15k RPM hard drives were the fastest spinning storage you could buy for most of the 2000s and early 2010s. SAS drives from Seagate (Cheetah/Savvio) and some Fujitsu and Hitachi models spun at 15,000 RPM instead of the usual 5,400 or 7,200, cutting seek times and boosting sustained transfer rates. If you’re reading this, you’ve probably either inherited some old enterprise gear, found a cheap SAS drive on eBay, or you’re trying to decide whether to keep running one in a workstation or NAS. The short answer: an SSD beats a 15k drive in every performance metric that matters today, and it’s not close. But there are a few reasons people still reach for 15k drives, so let’s go through the actual numbers.
What 15k drives were actually good at
A 15k RPM drive has an average seek time around 3.5-4ms, versus 8-10ms for a 7,200 RPM desktop drive. Sustained sequential throughput tops out around 200-280 MB/s depending on the model and where on the platter you’re reading. Random 4K performance is where it really falls apart by modern standards: you’re looking at maybe 1-2 MB/s and a few hundred IOPS at best, because you’re still bound by a physical head having to move to a track and wait for the platter to rotate into position.
That seek time advantage mattered enormously when these drives were new, because the alternative was a 7,200 RPM drive with worse seeks. In a database server or a RAID array handling lots of small random transactions, 15k drives in RAID 10 were genuinely the right call for a long time. That’s the whole reason they existed.
The SSD numbers
A basic SATA SSD today does sequential reads around 500-560 MB/s and random 4K reads in the tens of thousands of IOPS, often 90,000+ for a decent drive. An NVMe SSD on PCIe 4.0 will do 5,000-7,000 MB/s sequential and random 4K performance in the hundreds of thousands of IOPS. There’s no asterisk here, no “under ideal conditions.” Flash memory doesn’t have a physical head waiting for a platter to spin into place, so the random I/O penalty that defines mechanical drives just doesn’t exist in the same way.
Here’s a side-by-side with real-world numbers from drives of each type:
| Metric | 15k RPM SAS HDD | SATA SSD | NVMe SSD (Gen4) |
|---|---|---|---|
| Sequential read | 200-280 MB/s | 500-560 MB/s | 5,000-7,000 MB/s |
| Random 4K read (IOPS) | ~300-400 | 80,000-100,000 | 400,000-700,000 |
| Average seek/access time | 3.5-4 ms | ~0.05-0.1 ms | ~0.01-0.02 ms |
| Idle power draw | 8-10 W | <0.5 W | ~1-2 W |
| Typical capacity available new | Discontinued (used: 300GB-900GB) | 250GB-8TB | 500GB-4TB (common) |
| Noise/vibration | Audible whine, real vibration | Silent | Silent |
Note that 15k drives aren’t manufactured anymore in any meaningful volume. Seagate stopped new development years ago. Anything you buy now is used or old new-stock, and it’s SAS, which means you need a SAS controller or HBA, not a plain SATA port. That’s an extra cost and compatibility headache most people don’t need to take on in 2024.
Failure modes worth knowing about
15k drives run hot and loud because the platters are spinning so fast, and that heat and vibration is exactly what wears out bearings over time. Used ones you find secondhand have unknown power-on hours, often in the tens of thousands, and SAS drives don’t have the same consumer-friendly SMART reporting tools that make it easy to check drive health. When they fail, it’s often sudden; a head crash or bearing seizure, not a graceful slowdown. If you’re going to run one, do it in RAID with a hot spare, never alone.
SSDs fail differently. NAND wears out with write cycles, and you’ll generally get a warning via SMART attributes (reallocated sectors, wear leveling count) before total failure, though controller failures can still be sudden. For typical desktop use, a consumer SSD will outlast the system around it. Endurance is measured in TBW (terabytes written), and even a budget drive usually has a rating well beyond what a typical user writes in the drive’s useful life.
When a 15k drive still makes sense
Almost never, honestly. The one case I’d consider it: you have existing SAS infrastructure (an older server with an HBA and backplane already in place) and you need bulk-ish storage that’s faster than 7,200 RPM SAS/SATA for a specific legacy workload, and you don’t want to touch the budget for new SSDs. Even then, a cheap SATA SSD on an adapter often ends up both faster and cheaper once you account for the used-drive risk.
If you’re maintaining an old server as a hobby or homelab project and already have the drives, there’s no shame in running them until they die, as long as your data has a backup elsewhere. But buying 15k drives today as a performance upgrade for anything doesn’t make sense.
What to buy instead
For a desktop or laptop upgrade, a plain SATA SSD is the right call if you’re on an older board without M.2 slots, or you just want a cheap, reliable boot drive. If your motherboard supports it, an NVMe SSD gets you the sequential and random performance that 15k drives were never close to, at a fraction of the noise and power draw. For bulk storage where capacity matters more than speed, a 7,200 RPM internal hard drive is still the economical choice per terabyte, and it’ll outperform a 15k drive on sequential transfers anyway since modern 7,200 RPM drives benefit from higher areal density. The RPM number alone stopped being the right way to judge a hard drive’s speed once density caught up and overtook it.
If you’re specifically dealing with an old SAS server and need to replace a dying 15k drive, check whether the chassis supports SAS SSDs before you default to buying another spinning disk. It costs more up front, but you won’t be back here again in two years swapping out another bearing failure.
