15K vs 10K vs 7200 RPM: Choosing Server Hard Drives
Spindle speed is presented as a ladder — 15K at the top, 7200 at the bottom — and that framing causes more bad purchases than almost anything else in storage. The three tiers are not better and worse versions of the same product. They are answers to different questions.
This guide covers what each tier is actually optimised for, the arithmetic that decides between them, and where each belongs.
What each tier is answering
15,000 RPM answers: how do I minimise seek latency? It gives up capacity per drive, power efficiency and cost per gigabyte to do it.
10,000 RPM answers: what if the workload is mixed? Which describes most servers. It sits between the two extremes deliberately.
7200 RPM nearline answers: how do I store the most data per pound at sustained duty? It gives up seek performance to do it.
Read that way, "which is best" stops being a sensible question. A 15K drive in a backup target is a waste of money; a nearline drive under a busy virtualisation host is a performance incident waiting to be logged.
Why 15K drives have small capacities
This puzzles people who see a 300GB drive priced above a 2TB one, and the reason is physics rather than segmentation.
A platter spinning at 15,000 RPM experiences considerably more mechanical stress than one at 7200 RPM. Keeping the head flying stably at that speed requires tighter tolerances, which in practice means smaller platters and fewer of them. Less surface area means less capacity. Heat and power draw rise at the same time, which constrains how densely you can populate a chassis.
So a 15K drive is not a small drive that someone forgot to enlarge. It is a fast spindle, and the small capacity is the price of the speed.
The spindle count arithmetic
This is the single most useful thing to understand about performance storage, and it is the reason "upgrading" an array to fewer larger drives so often makes it slower.
Every drive has one set of heads and can service one request at a time from its mechanism. The number of concurrent operations an array can handle therefore scales with the number of drives, not with total capacity.
Sixteen 300GB drives and four 1.2TB drives both give roughly 4.8TB raw. The sixteen-drive array delivers approximately four times the IOPS, because it has four times the heads working in parallel.
Two consequences follow. Performance arrays are built with many small drives deliberately, not because larger ones were unavailable. And replacing a failed 300GB member with a 1.2TB drive is not an upgrade — the RAID group uses 300GB of it and strands the rest, so you have spent more and gained nothing.
Our 300GB 15K guide covers this in more depth.
Where each tier belongs
15K SAS — transactional databases, the latency-critical core of an application, log and index volumes, and any workload where a user or a transaction waits on a disk read. Built as high-spindle-count arrays. See our 600GB 15K guide for where 15K stands against flash today.
10K SAS — virtualisation datastores, application servers, mid-sized databases, active file shares. The general-purpose enterprise tier, and the right answer for most mixed workloads. Our 1.2TB 10K guide covers the capacity point that became the industry default.
7200 RPM nearline — backup targets, archive volumes, surveillance retention, media storage, bulk file serving, secondary tiers behind faster storage. Sequential workloads at sustained duty. Our 8TB nearline guide covers capacity-tier planning.
Rules that apply within a RAID group
Three constraints that catch people when mixing drives, and all three are absolute.
The group degrades to its slowest member. One 7200 RPM drive in a group of 15K members slows the entire group to nearline speed. This is why dropping a single SSD into a mechanical array achieves nothing.
The group uses the smallest member capacity. A larger drive contributes only as much as the smallest one, and the excess is stranded.
Tiering happens between groups, not within them. If you want fast and bulk storage in one chassis, build separate RAID groups from separate drive types. Mixing within a group gives you the worst of both.
Manufacturer, by contrast, does not need to match. Two drives of the same capacity, interface, form factor and speed will work together regardless of who made them — and mixing batches deliberately reduces the chance of correlated failure, since drives from one batch tend to fail at similar times.
Where flash changes the answer
An honest treatment has to address this, because for new deployments it changes the calculation.
SSDs have no seek time at all, so they beat any mechanical drive on latency by a margin no spindle speed can close. For a new latency-critical tier, flash is the correct starting point and 15K is a budget fallback rather than a technical choice.
Two places mechanical drives still hold ground. Existing arrays, where a partial swap achieves nothing because the group runs at its slowest member — replace like for like and treat migration as a project. And sustained write-heavy workloads, because enterprise SSDs are rated by drive writes per day and write-intensive models carry a real premium, while a mechanical drive has no equivalent wear ceiling.
For capacity tiers, mechanical storage remains far cheaper per terabyte and will for the foreseeable future.
Choosing without guessing
Work through it in this order.
1. What is the access pattern? Sequential means nearline. Random means 10K or better.
2. Does a user or transaction wait on it? If yes, 15K or flash. If no, 10K is usually enough.
3. How many spindles can the chassis hold? IOPS scale with drive count, so bay count sets your performance ceiling before capacity does.
4. What does the existing array use? If you are replacing rather than building, this overrides everything above — match it.
Then check the physical constraints: form factor, carrier generation and firmware coding on OEM platforms. Our compatibility guide covers those by platform, or send us the part number from a fitted drive and we will confirm the match.
Common questions
Why does a 300GB 15K drive cost more than a 2TB nearline one?
Because you are buying a fast spindle rather than storage. Platters at 15,000 RPM must be smaller to stay mechanically stable, so capacity is the price of the speed. In performance arrays, spindle count rather than capacity is what determines throughput.
Can I mix 10K and 15K drives in one array?
Within a single RAID group, no — the group degrades to its slowest member, so the 15K drives run at 10K speed. Build separate groups from separate drive types instead; tiering happens between groups, not within them.
Will fewer larger drives simplify my array?
It will also slow it down. IOPS scale with the number of drives, so consolidating sixteen 300GB drives onto four 1.2TB ones removes roughly three quarters of the concurrent throughput while keeping the same capacity.
Should I still buy 15K drives, or go to SSD?
For a new latency-critical tier, flash is the correct starting point. For maintaining an existing 15K array, replace like for like — a partial swap achieves nothing because the group runs at its slowest member. For sustained write-heavy work, mechanical drives have no wear ceiling while write-intensive SSDs carry a premium.
Does the drive manufacturer need to match in an array?
No, provided capacity, interface, form factor and spindle speed match. Mixing manufacturers or batches deliberately can even help, since drives from a single batch share accumulated hours and tend to fail at similar times.
Tell us the workload and the chassis and we will recommend a tier rather than a part number.
