Roofline benchmark: matmuls and elementwise operations on your GPU

Measure achieved FLOP/s and bandwidth for operations of different arithmetic intensity and compare them with a roofline drawn from your GPU’s datasheet.

The exercise happens on Self-directed (PyTorch benchmark utilities). Canonical link: https://docs.pytorch.org/tutorials/recipes/recipes/benchmark.html. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Needs an NVIDIA GPU (local or rented). Prepares the compute-bound versus memory-bound reasoning used in every AI-infrastructure interview.