Skip to content
Hi, Bot

NVMe SSD

Flash storage: no moving parts, keeps data when the power's off.

How big?About the size of a stick of gum.

Real: What it looks like.

Keys: arrows rotate · + / − zoom · 0 reset · 1–4 views · S signal · T tour · L labels · Space spin. Models are stylised and built from code: proportions are honest, details are simplified.

What it is

Flash memory traps electrons in cells stacked in 3D, with hundreds of layers per die and several dies per package. Trapped charge means the data survives with the power off.

A controller chip spreads reads and writes across all the dies at once and works around worn-out cells. A small DRAM cache keeps the map of where everything is.

NVMe is the protocol: the drive talks to the CPU directly over PCIe lanes, which is why it's so much faster than older SATA drives.

Why AI needs it

Each server keeps fast local storage for the training data it's working through right now and for checkpoint saves, so the GPUs never wait on the network.

Every labeled part

  1. 1

    Controller

    The drive's own processor. It manages wear, errors and parallel access.

  2. 2

    DRAM cache

    Holds the lookup table of which cell holds which data.

  3. 3

    NAND flash package

    X-ray it to see the stacked dies inside.

  4. 4

    3D cells

    Each die has hundreds of layers of memory cells built straight up.

  5. 5

    M.2 connector

    Four PCIe lanes run straight to the CPU.

Try it · concept lab

How far away is your data?

Switch to human time to feel the gaps.

  • Register · inside the core · bytes1.0 s
  • On-chip cache (SRAM) · on the die · MBs10 s
  • HBM / DRAM · beside the chip · GBs–TBs6 min
  • NVMe SSD · in the server · TBs3 days
  • Hard drive · storage rack · PBs193 days
  • Data lake over network · across the building · PBs–EBs3.2 years

Rough, typical figures (log scale). Every step down is bigger and cheaper, and much slower.

Big idea: Fast memory is small and slow memory is big. AI performance is mostly about keeping the right numbers close to the cores.

Swap it: other ways to do the same job

  • Hard drive

    Much cheaper per terabyte, much slower, with moving parts. Used for bulk data-lake storage.

  • Data-center SSD (U.2 / E3.S)

    Same flash in a bigger, hot-swappable case. More capacity and better cooling.

  • Tape

    Cheapest and longest-lasting archive. Too slow for anything but cold backups.

Talk about it

  1. Q1

    An SSD has no moving parts. Why might that matter for something that runs all day and all night?

  2. Q2

    A training computer saves its progress every so often. What do you save as you go, and what happens if you forget?

For grown-ups: there are no right answers here. Ask a question, then ask "why do you think that?" The reasons matter more than the answer.

Printable question sheet (PDF)

Connected exhibits