cpu_tuple_cost
No restart — settable per sessioncost factor (per row processed)
The planner's assumed CPU cost of processing one row.
What it does
cpu_tuple_cost (default 0.01) is how much the planner charges for handling each row, relative to seq_page_cost = 1.0. Together with the page costs it shapes whether the planner prefers plans that touch fewer rows versus fewer pages.
How to tune it
Rarely worth changing in isolation. The cost *ratios* (random_page_cost vs seq_page_cost, page costs vs CPU costs) are what shape plans, so adjust those one or two first. On a heavily-cached, CPU-bound system some operators raise cpu_tuple_cost — or lower the page costs together — so the model reflects that CPU, not I/O, is the bottleneck. Always validate with before/after EXPLAIN ANALYZE on representative queries.