Classical Baseline · Qubitwerk
We simulate quantum circuits on an Intel Core i9 with NVIDIA RTX 3090. See measured runtimes, accuracy checks and the limits of the workloads we tested.
Illustration · timings from 21 Oct 2026, follow-up requests.
Recorded timings · 21 October 2026
1.62×as fast in this selected test
The lights show the timing ratio in slow motion. We replay recorded timings; your click does not start a new calculation. The values remain readable without animation.
Follow-up requests · double precision · measurement conditions →To support a quantum-advantage claim, a quantum method must outperform a suitable classical baseline. A faster simulator alone does not establish that advantage.
XPRIZE Quantum Applications asks teams to compare with the best available classical approach. Our reference runs can support part of that work. We are not an official competition evaluator.
Source: competition requirements ↗Selected tests · 21 October 2026
We selected these examples after the measurement series. They show advantages for these circuits and readout queries, not a general lead over Aer.
1.62 × as fast in this test
We checked 64 selected amplitudes, the norm and 22 Z expectation values. Not a full-state comparison.
Largest checked amplitude deviation: 2.638e-18. Acceptance threshold: 1e-12.
1.51 × as fast in this test
We checked the full state of 512 amplitudes, the norm and 9 Z expectation values, also against the original circuit.
Largest checked amplitude deviation: 9.045e-15. Acceptance threshold: 1e-12.
2.39 × as fast in this test
We checked the full state of 64 amplitudes, the norm and 6 Z expectation values, also against the original circuit.
Largest checked amplitude deviation: 3.353e-15. Acceptance threshold: 1e-12.
Intel Core i9-10900K and NVIDIA RTX 3090. Both simulators used the same circuit, the same readout queries and complex128. Reference: Qiskit Aer 0.18.0 with Qiskit 2.6.0, GPU statevector with cuStateVec.
Timings are medians of four follow-up requests per simulator and example across two separate processes. Both simulators recomputed every request. These timings exclude process startup, the public queue, internet and animation; they are not isolated GPU kernel timings either.
Full-state checks are separate from the timings above. Four follow-up values form a small sample. We keep our implementation and the full measurement series private; public rerun scripts are not available at present.
All published conditions and checks →Understand memory
16 GiBState data at complex128
Calculated: 16 bytes × 2 to the qubit count. Each block represents 8 GiB; smaller values fill only part of the first block. This is the statevector alone, excluding working memory. The graphic is not a utilisation monitor.
Separate capacity test · 20 September 2026
We computed the statevector in double precision without truncating amplitudes, storing it in blocks on the SSD. The GPU processed those blocks. This is a capacity test, not a runtime advantage over Aer at 35 qubits.
We compared the smaller 30-qubit circuit against Aer. At 35 qubits we checked the norm, distribution and reverse evolution. These checks are not a full independent reference for the 35-qubit state.
35-qubit evidence and checks →Your circuit
Email OpenQASM 2 or 3, or a short description, to info@coolblack.gmbh. Include the output you need and the required accuracy. Subject: “Classical baseline”.
We first check whether our current simulator supports your circuit and requested readout. For a supported run, we return runtime, memory use and appropriate accuracy checks. If we cannot run it, we explain the limit. We handle the exchange by email.
Send us your circuit ↗Start with material you have permission to share. An enquiry does not mean we support arbitrary OpenQASM programs or sizes.
For a full statevector, each additional qubit doubles the memory requirement. The 35-qubit circuit was an SSD-backed experiment with a long runtime, not a general promise for arbitrary 35-qubit programs. Runtime and memory depend on the circuit, readout and method.
A comparison with Aer alone does not cover the best classical methods for every application. A credible quantum-advantage verification must consider the problem and suitable specialised methods. Our current measurements establish neither quantum advantage nor a world record.
XPRIZE is a trademark of the XPRIZE Foundation. Qubitwerk is not affiliated with, sponsored or endorsed by XPRIZE.