Omnisolver-bruteforce plugin delivers 8x GPU scaling for exact QUBO solves up to N=60

The update packages the 2021 CPC single-GPU kernel as an Omnisolver plugin and layers a Ray dispatcher that partitions the 2^N search into 2^k fixed-variable subproblems across hosts. Numerical stabilization via compensated summation and periodic re-anchoring keeps float32 errors below the ground-state threshold for N>=40. On random dense instances the measured wall-clock time follows the expected doubling rule, confirming the distributed layer saturates the aggregate GPU throughput without communication bottlenecks.

Existing coverage treats the work as a narrow software release. In reality it supplies the first publicly available, hardware-agnostic oracle capable of certifying optimality for the Simulated Bifurcation Machine and other quantum-inspired heuristics on instances large enough to expose their approximation gaps. This directly affects benchmarking pipelines used by logistics, finance, and materials-discovery teams that currently rely on unverifiable heuristic outputs.

The chief limitation remains the exponential wall: even 8x H100 hardware reaches only N=60. Extending the same interface to tensor-network contraction or hybrid quantum-classical sampling will be required before the tool can serve as a routine ground-truth generator beyond the current size frontier.

Next milestones include integration with NVIDIA's cuQuantum SDK and release of a public benchmark suite of N=55-65 instances with certified optima.