Lab

Interactive tools

The best way to explain how something works is to build a tool that makes you prove it. Two below — a third one's in the corner of every page (look for "Ask about me").

AllReduce Visualizer
Ring vs. Recursive Doubling — watch GPU collective communication algorithms move data step by step, with a live cost model.

Open →
Inference Throughput Estimator
Prefill (compute-bound) and decode (memory-bandwidth-bound) throughput for Llama-3 on H100 / H200 / B200 — every assumption adjustable.

Open →