Streaming Multiprocessor
NVIDIA's core building block: a cluster of CUDA cores, Tensor Cores, schedulers, registers and on-chip SRAM. AMD's equivalent is a Compute Unit (CU).
A GPU die is partitioned into Streaming Multiprocessors on NVIDIA, or Compute Units on AMD. Each SM bundles CUDA cores, Tensor Cores, register files, warp schedulers, and a slice of fast on-chip SRAM (shared memory / L1).
That organization is what makes a GPU a GPU: not a few fast cores, but a sea of simple ones running thousands of threads in parallel under the SIMT model.
In TensorLight’s matrix-multiply walkthrough, the output is cut into tiles and each tile is handed to its own Streaming Multiprocessor so they compute at once.
TensorLight glossary · Streaming Multiprocessor · Tensor Core · tile · HBM · also called Streaming Multiprocessors, SM