📖 Book · hpc · advanced

🟩 The GPU / CUDA execution model

SIMT, the memory hierarchy, coalescing, shared-memory tiling, warp shuffles, atomics & streams — in complete, compile-verified CUDA.

A deep, code-level tour of how GPUs execute: the SIMT model and a full SAXPY program, grid-stride and 2D indexing, the memory hierarchy with a shared-memory tiled GEMM, warp coalescing / divergence / bank conflicts, reductions with warp-shuffle intrinsics, atomics and streams, and occupancy. Every kernel is compile-verified with nvcc.

7 chapters▶ runnable code✦ in-depth animation

What this book covers

  1. 01SIMT: one kernel, thousands of threads
  2. 02Indexing: grid-stride loops & 2D grids
  3. 03The memory hierarchy is the whole game
  4. 04Warps: coalescing, divergence & bank conflicts
  5. 05Reductions: shared memory → warp shuffle
  6. 06Atomics, synchronization & streams
  7. 07Occupancy & when the GPU actually wins

Read it in full — runnable C++ + step-through animations

Every chapter of The GPU / CUDA execution model ships complete, editable code you compile and run in the browser, plus the same in-depth animations as the lessons.