Definition
Plain language
How much work a system gets done per second, regardless of whether it's the useful kind.
As stated in the literature
The raw rate of completed operations or tokens per unit time; contrasted with goodput, which counts only work completing usefully within deadlines and can collapse even while throughput stays high.
Why it matters: It measures raw productivity but can look healthy even when little of the work actually helps, which is why useful-work metrics matter too.
For example, a server might churn out thousands of tokens per second even while many of those responses arrive too late to be useful.
Heard on the show
“GPU utilization pinned in the high nineties, tokens streaming out at full bore, throughput numbers that would have looked impossible two years ago.”Episode 016 — Why Your Coding Agent Stalls While the GPU Runs Hot