NEROX Blog
Benchmark results, algorithm analysis, and implementation notes on GPU-accelerated QUBO optimization and quantum-inspired methods.
NEROX v0.9.3 ships the Parity Twine Chain ansatz — fused SWAP·ZZ bricks that collapse dense-QUBO phase-separator two-qubit depth from O(n²) to O(n). At n=120 the circuit is 2.5% of the baseline depth. Ported from Monbroussou et al. (arXiv:2505.17944); the same parity-propagation family powered IBM + ParityQC's 52-qubit QFT benchmark on Heron r3.
Architecture and solver stack overview: massively parallel GPU annealing across 4,096 independent Markov chains, hybrid decomposition for large instances, QAOA via cuStateVec circuit simulation, Tabu Search with domain-specific move operators, and VQE for Hamiltonian ground-state problems.
Controlled comparison of GPU-SA, QAOA (p=4), Tabu Search, and CPU-SA on TSPLIB (eil51, pr1002), Biq Mac MaxCut (G01, G22), OR-Library FT10 job shop, and a portfolio QUBO. Fixed hardware (A100), default solver configs, 5 trials per instance, median reported.
GPU-native quantum-inspired optimization for combinatorial problems at any scale.