hdparm

What it does

hdparm gets and sets parameters on SATA/IDE (ATA-family) hard disks on Linux, operating directly on a block device like /dev/sda. It can report drive identification data, benchmark raw read throughput, toggle features like write caching or the drive's acoustic management, and control power states (spin-down/standby). It's a low-level tool that talks to the drive's firmware through the kernel's ATA ioctl interface, so it mostly applies to spinning disks and SATA SSDs rather than NVMe.

Beginner examples

• sudo hdparm -i /dev/sda — show drive identification (model, firmware revision, geometry)
• sudo hdparm -I /dev/sda — show the full detailed ATA IDENTIFY data
• sudo hdparm -tT /dev/sda — quick read-speed benchmark (cached and buffered)
• sudo hdparm -C /dev/sda — check whether the drive is active, idle, or in standby

sudo hdparm -tT /dev/sda

Advanced examples

• Compare read throughput across two drives to spot a failing one: sudo hdparm -tT /dev/sda /dev/sdb
• Disable write caching for safety on a drive without a battery-backed controller: sudo hdparm -W 0 /dev/sda
• Force a drive to spin down into standby to save power or reduce noise: sudo hdparm -y /dev/sda
• Set the standby (spin-down) timeout so the drive sleeps after inactivity: sudo hdparm -S 120 /dev/sda
• Check and set the Advanced Power Management level (lower = more aggressive power saving): sudo hdparm -B 127 /dev/sda

sudo hdparm -tT /dev/sda /dev/sdb

Common mistakes

• Running hdparm -tT on an SSD and drawing conclusions about "disk speed" — the numbers are useful for relative comparisons but don't reflect real-world random I/O performance.
• Forgetting sudo, which causes silent permission errors or truncated output instead of a clear failure.
• Assuming hdparm works the same way on NVMe drives — most feature flags and timing commands are ATA-specific and simply don't apply.
• Changing write-cache or power settings (-W, -B, -S) on a production disk without understanding the durability tradeoff — disabling write cache is safer but slower; enabling it is faster but risks data loss on power failure without a UPS.

Tips

• Run -tT a few times in a row; the first run may be affected by prior cache state, so a couple of repeats give a more stable benchmark.
• Use -I (capital) instead of -i when you need the full raw IDENTIFY payload, including supported features, security status, and SMART capability flags.
• -C is a quick, non-destructive way to check if a drive is spun down without waking it (unlike issuing a read, which would force it to spin up).
• Combine with smartctl (from smartmontools) for actual health/SMART data — hdparm doesn't do health diagnostics, only parameter get/set and basic timing.

Best practices

• Always double-check the target device (/dev/sda vs /dev/sdb) before changing settings — running the wrong flag against the wrong disk can silently degrade performance or reliability.
• Avoid destructive-sounding flags (like security erase commands) unless you fully understand ATA security features; a misused command can lock a drive.
• Persist power-management changes (like -B, -S) via a udev rule or systemd unit, since hdparm settings are typically not retained across reboots by default.
• Use -tT for quick sanity checks only; for serious benchmarking use fio or dd with oflag=direct to get more controlled, reproducible numbers.

Real-world use cases

• Diagnosing a suspected failing hard drive by comparing its buffered read throughput against a known-good drive of the same model.
• Tuning laptop power consumption by adjusting a spinning disk's standby timeout and APM level to reduce battery drain.
• Verifying that write caching is enabled/disabled as expected after a firmware update or RAID controller configuration change.

Common interview questions

• What's the difference between hdparm's -t and -T flags? -T measures cached reads from the buffer cache (memory speed); -t measures buffered/physical reads directly from the disk, bypassing cache, giving a real disk throughput estimate.
• Why doesn't hdparm work well on NVMe drives? hdparm speaks the legacy ATA command set over an ioctl interface; NVMe drives use a different command protocol over PCIe, so most hdparm features don't apply — nvme-cli is the NVMe equivalent.
• How would you check if a drive has spun down without waking it up? hdparm -C /dev/sda queries the power state without issuing an I/O request, so it won't force the drive to spin up.