Alibaba’s latest breakthrough in industrial software has really “broken the bottleneck”—it’s got GPU acceleration plus long-tail optimization, and it finally makes it possible to run linear programming on systems with on the order of billions of variables.

CryptoWorld News: Alibaba's DAMO Academy announced today the official release of its "MindOpt" solver (mindopt) GPU version, fully leveraging GPU parallel acceleration features and introducing new algorithms to break through the "long tail effect" challenge. Testing on approximately 2,000 general cases shows that MindOpt can stably solve over 99% of problem types to high precision, and even support traditionally "unsolvable" linear programming problems with billions of variables. The solver is hailed as "the core of industrial software," responsible for complex calculations in critical fields such as power dispatch, flight scheduling, high-end manufacturing, and financial management. Traditional linear programming solvers are CPU-based, relying on complex matrix decomposition calculations. As problem sizes grow, memory requirements explode, and limited parallelism in traditional algorithms leads to problems that cannot converge in hours or even crash outright.