SSD and NVMe drives wear out differently from mechanical disks: there are no moving parts, but there is a limited number of write-erase cycles for the flash memory cells. This article explains what TBW means, how to read SMART attributes, and when to plan a drive replacement on a VDS or a dedicated server.
What TBW is and how it is calculated
TBW (Total Bytes Written) is the rated endurance of a drive, expressed as the total volume of data that can be written over its entire service life under warranty conditions. The vendor states this figure in the documentation for a specific model, and it is not universal: a consumer SATA SSD usually has noticeably lower endurance than a server-grade NVMe drive of the same capacity.
Alongside TBW, vendors often quote DWPD (Drive Writes Per Day) — how many times a day the whole drive can be rewritten without exceeding the warranty endurance over the full warranty period. It is the same figure recalculated differently, and it is convenient when comparing drives of different capacity from different vendors.
Types of SSD wear
Drive wear is not a single process but several parallel factors worth telling apart when reading SMART attributes.
- NAND cell wear — the physical exhaustion of write-erase cycles in flash memory, growing with the volume of data written.
- Controller degradation — failures in the drive's control chip that are not directly tied to the amount of writes.
- Lack of spare blocks — exhaustion of the reserve of cells for wear-leveling, which speeds up visible wear at the same write volume.
- NVMe overheating — controller throttling under high temperature, which looks like a performance drop rather than wear.
Comparing ways to check drive endurance
Different tools suit diagnostics, and they complement each other rather than replace one another. The differences between NVMe, SATA SSD, and HDD in servers are covered in detail in the article about server disk types.
| Method | What it shows | Where it is available | When to use it |
|---|---|---|---|
| smartctl | Percentage Used and a counter of bytes written | console, Linux and Windows | basic check on any server |
| nvme-cli | health attributes specific to NVMe | console, NVMe drives only | detailed NVMe diagnostics |
| provider panel | aggregated alerts about drive health | hosting web interface | when console access is unavailable |
| load monitoring | indirect signs through IOPS and latency | external monitoring systems | early detection of anomalies |
An overview of utilities for day-to-day VDS monitoring, including CPU, memory, and disk, is available in the article about monitoring VDS resources.
How to check the actual wear in the console
The commands differ for NVMe and SATA SSD drives, but the point is the same: find the percentage of endurance already used.
# Checking SMART attributes of an NVMe drive
sudo nvme smart-log /dev/nvme0
# Checking a SATA SSD through smartctl
sudo smartctl -a /dev/sda | grep -i wearFor NVMe drives the key field in the output is percentage_used: while it stays below 100, the vendor still guarantees the drive's endurance, and further operation is possible but no longer under warranty. For SATA SSD a similar meaning is carried by a SMART attribute named something like Wear_Leveling_Count or Media_Wearout_Indicator — the exact name depends on the controller vendor.
What to do when endurance is running out
- Read the SMART attributes and record the current value of Percentage Used or a similar attribute to track the trend over time.
- Compare the rate of wear growth with the nature of the workload: databases and logs write to the disk noticeably more actively than static site files.
- Move the data with the heaviest write activity — logs, temporary tables, cache — to a separate partition or disk if that is possible.
- Set up backups with integrity checks instead of waiting for the moment when the drive fails without warning.
- Plan the drive replacement in advance, once wear approaches the rated limit, rather than after the first read errors appear.
To evaluate drive performance under load, not only its wear, the tests from the article about server benchmarks with fio and sysbench come in handy: high wear combined with growing latency is a signal to prepare a replacement ahead of schedule.
Frequently asked questions
How much write volume does an ordinary SSD withstand?
The exact figure depends on the model and is listed in the vendor's specification: consumer SATA SSDs usually have noticeably lower endurance than server-grade NVMe drives of the same capacity. The exact value can only be found in the documentation for a specific model, not from general estimates found online.
Why does an SSD show 100% load when there is little data on it?
A 100% load in monitoring is the saturation of the input-output request queue, not how full the drive is with data. The most common cause is a large number of small random write or read operations that the queue cannot keep up with, not the physical volume of stored files.
Can the drive still be used after reaching 100% Percentage Used?
Technically the drive usually keeps working after reaching the rated limit, but the vendor no longer guarantees data integrity for further writes. For production workloads such a drive should be replaced in advance, leaving it at most for secondary tasks.