Bad Apple running on a Trainium chip: real kernel instruction trace executes every frame

daniel_c0deb0t · x · 2026-09-05

The author shared a real profiling trace of instructions and DMAs in a kernel running on an AWS Trainium NeuronCore: one engine emits a steady stream of instructions that other engines use like a clock, and the kernel "executes" all frames of Bad Apple one after another.

Related event: Bad Apple Runs on AWS Trainium Chip via Raw Kernel Instructions(4 posts)→

Original post →

More from Fun

Fun channel →