Scaling Decision Optimization to 100 Million Variables and Beyond with mPDLP in NVIDIA cuOpt
NVIDIA cuOpt scales decision optimization to 100M+ variables via GPU-accelerated mPDLP
NVIDIA announced mPDLP support in cuOpt, enabling GPU-accelerated linear programming at 100 million+ variable scale for supply chain and energy grid problems. The technique delivers 10x+ speedups over CPU solvers. While technically impressive, this is an engineering/HPC story rather than a core AI/ML signal relevant to conference audiences.