The NISQ Era & Quantum Volume
How to actually compare today's noisy, imperfect quantum computers to each other
By the end of this topic you'll be able to
NISQ — Noisy Intermediate-Scale Quantum — is the term John Preskill coined in 2018 to describe the current era of quantum hardware honestly: devices with a meaningful number of qubits, tens to low thousands, but without full fault tolerance.
2^5
Quantum Volume ≈ 32
This is a simplified illustration, not the official QV protocol — but it shows the real effect: push the error slider up while qubit count stays fixed, and the usable depth (and QV) collapses fast, because fidelity drops exponentially with both circuit width and depth together.
In a NISQ device, every gate carries a nontrivial error rate, so only circuits of modest depth can be run before accumulated noise overwhelms the signal being computed. Shor's algorithm at cryptographically relevant scale is firmly beyond what NISQ hardware can do; VQE and QAOA, by contrast, were specifically designed to extract something useful within exactly these constraints.
Given that reality, comparing two different NISQ devices to each other turns out to be a genuinely subtle question. Raw qubit count alone is a famously misleading metric: a 100-qubit device with high gate error rates and poor qubit-to-qubit connectivity can run meaningfully less useful computation than a 50-qubit device with lower error rates and better connectivity.
Quantum Volume, introduced by IBM, is one widely used answer to this problem. It is measured by running randomized 'square' circuits — n qubits deep and n qubits wide, for increasing values of n — designed to stress-test a device's qubits, gates, and connectivity simultaneously, and finding the largest n the device can run successfully at high enough fidelity.
The result is reported as 2ⁿ, and it captures a device's genuinely usable computational power far better than qubit count alone does: a device's Quantum Volume can, and often does, fail to double even when its raw qubit count doubles, if error rates or connectivity haven't kept pace.
Other benchmarks round out the modern toolkit for comparing NISQ hardware honestly — randomized benchmarking, for characterizing individual gate error rates in isolation, and application-specific benchmarks that measure performance directly on real algorithmic workloads rather than on an abstract random circuit.
Try It Yourself
A device passes the n=6 square random-circuit test (6 qubits, 6 layers deep) with high fidelity, but fails the n=7 test. What Quantum Volume is reported?
- 1By definition, Quantum Volume is reported as 2ⁿ, where n is the largest square (n-qubit-by-n-layer) random circuit the device can run successfully at high fidelity.
- 2Here, the largest passing n is 6.
- 3So Quantum Volume = 2⁶ = 64.
Quantum Volume = 64 — meaning the device has enough combined qubit count, gate fidelity, and connectivity to reliably run a genuinely demanding random circuit 6 qubits wide and 6 layers deep, not merely 'has 6 qubits.'
Reference
| NISQ | Term coined by John Preskill, 2018 | |
| Quantum Volume | Largest square (n×n) random circuit run with high fidelity | |
| Randomized benchmarking | Characterizes individual gate fidelities |
Quick Check
Why is raw qubit count alone a misleading way to compare two NISQ devices?
What does Quantum Volume measure?