NAS drive dataset

Cite: NASdisks (2026), NAS Drive Dataset, Zenodo. doi:10.5281/zenodo.21225049

An original dataset of NAS and enterprise hard drives: full specs, CMR vs SMR classification, and annualized failure rates derived from Backblaze Drive Stats. 273 drives.

The core of this dataset - the CMR/SMR classification and drive specs - is our own original research, compiled from manufacturer datasheets and model-number analysis. Manufacturers do not publish a unified CMR/SMR list; we built one. The failure-rate column enriches it with rates derived from Backblaze's raw Drive Stats data (our computation, their underlying counts).

Maintained at nasdisks.com, where the same data powers a NAS drive comparison across seven Amazon regions.

These files are a dated snapshot and may be out of date. The current, always-up-to-date dataset lives at nasdisks.com/data/ - download the latest there, or call the CORS-enabled endpoints below. Each file in data/ carries its export date in the filename so you always know how old your copy is.

What's inside

Price trends: price-trends/ - two charts of the NAS drive market over time: the monthly median $/TB (consumer vs enterprise, the storage price crunch), and how often each drive goes on sale and how deep. Computed data plus the charts.

Quick start

The files in data/ are ready to use as-is. If you'd rather hit a live endpoint (CORS-enabled, cached, no key, no rate limit):

# JSON: { source, license, attribution, count, drives: [...] }
curl -s https://www.nasdisks.com/data/drives.json | jq '.drives[0]'

# Flat CSV
curl -s https://www.nasdisks.com/data/drives.csv

Columns (drives-2026-09-04.csv / drives-2026-09-04.json)

Methodology

• CMR vs SMR is our original classification, built from manufacturer datasheets and model-number decoding. Manufacturers ship SMR drives into NAS lines without labelling them as such; this dataset exists because there was no single authoritative per-model reference. The confidence column in cmr-reference-2026-09-04.csv flags certainty per row.
• Noise and helium are read from manufacturer datasheets (idle/seek A-weighted sound power; helium-seal designation). Our compilation; licensed CC BY 4.0 alongside the other specs.
• Failure rates are computed from Backblaze's raw quarterly Drive Stats: we aggregate drive-days and failure counts into a single annualized rate over a rolling window through Q2 2026 (full-year 2025 plus the Q1 and Q2 2026 quarters), which keeps per-model samples large, and validate against Backblaze's published fleet figures (~1.41% over that window). The computation and methodology are ours; the underlying counts are Backblaze's (see licence section). Models Backblaze doesn't track are left null rather than guessed.
• series-reliability-2026-09-04.csv gives manufacturer-rated context (warranty, workload TB/year, MTBF) for lines Backblaze doesn't run.

Licence & attribution

Short version: the specs and CMR/SMR data are fully open (CC BY 4.0). The failure-rate fields are free to use but must credit Backblaze and cannot be resold on their own.

1. Drive specs, CMR/SMR classification, and acoustics - CC BY 4.0

Our original work (every field except the failure-rate ones). Licensed under Creative Commons Attribution 4.0 (CC BY 4.0); full text in LICENSE.

• You can: copy, modify, redistribute and build on it - including commercially.
• You must: credit the source, e.g.

  Drive specs & CMR/SMR by nasdisks.com, CC BY 4.0
  

  A link back to https://www.nasdisks.com/ is enough.

2. Failure-rate data - Backblaze's terms (NOT CC BY 4.0)

The fields afr_pct, reliability_drive_count, reliability_drive_days, reliability_failures (in drives-2026-09-04.csv / drives-2026-09-04.json) and all of backblaze-afr-2026-09-04.csv are derived from Backblaze Drive Stats. The computation is ours; the underlying counts are Backblaze's.

• You can: use it for free, and build (and even sell) derivative works from it.
• You must: credit Backblaze as the source.
• You can't: sell the raw data itself.
• Backblaze does not endorse this project. Details in NOTICE.

Updates

This repo is a point-in-time snapshot; the date in each filename is its export date. For current data use nasdisks.com/data/ or the endpoints above.

Changelog

• 2026-09-04 - v1.0.4 - Refreshed Backblaze failure rates to a rolling window through Q2 2026 (full-year 2025 plus the Q1 and Q2 2026 quarters, ~1.41% fleet), computed from Backblaze's raw Q2 2026 Drive Stats; 273 drives (167 HDD + 106 SSD), 25 with a failure rate. Added recent PCIe 5.0 and NAS-cache SSDs (Crucial P510, Corsair MP700 Elite, Synology SNV3410/SNV3510); sub-TB drive capacities are now reported correctly (were 0). NAS enclosures are no longer included.
• 2026-07-16 - v1.0.3 - Refreshed Backblaze failure rates to a rolling window through Q1 2026 (full-year 2025 plus the Q1 2026 quarter, ~1.34% fleet); 208 drives, 25 with a failure rate.
• 2026-06-26 - v1.0.2 - Refreshed to the current live catalog (149 drives); minor catalog reconciliation since v1.0.1.
• 2026-06-21 - v1.0.1 - Added per-drive noise (idle/seek A-weighted sound power, dB) and helium fields. Corrected WD40EZAX / WD60EZAX to CMR (newer EZAX is CMR; only older EZAZ is SMR).
• 2026-06-16 - v1.0.0 - Initial public dataset: full specs, CMR/SMR classification, full-year 2025 Backblaze annualized failure rates.