Slow file access, strange freezes, users complaining the server 'hangs' — these can be a dying drive, a bad cable, or a busy network. Two minutes with the right numbers separates a real disk failure from a false alarm.
Technicians lose more data to wrong first moves than to the failure itself. They run a full disk scan on a drive that was quietly dying, and the scan finishes it. The correct first move is the opposite: read the drive's own health counters — a light, read-only check — and only then decide. This page shows which numbers matter and what each one is really telling you.
A drive rarely dies silently. The classic pattern is a server that gets slower in a particular way: opening shared folders takes seconds, copying a file stalls at the same place, backups that took an hour now take four, or the server freezes for 20–30 seconds and then recovers as if nothing happened. Those pauses are often the drive retrying a weak sector again and again before finally reading it. Random once is nothing; the same behaviour every day is the drive asking for help.
Every modern drive keeps health counters called SMART attributes. CrystalDiskInfo shows them free, in plain rows. You can also ask Windows directly from an admin PowerShell:
Get-PhysicalDisk | Select FriendlyName, HealthStatus, OperationalStatus — a fast health headline per disk.Get-WmiObject -Namespace root\wmi -Class MSStorageDriver_FailurePredictStatus — PredictFailure: True means the drive itself is predicting its own death.Judge the raw values: on a healthy drive, attributes 05, C5 and C6 are all zero. Reallocated sectors (05) are spots the drive already gave up on. Pending sectors (C5) are spots it cannot read right now. Uncorrectable (C6) are spots it tried and failed. Any of them rising week to week is a failing surface — full stop.
Windows logs what it sees. In Event Viewer under Windows Logs, System, filter for sources disk and storahci / stornvme: Event ID 7 (bad block), Event ID 51 (error during a paging operation) and Event ID 153 (an I/O operation was retried) are the fingerprints of a struggling drive. A handful across a year is noise. Dozens in a week, always naming the same disk, is evidence. Note which physical disk the messages point at — on a multi-drive server, cloning the wrong disk wastes the whole window.
Attribute C7 (UltraDMA CRC Errors) tells a different story: data is arriving corrupted between the drive and the board, which usually means a loose or damaged SATA cable, a stressed connector, or a failing port — not a dying surface. If C7 is high but 05/C5/C6 are all zero, reseat or replace the cable first and watch whether C7 keeps rising. Old C7 errors never reset to zero; only new growth matters. This is the one 'failing drive' alarm that a low-cost cable can honestly fix.
Until your data is safely copied, treat a suspect drive like a patient with a weak heart — no stress tests.
The order that saves data is always the same: prove it lightly, clone it once, diagnose the copy.
The video starts from exactly these warning signs on a real server and takes the drive through to a completed clone.
▶️ Watch Full VideoYes — plan the clone now, while it still runs fine. Caution almost always means 05, C5 or C6 has left zero. A drive in that state can work for months or fail this week; nobody can promise which. Cloning while it is healthy enough to read everything is the whole game.
Use the controller's own management tool (or its boot-time utility) — it reports each member disk's health and predictive-failure state. The same rule applies: a member flagged predictive-failure gets replaced by rebuild, not debated with.
Not necessarily. SMART is a good witness, not a perfect one, and it says nothing about RAM errors, a failing cable or a sick file system. Check C7 and the Event Viewer disk errors, test with a known-good cable, and if corruption follows the drive to another machine, treat the drive as guilty despite the green label.