Interactive Visualizations
Interactive charts and calculators for the research notes. Click any title to go to the full article.
GPU MatMul — prefill vs decode
Interactive GPU MatMul visualizer — toggle prefill vs decode, zoom into SMs and tensor cores, see the 64×64 systolic wavefront. Requires JavaScript.
B200 die map
Interactive B200 die map — hover over regions to explore transistor breakdown. Requires JavaScript.
memory hierarchy energy diagram
Energy cost of data movement (Horowitz diagram) for B200. Requires JavaScript.
KV cache memory calculator
KV cache memory estimator — adjust model params, sequence length, batch size. Requires JavaScript.
roofline model
Interactive roofline plot — B200 vs H100, shows compute vs bandwidth ceilings. Requires JavaScript.
inference cost estimator
Inference cost calculator — estimate $/M tokens for different models and hardware. Requires JavaScript.
systolic array efficiency
Systolic array efficiency by size — Roune’s scaling insight. Requires JavaScript.