Definition
Operational intensity is the metric used as the horizontal axis of the roofline model: the number of operations performed per byte of DRAM (off-chip main-memory) traffic. It was defined precisely in Williams, Waterman, and Patterson's 2009 roofline paper to make performance modeling concrete and measurable on real hardware.
How it differs from arithmetic intensity
In casual use the terms are interchangeable, but the distinction is deliberate. Arithmetic intensity is the general idea of work-per-byte. Operational intensity pins down exactly which bytes count — only traffic that crosses the chip boundary to and from DRAM — and which operations count for a given study. Data served from on-chip cache or reused inside registers does not add to the byte total, which is why effective tiling and caching raise a kernel's operational intensity and push it rightward on the roofline toward the compute-bound region.
Why the off-chip framing matters
Off-chip memory is the slowest, most power-hungry data path in the system, so it is the bottleneck the roofline cares about. By measuring operations per DRAM byte, operational intensity directly answers the practical question: "is this workload limited by how fast the chip can compute, or by how fast it can feed itself from main memory?" Raising it — through fusion, tiling, and reuse — is the central lever for AI-kernel optimization.
This is the metric you plot to find where a workload lands on the curve. See the roofline model and the broader concept of arithmetic intensity.
In Simple Terms
Operational intensity is the metric used as the horizontal axis of the roofline model: the number of operations performed per byte of DRAM (off-chip main-memory)…