Enterprise Hard Drive Buying Guide: SAS vs SATA, 7.2K vs 10K, CMR vs SMR

SAS vs SATA: What the Interface Actually Changes

At 7,200rpm, a SAS drive and a SATA drive with the same platters have the same mechanics and the same sequential speed. The interface doesn't make the drive faster. What SAS adds is around the drive rather than inside it:

• Dual ports. A SAS drive has two independent connections, which matters in dual-controller storage arrays and some high-availability designs. Irrelevant in a single server, but it's why SAS is the enterprise default.
• Full-duplex command handling and deeper queues. SAS handles many outstanding commands more gracefully than SATA's 32-deep queue, which helps under concurrent load from many VMs or users.
• Better error reporting and recovery. SAS drives report errors in more detail and controllers handle SAS faults more predictably.
• High spindle speeds exist only on SAS. 10K and 15K drives are SAS-only; the last 10K SATA drive was discontinued years ago.
• Backwards compatibility runs one way. A SAS controller accepts SATA drives. A SATA controller cannot use SAS drives. PERC and Smart Array controllers are SAS, so both work.

7.2K vs 10K vs 15K: Why Faster Spindles Are Nearly Gone

Spindle speed sets random access latency, and for years 10K and 15K SAS drives were how you got IOPS out of a database server. SSDs ended that. A SATA SSD delivers hundreds of times the random IOPS of a 15K drive at lower cost per IOP, and the high-speed spinning drive market has shrunk to small capacities as a result.

The practical consequence: buy 7,200rpm drives for capacity and SSDs for performance, and don't buy new 10K or 15K drives at all. The one exception is replacing a failed drive in an existing 10K array on a refurbished server, where matching the array matters more than the spindle speed. Those replacement drives are readily available and cheap precisely because the market has moved on.

CMR vs SMR: The Trap That Breaks Arrays

This is the section to read if you read nothing else. Two recording technologies exist on modern hard drives, and one of them is unsafe in any array.

An SMR drive hides the penalty behind a small conventional-recording cache area. For light desktop use, writes land in the cache and get shuffled into the shingled zones during idle time, and the drive behaves normally. A RAID rebuild or ZFS resilver is the opposite of light use: sustained writes across the entire drive with no idle time. The cache fills within minutes, every subsequent write becomes a read-modify-write of a whole shingled band, and throughput collapses.

What happens next is worse than slowness. RAID controllers and ZFS have command timeouts. A drive that stops responding for long enough is treated as failed and ejected. During a rebuild, that means a physically healthy replacement drive gets dropped from the array partway through, the rebuild aborts, and the array is back to degraded or gone. Rebuilds that should take hours take days or never finish, and the window during which a second failure loses everything stretches accordingly.

How to check what you're buying

• Enterprise and NAS-class drives are CMR. Seagate Exos and IronWolf Pro, WD Ultrastar, Gold, and Red Plus/Pro, Toshiba MG and N300 series. Since a 2020 disclosure controversy, manufacturers label these explicitly.
• Desktop, external, and "shucked" drives are frequently SMR and often not labelled. Plain WD Red (not Plus or Pro), many Barracuda models, and most large external USB drives are SMR.
• Check the manufacturer's datasheet by exact model number, not the product family. Several families mix CMR and SMR across capacities.
• If a large-capacity drive is unusually cheap, SMR is the usual reason, in the same way that copper-clad aluminium explains cheap network cable.

The Second Trap: Error Recovery Timeouts

Less famous than SMR but the same failure mode. When a hard drive hits a sector it can't read cleanly, it retries. A desktop drive will retry for thirty seconds or more, trying heroically to recover the data on its own. An enterprise or NAS-rated drive caps that at around seven seconds, reports the error, and lets the RAID controller reconstruct the sector from parity. The feature is called TLER by Western Digital, ERC by Seagate, and goes by other names elsewhere.

The reason it matters: most RAID controllers time out a non-responding drive at around eight seconds. A desktop drive in deep recovery looks, to the controller, identical to a dead drive. It gets ejected, the array degrades, and a rebuild starts, all because of one marginal sector that parity could have fixed in milliseconds. In a rebuild, the same thing on a surviving drive fails the rebuild.

Drive Classes: What You're Paying For

The workload rating is worth understanding. It's the volume of data the drive is warranted to read and write per year. A backup target that ingests 2TB nightly moves over 700TB a year, which exceeds a NAS-class rating and sits comfortably inside an enterprise one. Match the class to the volume, not just to the interface.

Two other things to know about modern large drives. Capacities above roughly 10TB are helium-filled, which reduces drag and heat and is entirely standard. And enterprise 3.5" drives now commonly ship at 16TB, 20TB, and beyond, which brings the RAID 6 rebuild-window argument into force on every array built from them.

Dell and HPE: Caddies, Firmware, and Sector Sizes

Buying a bare drive for a PowerEdge or ProLiant involves three things that waste afternoons if they're not known in advance:

• Caddies are generation-specific. A Dell 14G caddy doesn't fit a 13G bay, and HPE Gen10 smart carriers differ from Gen9. A bare drive without the correct caddy can't be installed. Buy the drive with the caddy, or confirm the caddy part number for your chassis.
• Vendor firmware and drive identification. Dell and HPE sell drives with their own firmware. Third-party drives work in almost all cases, but iDRAC or iLO may flag them as uncertified, and on some HPE generations the amber "not authenticated" indicator lights. The drive functions; the warning is cosmetic. Some older PERC firmware was stricter, and a matched vendor drive is the path of least friction for a production array.
• Ex-array SAS drives may be formatted at 520 or 528 bytes per sector. Drives pulled from NetApp, EMC, and similar storage arrays use non-standard sector sizes with embedded checksums. A PERC or Smart Array won't use them until they're low-level reformatted to 512 bytes, which takes hours per drive. A reputable seller has already done this; a cheap listing often hasn't.

Buying Used Hard Drives

Hard drives are the one server component where "used" needs the most scrutiny, because they're the component with a mechanical life and the most likely to fail. That doesn't make used drives a bad buy; it makes the SMART data mandatory.

• Reallocated, pending, and uncorrectable sectors must all be zero. Any non-zero value means the drive is already remapping bad areas. Reject it.
• Power-on hours disclosed and priced accordingly. Enterprise drives are rated for years of continuous operation; 30,000 hours in a cool data centre is not disqualifying. Undisclosed hours are.
• Buy from the same batch where possible, but stagger replacements. Identical drives of identical age fail at similar times. Some administrators deliberately mix batches in an array for that reason.
• Keep one spare on the shelf. Always. Our inspection checklist covers the full SMART review.

Heat and Spinning Drives in Qatar

Of every component in a server, the hard drive is the one most sensitive to sustained heat. Motors, bearings, and lubricants all age faster, and the failure statistics bear this out: drives run above roughly 40°C fail measurably sooner than drives kept in the low 30s. A full 12-bay chassis of 7,200rpm drives generates its own heat on top of the room's.

• Check per-drive temperature in iDRAC or iLO, not just the inlet. Drives in the middle of a densely packed backplane run warmer than the ones at the edges.
• Helium drives run cooler than air-filled equivalents at the same capacity, which is a small but real advantage in a warm room.
• Populate bays with blanks where drives aren't installed, so airflow passes over the drives that are.
• A drive-heavy server belongs in the conditioned room. Of all the servers to keep out of the unventilated cupboard, the backup target full of hard drives is the one. Our cooling guide covers placement.

What to Buy for Which Job

Frequently Asked Questions

Is SAS faster than SATA?

Not at the same spindle speed. A 7,200rpm SAS drive and a 7,200rpm SATA drive have identical mechanics and sequential throughput. SAS adds dual ports, deeper command queues, and better error handling, which matter under concurrent load and in dual-controller arrays, not in raw speed. High-speed 10K and 15K spindles exist only on SAS.

What is the difference between CMR and SMR?

CMR writes tracks side by side so any track can be rewritten independently. SMR overlaps tracks like roof shingles to fit more data, so rewriting one track means rewriting its neighbours. SMR is cheaper but collapses under sustained random writes, which is exactly what a RAID rebuild or ZFS resilver generates. Never use SMR in an array.

Can I use desktop hard drives in a server?

Not in an array. Desktop drives lack capped error recovery, so a marginal sector makes the drive retry for 30 seconds or more and the RAID controller ejects it as failed. Many are also SMR. They're rated for a fraction of the annual workload of an enterprise drive. Enterprise or NAS-class drives cost more for reasons that show up during the first rebuild.

Should I buy 10K SAS drives?

Only to replace a failed drive in an existing 10K array. For any new purchase, 7,200rpm drives provide capacity and SSDs provide performance, and 10K drives are worse at both. They remain cheap and available precisely because the market has moved on.

Do I need Dell or HPE branded drives?

Usually not. Third-party enterprise drives work in almost all cases, though iDRAC or iLO may flag them as uncertified and some HPE generations show an amber indicator. The drive functions normally. For a production array where every warning matters, vendor-branded drives are the path of least friction. Either way, you need the correct generation-specific caddy.

Where can I buy enterprise hard drives in Qatar?

ServerDove supplies enterprise SAS and SATA hard drives for Dell and HPE servers across Qatar, with caddies, CMR and sector format confirmed, and SMART data disclosed on used drives, delivered in Doha and nationwide with local warranty.

Final Thoughts

Three rules cover nearly every hard drive purchase for a server. Buy 7,200rpm for capacity and leave performance to SSDs. Buy enterprise or NAS-class, never desktop, because capped error recovery and CMR recording are what keep an array alive through a rebuild. And confirm CMR by exact model number before ordering, because the cheapest large drive on the listing is cheap for a reason.

Then match the caddy to the chassis, keep a spare on the shelf, and keep the drive-heavy server in the cool room. The drives will outlast the server around them.