Blackwell Thread Hierarchy
click a level — thread → warp → warpgroup → CTA → cluster → grid
GRID — one kernel launch
CLUSTER — CTAs across SMs (DSMEM)
CTA — thread block · one SM
WARPGROUP — 4 warps · 128 threads
WARP — 32 threads (SIMT)
warp 1
warp 2
warp 3
CTA 1
…