording media. One or more actuators give the drive physical access to the data on it., a head that flies over each face and an arm that moves the head to a track. Nearly every label on a spec sheet buys either more bits on those platters or more reads from that one arm. Conventional magnetic recording (CMRConventional magnetic recording writes a hard drive’s tracks side by side without overlap, so any track can be rewritten on its own.) and shingled magnetic recording (SMRShingled magnetic recording overlaps each track on the one before to fit more data. Rewriting part of a band means rewriting the rest of it, which slows sustained writes.) trade write freedom for density, heliumA helium-sealed hard drive is filled with helium instead of air. The thinner gas lets the platters spin with less drag and flutter, so more platters fit in the same case. makes room for more platters, and heat-assisted magnetic recording (HAMR)HAMR heats a tiny spot of a hard drive platter with a laser as the head writes, so media that holds its bits more firmly at room temperature can still be written, and bits can sit closer together. and energy-assisted perpendicular magnetic recording (ePMR)ePMR is Western Digital’s perpendicular recording with an electrical current applied in the write head, which makes the head switch faster and more consistently, with no laser. shrink the bit, and a second actuatorAn actuator is the arm assembly that moves a hard drive’s read and write heads across the platters. A dual-actuator drive splits the heads between two arms that can work at the same time. adds reads.

Part 1 put the hard drive where it keeps the lowest cost per bit and loses on access time. Access time comes from one arm chasing a spinning surface, and cost per bit from how many bits each platter face holds. The HDD Timeline holds the capacity and recording method of each maker release by year, and The Capacity Drives in Memory and Storage for SI sets the largest quad-level cell (QLC) SSDs against a hard drive on power and rack space.

Where a bit sits on a platter

Each platter is coated on both faces, and each face has its own head. The IEEE’s International Roadmap for Devices and Systems (IRDS) counts two heads per disk and puts the largest nearline drives at up to 10 disks for Seagate and 11 for Western Digital.

A head rides on a slider that flies a hair above the spinning surface, and an arm swings it to a track, which is one of the concentric rings that carry the data. Lithography borrowed from chipmaking shapes the write and read elements, and their width sets the track pitch. Each track is cut into sectors, the unit the host addresses, and Seagate’s Exos 2X18 data sheet lists 512 bytes or 4,096 bytes per sector.

Density runs in two directions. Bits per inch (BPI) counts how short each bit is along the track, and tracks per inch (TPI) counts how narrow the tracks are across it. Raising areal densityAreal density is how much data a hard drive platter stores in a given area of its surface. Raising it is how makers fit more data on each platter. means writing smaller bits and using narrower tracks. Dual-stage actuators (a coarse arm plus a fine positioner) reached enterprise drives in 2005 and are in every hard drive now, and a piezoelectric micro-actuator at the arm tip raises the servo bandwidth and with it the track density a drive can hold.

Mechanics also set access time. At 7,200rpm, Seagate’s Exos X20 data sheet lists an average latency of 4.16ms, half a turn of the disk, with the seek on top. On my test system, CrystalDiskMark 8.0.4 measured a Western Digital (WD) Red Pro 20TB at a median 4,215µs for 4KiB random reads at queue depth (QD) 1, with single runs from 2,996µs to 5,001µs over four runs. That default test reads inside a 1GiB test file, which would keep the seeks short. That is about 238 input/output operations per second (IOPS), and about 81 times the latency of the Kioxia XG8 SSD in Part 1.

Conventional and shingled tracks

Spec sheets label the track layout CMR or SMR, which Toshiba’s product pages define as conventional and shingled magnetic recording. CMR keeps the tracks side by side with guard space between them, per Backblaze’s Q2 2026 Drive Stats, so a drive can rewrite one track without touching the next.

SMR sets the track pitch far below the width of the write field, so each new track overlaps the last like roof shingles. Random reads still work, but a write overwrites part of the neighboring tracks in the direction of shingling, and the roadmap’s verdict is that SMR does not readily accommodate random writes.

Density is the payoff. IRDS puts the gain at 10 to 18%, with 17 to 18% for Western Digital’s UltraSMR. Toshiba’s September 2024 release shows a pair built on a common 10-disk architecture, the 24TB MG11 in CMR and the 28TB MA11 in SMR, and the SMR model holds 17% more. Western Digital’s October 2024 release pairs an 11-disk, 26TB CMR drive with a 32TB UltraSMR one, which works out to 23% more and sits above the roadmap’s range.

Writes pay for it. A write disturbs the tracks after it, so changing data in the middle of a shingled band means rewriting the rest of the band, and, per Backblaze, updating data may require the drive or its software to reorganize a larger section. An SMR drive keeps zones with a larger track pitch for random writes or caches them in non-volatile memory, needs special algorithms and metadata, and can hurt write-intensive work, particularly write latency. Toshiba’s March 2026 release makes the same point from the maker’s side, that random writes can degrade performance because of the overlap.

Who does the rewriting splits SMR three ways. In drive-managed SMR the firmware places data and rewrites bands, so the drive looks ordinary and write performance can become unpredictable. Host-managed SMR makes the operating system or application write sequentially inside zones, which is how Toshiba’s M12 SMR drives work, and host-aware drives sit between the two. Standards let the host either be aware of an SMR device or manage it directly.

SMR is common in data centers and in external backup drives. Seagate said in 2022 that nearly 25% of its nearline drives were SMR. Backblaze’s Q2 2026 report counts no shingled drives in its fleet.

A desktop or network-attached storage (NAS) array writes wherever the file system puts a file, so CMR is the safe label there, and SMR belongs where software writes in order.

Why drives are sealed with helium

A stack of spinning disks drags on the gas around it. Helium has one-seventh the density of air, which cuts the drag on the stack and so the power the spindle motor needs, and softens the flow forces that buffet the disks and arms, according to IRDS. With that turbulence gone, the disks and the data tracks can both sit closer together, and the drive runs cooler and quieter.

HGST (Hitachi Global Storage Technologies, now part of Western Digital) began shipping production helium drives in 2013, and Seagate followed in 2015. Seagate’s Exos 2X18 data sheet says the helium design lowers total cost of ownership through lower power and weight, which is a maker’s claim and not a measurement.

Tighter spacing turns into more disks. Seagate’s 10TB helium drive of January 2016 had seven platters and 14 heads. Toshiba’s 14TB MG07ACA of December 2017 used nine disks and laser welding to keep the helium in. IRDS reports that Western Digital disclosed a 10-platter drive for its 22TB Ultrastar in 2022 and had 11 platters by 2024, and that Toshiba says it was first to verify a 12-platter technology to be commercially available by 2027.

Each added disk is a mechanical problem. Fitting more platters into a standard drive enclosure is pushing makers from aluminum substrates to thinner, more rigid glass-ceramic ones and forcing a redesign of the suspensions and stack, and Toshiba’s 11-disk M12 series already uses glass. Higher capacities now come mainly from more platters, from helium that reduces disk flutter, from better head-disk interface smear properties and from higher areal density.

Heat and current help the head write

Most drives write with perpendicular magnetic recording (PMR). Plain PMR cannot be extended much beyond 1.3Tb/in², although 2Tb/in² has been demonstrated, and the roadmap calls the cause the trilemma. Denser bits need smaller magnetic grains, small grains need more magnetic anisotropy to stay stable, and a head can only produce so much write field. Energy-assisted recording adds energy during the write so the head can switch media that a plain field cannot.

HAMR takes the heat route. Seagate’s HAMR page explains that at high density the magnetism of one grain can disturb its neighbors, that media stable enough to prevent it are hard to flip, and that a small laser diode on each head momentarily heats a tiny spot so the head can flip one bit at a time.

Heating lowers the media’s coercivity below the head’s write field, and the spot then cools back to ambient. IRDS puts the spot on Seagate’s 30TB drive at near 450°C, on iron-platinum (FePt) media. Seagate quotes areal densities of 3TB per disk and beyond.

HAMR reached volume in late 2024, and Seagate’s Mozaic 4+ release puts drives of up to 44TB in volume shipments to two hyperscale cloud providers, with a roadmap from 4TB-plus per disk toward 10TB per disk and 100TB drives. Each head needs its own laser. IRDS estimates that 71 million nearline drives shipped in 2024, and at its assumed 10 disks and two heads each that is about 1.4 billion heads, so roughly 1.4 billion lasers if every one were HAMR.

Western Digital took a different route. Its ePMR applies an electrical current to the main pole of the write head throughout the write, which makes the head switch faster and more consistently, cuts timing jitter and so raises BPI, with no laser. In its February 2026 announcement, Western Digital said its 40TB UltraSMR ePMR drive is in qualification with two hyperscale customers, that it plans to extend ePMR to 60TB, and that HAMR will scale to 100TB by 2029 with ramp production in 2027.

Microwave-assisted magnetic recording (MAMR)MAMR adds a radio-frequency field from a spin torque oscillator in the write head to the write field, which helps the head write media that holds its bits more firmly. adds a radio-frequency field from a spin torque oscillator to the write field, and Toshiba uses a flux-control version called FC-MAMR. Modeling suggests MAMR could reach near 4Tb/in² if the oscillator requirements are met and suitable media are engineered, but very limited experimental gains have been shown, and Toshiba’s March 2026 release says it plans HAMR and 12-disk drives later.

These labels answer a different question from CMR and SMR. CMR and SMR describe how tracks sit, while PMR, ePMR, MAMR and HAMR describe how the head writes a bit. Seagate’s IronWolf Pro 24TB with HAMR is the proof, a HAMR drive with a CMR layout.

Two actuators in one drive

Capacity per drive grows faster than the number of random requests one arm can serve, so IOPS per terabyte falls. With a single actuator and one active head at a time, as IRDS frames it, the drive has to service ever more data. Putting more read and write elements on a slider raises throughput but not IOPS. More IOPS need more actuators, either by splitting the heads across several voice coil motors, as in Seagate’s MACH.2, or by giving each surface a second head and actuator.

Seagate’s Exos 2X18 is the first design. Its 18TB appears as two independently addressable 9TB logical units on Serial Attached SCSI (SAS) and as one device on SATA. On SATA, Seagate’s FAQ explains, the first 50% of the logical block addresses (LBAs) belong to one actuator and the second 50% to the other. Hosts must send commands to both actuators at once to see any gain. A SATA log page lists each actuator’s range, and Linux 5.19 shows the ranges under independent_access_range.

Seagate rates the drive at up to twice the performance of a single-actuator drive when both actuators work. Seagate’s data sheet lists 304 random reads and 560 random writes per second at queue depth 16, against 168 and 550 on the single-actuator Exos X20. Seagate labels the X20’s row WCD without defining it, so the comparison is indicative.

Per terabyte, the 2X18 gives 16.9 random reads against 8.4, which doubles them. Random writes go from 27.5 to 31.1, a much smaller step.

Two actuators cost a disk. The 2X18 has nine disks to the X20’s 10 at similar areal density, so the second actuator currently takes one disk of capacity, and by Seagate’s account, at 4KB random reads and queue depth 16, the drive draws less than 20% more power for up to twice the performance.

Western Digital’s Dual Pivot design aims at up to 2x sequential IO gain in a 3.5-inch drive and, the company says, avoids the capacity loss of earlier dual-actuator designs, but it is a February 2026 plan and not a drive on sale. The dual actuator reads as a data center part. Gains exist only when software addresses both halves, so a desktop that sends one stream to an 18TB SATA device may keep one actuator busy while the other idles.

What a current spec sheet tells you

A data sheet answers four questions about a hard drive. It says how the tracks are laid out (the CMR or SMR line), what gas the drive holds, how fast it turns and streams, and how hard the maker rates it to work. The workload ratingA workload rating is the most data a year, read and written, that a maker says a hard drive is built to move. Going over it can hurt its reliability., in terabytes a year, is that last one, and it sits beside the mean time between failures (MTBF) and the warranty. Four CMR drives follow, straight from the makers’ own sheets and manuals.

Source: Western Digital data sheets and Toshiba product manuals.

All four carry the same workload rating. One name can also hide two drives. Seagate’s IronWolf Pro 24TB comes as the helium ST24000NT002, with a 285MB/s maximum sustained rate, and as the HAMR ST24000NT031, with a 240MB/s maximum.

The helium sheet puts Helium on its Drive Design line and the HAMR sheet puts HAMR there, while both list a CMR layout and a workload rate limit of 550. Read the model number and not the product name. A maximum sustained rate is a maker’s rating, a different kind of figure from a measurement. On my test system, CrystalDiskMark 8.0.4 read my WD Red Pro 20TB sequentially at a median 269.6MB/s (1MiB blocks, queue depth 8, two runs), which is a result for one drive and not a rating.

What to watch

Capacity is moving by the disk count and by the density per disk at once. Seagate’s Mozaic 4+ release has HAMR drives in volume shipments, Western Digital’s announcement has an ePMR UltraSMR drive in qualification and Toshiba’s M12 release has SMR drives sampling.

The longer bars are plans, and the HDD Timeline records which ones arrive. The Best HDDs picks apply the labels in this guide to drives a buyer can order.

The IronWolf Pro pair leaves one question open. Its HAMR model lists a 240MB/s maximum against 285MB/s for the helium one, and the sheets do not say whether the recording method or the platform behind it explains the gap.

Below the hard drive in the hierarchy sits tape, cheaper per terabyte and slower to reach, and Part 3 covers it with the optical, glass and DNA media aiming at the same cold data.