NVIDIA cuOpt 推出 mPDLP 多 GPU 线性规划求解器
原标题：Scaling Decision Optimization to 100 Million Variables and Beyond with mPDLP in NVIDIA cuOpt
AI 摘要
NVIDIA 发布 cuOpt 中的 mPDLP 求解器，将线性规划问题分布到通过 NVLink 连接的多个 GPU 上，解决超大规模 LP 问题。相比单 GPU PDLP，每 GPU 峰值内存最多降低 6 倍，求解速度比一年前快近 10 倍，并在 zib03（超 1.04 亿非零元）基准上验证。该方案已用于 Kinaxis 的供应链规划与 PSR 的能源系统容量扩展模型。
正文节选
Supply chain problems are expanding across more SKUs, lanes, and constraints than ever before, while energy grids are balancing more distributed sources in real time. To meet these challenges, teams need to evaluate larger models and more uncertainty within a practical planning window. NVIDIA cuOpt GPU-accelerated decision optimization can already deliver speedups of more than 10x over CPU solvers on a single GPU for large-scale linear programming (LP) problems. However, today’s largest planning