Are GPUs the final form of AI compute?

An examination of why AI uses GPUs, when specialized chips offer advantages, and how memory, manufacturing, software and cost constrain alternatives ranging from TPUs to optical, neuromorphic, thermodynamic, biological and quantum computing.

4 sections ยท 16 lessons

Course outline

AI computing systems

  1. What AI actually computes
  2. Why graphics chips suit neural networks
  3. Why advanced chips are difficult to make
  4. When faster math stops helping

Choosing accelerators

  1. When a TPU earns its place
  2. Other digital routes to efficiency
  3. Compare the cost of useful work
  4. Read a GPU replacement benchmark
  5. Choose hardware for a changing service

New physical approaches

  1. Calculating where weights are stored
  2. Light can connect or calculate
  3. When brain-inspired chips save work
  4. Computing with controlled randomness

Evidence and outlook

  1. What quantum progress does not prove
  2. What neurons playing Doom demonstrate
  3. Could new hardware change how AI learns?
  4. What would actually displace GPUs?
  5. Evaluate a GPU replacement headline

Start learning with Wondering