cuOpt consume a large amount of GPU memory when solving MIP problems, even if they are very small (e.g., markshare_4_0). In fact, solving markshare_4_0 with just 34 variables and 4 constraints consumes the same amount of memory as savsched1 with 328575 variables and 295989 constraints. As a result, the solver runs out of memory when running on a RTX ADA 1000 (6 GB of VRAM) and always consume 90 GB of VRAM when running on a GB100.
cuOpt consume a large amount of GPU memory when solving MIP problems, even if they are very small (e.g., markshare_4_0). In fact, solving markshare_4_0 with just 34 variables and 4 constraints consumes the same amount of memory as savsched1 with 328575 variables and 295989 constraints. As a result, the solver runs out of memory when running on a RTX ADA 1000 (6 GB of VRAM) and always consume 90 GB of VRAM when running on a GB100.