On May 28, the official Weibo account of Alibaba DAMO Academy announced that its self-developed 'MindOpt' solver has officially launched a GPU version. This version fully leverages the parallel acceleration capabilities of GPUs and introduces new algorithms, effectively addressing the challenge of the 'long-tail effect.' Tests show that among approximately 2,000 general test cases, the MindOpt solver can stably solve over 99% of the problems to high precision and supports solving traditionally difficult linear programming problems with hundreds of millions of variables. Solvers are the 'heart' of critical industrial software, widely used in fields such as power dispatch, flight scheduling, high-end manufacturing, and financial management. Traditional solvers, designed based on CPUs, rely on complex matrix decomposition. As the problem size increases, memory demands surge, and parallelism is limited, often leading to prolonged non-convergence or crashes.