A benchmark number means nothing on its own. In this lab you write one small portable benchmark, run it with uv on the DGX and on a second device such as a Jetson Thor or a laptop, then merge the results into one tidy table and compare them. You measure CPU compute, memory bandwidth, interpreter speed, memory capacity, and sustained behavior under load, and you build the figures in the same house style as the figures lab: a per-device compute chart, a speedup table, a memory-capacity comparison, and a thermal-throttling plot.

It puts the two data labs to work on real hardware: the reproducible method from the data-collection lab and the publication figures from the figures lab, now used to compare very different machines in the same NVIDIA edge family.

Click Launch in JupyterHub above to open this lab. It pulls the latest version into your ~/EdgeNotebook and opens the notebook, ready to work through top to bottom. The first time, sign in with your class account.