WORLDSUPERCOMPUTING

WORLD SUPERCOMPUTING

Scaling Lab

Explore an Amdahl-style model of runtime, efficiency and resource use.

Fixed problem size
Added to the runtime fraction at the selected size; set from measurement. At one node it is zero.

Hypothetical starting inputs. Nodes are equivalent processing units in this simplified model; no hardware performance is assumed.

The model

Runtime fraction = serial fraction + (1 − serial fraction) ÷ nodes + extra overhead fraction. Extra overhead is zero at one node. Runtime = baseline × runtime fraction. Speedup = 1 ÷ runtime fraction. Efficiency = speedup ÷ nodes. Node-hours = runtime × nodes.

The idealized comparison uses zero extra overhead. Real communication, memory, I/O and queueing require measurement. This model is not a machine benchmark.

PRIMARY SOURCES · CHECKED 9 NOVEMBER 2026

NERSC: parallelism →Living documentation; checked 2026-11-09 · Strong scaling fixes total problem size; weak scaling fixes work per processor.