How NVIDIA GPUs and Google TPUs Work Under the Hood

How do AI chips actually work? In under 3 minutes, a stickman explains what’s inside a GPU (cores, warps, tensor cores, memory) and a Google TPU (the systolic array), and why they’re built differently. 0:00 How do AI chips think? 0:08 AI is just math 0:16 The CPU 0:22 The GPU 0:28 Inside the GPU: SMs and warps 0:37 Tensor cores 0:44 HBM memory 0:52 CUDA software 0:58 Meet the TPU 1:05 The systolic array 1:10 How data flows through a TPU 1:17 TPU pods 1:24 GPU vs TPU 1:35 Outro