The scene is 1180 px wide: swipe it sideways.
SM · idle
cycle 0
Pick a scenario or open the kernel editor to launch a kernel.
The scene is 1180 px wide: swipe it sideways.
A Python-subset kernel — every one of the 32 lanes in a warp runs it for real. Use tid (global thread index) and the input arrays by name. Supported: if/elif/else, for i in range(), while, def, lists & dicts, arithmetic, and CUDA intrinsics (cuda.grid(), cuda.shared.array(), cuda.syncthreads(), cuda.atomic.add()).
The task size is n in the inputs: the distributor reads the element count, the FLOP and byte budget per element, the access pattern and the branch structure from the kernel, and routes it to the cheaper engine. There is no launch geometry here: on GB10 the task is placed on a CPU or a GPU, not on a grid.