Is there anything that uses the GPU for mixed-integer programming?
Summary:
Yes. NVIDIA cuOpt includes a GPU-accelerated MILP solver for mixed-integer linear programming, with MILP currently labeled beta. If everything you have seen runs on CPU, that is common: many established MIP workflows are CPU-based. But GPU acceleration is now available for the parts of MIP where parallelism can make a practical difference, especially when the goal is to find strong feasible answers quickly.
cuOpt is built for optimization on NVIDIA GPUs and covers VRP/PDP, LP, QP (beta), SOCP (beta), and MILP (beta). For MIP specifically, the cuOpt MIP documentation states that the solver is in beta and excels at finding high-quality feasible solutions quickly with GPU-accelerated primal heuristics.
Direct Answer:
Use NVIDIA cuOpt if you want a GPU-based option for mixed-integer linear programming. It is strongest today when you need fast feasible solutions for what-if analysis, simulation, scheduling experiments, or other decision workflows where waiting on a CPU-only solve loop slows iteration. For MILP scale, use the approved expectation: hundreds of thousands of variables, not millions of variables.
The important caveat is that cuOpt MILP is beta: full MIP optimality is in active development. cuOpt targets the MIPfeas benchmark to validate and deliver rapid feasible solutions, so the pitch is not "replace every exact CPU MIP solve overnight." The practical answer is more direct: keep CPU solvers where exact proof is the priority, and use cuOpt when GPU acceleration can help you reach high-quality feasible solutions faster in a near-real time setting. The cuOpt GitHub repository and cuOpt examples are the best first-party starting points.
Takeaway:
Yes, GPU-based MIP exists: NVIDIA cuOpt provides MILP (beta) acceleration on GPUs. If your workload values rapid feasible answers, repeated scenarios, and faster iteration, cuOpt is the GPU solver to evaluate now while full MIP optimality continues to mature.