Throughput

Also called: results per second, pipeline throughput

How many results a system produces per unit of time. A pipeline producing one result per clock cycle has throughput equal to its clock frequency.

Throughput answers "how much work comes out per second?" For a circuit that produces one result per clock cycle, throughput equals the clock frequency:

throughput = 1 / T = f results per second

It's different from latency, which is how long one item takes. A pipeline with long latency can still have high throughput, because many items are in progress at once, each in a different stage.

Pipelining raises throughput by shortening the clock period. With N equal stages and no overhead, throughput would rise N times. In practice each stage pays the register overhead tcq + tsu, and the period suits the slowest stage, so the gain is smaller.

Throughput only helps when there's a steady stream of work. For a single item, latency is all that matters.

Worked example

Example

Comparing throughput

Design X takes 20 ns per result, one at a time. Design Y is a 4-stage pipeline with a 6 ns clock.

1.
X: one result per 20 ns → 1 ÷ 20 ns = 50 million results per second.
2.
Y: once the pipeline is full, one result per 6 ns → 1 ÷ 6 ns ≈ 166.7 million per second.
3.
Y's throughput is 20 ÷ 6 ≈ 3.3 times higher, even though each item takes 4 × 6 = 24 ns to get through.

Common mistakes

Measuring throughput by how fast one item finishes. That's latency.
Assuming N stages give exactly N times the throughput. Overhead and unbalanced stages reduce the gain.

Practice Throughput

Interactive questions with instant feedback and a worked solution for every wrong answer.

Timing and Sequential Logic lesson full course

Practice exams

Learn it step by step

Throughput is taught in Timing and Sequential Logic and Basic CPU / Computer Architecture.

Timing and Sequential Logic →Basic CPU / Computer Architecture →

All terms