Enterprise SSDs

Why an enterprise drive costs more than a consumer one of the same size

The difference is not capacity but how many times you can write to the drive and how evenly it holds speed. A consumer SSD is fast in short bursts: it has a cache, and while the cache holds, the numbers look excellent. Under sustained writes the cache runs out and speed drops several times over.

An enterprise drive behaves differently. It starts slower but holds a steady rate for hours, and the vendor quotes endurance in full drive writes per day. For a server writing logs, checkpoints and datasets around the clock, that figure is the one that matters.

Three things worth checking

• Write endurance. Measured in DWPD or in terabytes written. A database drive and an archive drive are different classes, and paying for endurance you do not need is as much a mistake as skimping on endurance you do.
• Power-loss protection. Capacitors on the board let the drive finish writing whatever it already accepted from the controller. Without them a sudden outage means corrupted data.
• Form factor and connector. U.2 goes into the server hot-swap bay, M.2 sits on the board and usually holds the operating system. Mixing them up at order time is easy, swapping them later is not.

How drives are split in a GPU server

The working layout looks like this: a mirrored pair of M.2 for the operating system, so one failed drive does not stop the machine, plus several large U.2 drives for data. Datasets and checkpoints are written in large blocks, and that is exactly where steady speed and endurance matter.

Send us the server model and the nature of the load. We will point out where a read-oriented drive is enough and where you need high write endurance.