QPerfect recently participated in a quantum emulator benchmarking challenge organized by Quantinuum, a leading full-stack quantum computing company that uses emulation extensively in their R&D.
The result? — Our own emulator, MIMIQ, crushed it.
This challenge was designed to identify the best quantum circuit simulators for their research teams to use. Quantinuum asked their research teams to provide a set of application-oriented benchmark circuits, with varying difficulty levels, from 28 to 280 qubits and from 71 to more than 1 million multiqubit gates.
Benchmark Design
The benchmark study covered a broad spectrum of use cases including: Trotterized Hamiltonian simulations, UCCSD ansatz state preparation for chemistry, Multivariate state preparations, Quantum Monte Carlo integration, and Quantum error correction codes.
Results
We’re excited to share that MIMIQ had the fastest runtimes on almost all the challenge circuits, often 10 to 100 times faster than the others! MIMIQ was even able to solve some of the circuits in the “impossible” category.
Notably, MIMIQ was the only submission running solely on CPUs (for now), demonstrating that a well-optimized CPU-based emulator can outperform even the best GPU-based emulators for many tasks.
“MIMIQ is an impressively fast tensor network simulator that has pushed the boundaries of what I thought could be efficiently simulated on CPUs.” — Pablo Andres-Martinez, Sr. R&D Scientist, Quantinuum
MIMIQ’s Secret
- Low-level CPU implementation: We have squeezed every microsecond by optimizing memory usage, number of operations, and parallelization.
- MPS methods: Benefiting from our decade-long expertise in MPS methods, we have implemented the best and most versatile methods for compressing and applying gates to the MPS state.
- Circuit pre-optimization: We have developed several functions that compress and transform the circuit, reducing necessary resources before the circuit is even run.
- MPS parameter optimization: The performance of MPS can sensitively depend on various parameters and we have developed methods to optimize them.
- User expertise: Our extensive use of MIMIQ has taught us how to best apply it depending on the problem.
The full benchmark results are publicly available in the CQCL/public_tn_sim_challenge_results repository on GitHub. For a deeper dive into MIMIQ’s tensor-network methodology, see our peer-reviewed paper, Comparative Benchmarking of Utility-Scale Quantum Emulators (Leonteva, Whitlock et al., ACM Trans. Quantum Computing 2025).