Micron rolls out 276-layer SSD trio for speed, scale, and stability

Original Article ↗ Hacker News Discussion ↗

SSD & HDD Pricing Floors and Market Structure

  • Multiple commenters report a clear “floor” in consumer SSD prices (e.g., 2TB drives not dropping over 18+ months) and similarly for HDDs, where small-capacity drives rarely fall below ~$50–80.
  • Debate over cost structure:
    • One view: the real floor is controller + PCB + casing; NAND is the variable part.
    • Counterview: controller/PCB are only cents; NAND and DRAM are the real cost drivers, but retail prices are mostly “what the market will bear,” not tightly tied to chip cost.
  • Several argue SSD pricing looks like an oligopoly/triopoly with brand premiums and potential quiet coordination, analogous to past DRAM price-fixing; others emphasize demand/supply cycles and overcapacity hangovers.
  • HDD $/TB has also flattened; current fluctuations for both HDD and SSD are framed as demand/supply dynamics more than manufacturing breakthroughs.

Capacity Stagnation and Form Factors

  • Frustration that consumer capacities have stalled around 2–4TB for years, with 8TB NVMe only recently appearing and at high prices.
  • Larger capacities (12–30TB+) exist mainly in server-oriented U.2/U.3/E1.S formats; they are expensive, hot, and require airflow, making them awkward for home/NAS use.
  • Disagreement on “no consumer market for 4TB+”:
    • One side: most people rely on cloud/phones, don’t need >2TB; gamers can uninstall/move games to HDD.
    • Other side: modern games and media workflows easily justify 4–8TB; some want to retire noisy HDD NAS but SSD prices make that unrealistic.
  • M.2 size/power limits and lack of fast 2.5" adoption in consumer gear further constrain practical large SSD options.

3D NAND Technology and Layer Counts

  • Discussion of how 3D NAND is fabricated: many layers are deposited at once, then vertical holes etched and filled; layers are largely identical and accessed via a “staircase” at the edge.
  • High layer counts improve density and lower variable cost per bit but do not eliminate fixed non-NAND costs.

Endurance, Retention, and Enterprise vs Consumer

  • Concern that marketing for the new 276‑layer drives omits data retention, while quoting very high P/E cycle numbers (6.6k–11k) that look more like MLC than typical TLC.
  • Explanation that endurance and retention trade off: enterprise SSDs often achieve higher DWPD by accepting lower guaranteed unpowered retention (e.g., months instead of a year at end‑of‑life).
  • Commenters note:
    • Worn flash has significantly worse retention than fresh flash; some tests show even once‑programmed TLC can exhibit retention issues over long unpowered periods.
    • Highly publicized “SSD endurance” torture tests are seen as misleading, because they measure failure at extremely low retention times.
    • The same underlying NAND can be sold as consumer vs enterprise largely via different warranty and rating points, plus extra enterprise features (power-loss protection, firmware QA, consistent throughput).

Workloads, DWPD, and Practical Reliability

  • Comparison: enterprise HDDs often tolerate ~0.05–0.1 drive writes per day, while enterprise NVMe commonly advertises 1–3 DWPD—orders of magnitude higher write tolerance.
  • Effective endurance improves for SSDs under sequential or zoned-write workloads, since specs are based on worst-case small random writes.
  • Some argue large single SSDs are preferable to striped arrays (RAID 0/others) due to fewer components and lower power per TB; others counter that distributing chips across multiple PCBs doesn’t fundamentally change chip-level risk.