GPUs and AI accelerators

Why GPUs won AI, what TPUs do differently, memory bandwidth and the economics of AI chips.

5 lessons planned · none written yet

  1. 1CPU versus GPUNot written yetLatency versus throughput.
  2. 2Matrix multiply hardwareNot written yetTensor cores and systolic arrays.
  3. 3The memory wallNot written yetHBM and bandwidth.
  4. 4InterconnectNot written yetScaling across thousands of chips.
  5. 5EconomicsNot written yetWho makes money and why.