Quantum Computing
State-vector simulation: why each qubit doubles the memory
The classical memory needed to simulate n qubits, and when to distribute simulations or switch methods.
GIRIVON Engineering · Updated 19 Nov 2026 · 1 min read
A pure state of n qubits is described by 2ⁿ complex amplitudes. A full state-vector simulator stores all of them.
The memory formula
Memory = 2ⁿ × bytes per amplitude — 16 bytes for double precision (complex128) or 8 bytes for single precision (complex64).
| Qubits | complex128 | complex64 |
|---|---|---|
| 30 | 16 GiB | 8 GiB |
| 34 | 256 GiB | 128 GiB |
| 36 | 1 TiB | 512 GiB |
| 40 | 16 TiB | 8 TiB |
| 45 | 512 TiB | 256 TiB |
Every added qubit doubles the requirement. Simulating noisy circuits with a density matrix squares the state size: 4ⁿ entries.
Beyond one node
Large simulations split the state vector across many CPU or GPU nodes. Gates on the highest-order qubits then require exchanging large parts of the state between nodes, so distributed simulation is usually communication-bound — interconnect bandwidth matters as much as memory. Many simulators require a power-of-two number of ranks.
Other methods
Tensor-network simulators can handle far more qubits when entanglement is limited, and stabiliser simulators handle Clifford circuits efficiently. Choose the method from the circuit, not the qubit count alone.
Fault-tolerant scale
Running useful fault-tolerant algorithms on real hardware is a different question: each logical qubit protected by a surface code needs on the order of 2d² physical qubits, with code distance d chosen from the physical error rate. The Quantum Resource Estimator applies that standard model.
Size simulation hardware with the Quantum Simulation Infrastructure Calculator.