NVIDIA cuOpt is a GPU-accelerated decision optimization engine for LP, MILP, VRP, and QP problems. Powers real-time logistics for companies like Lowe's, with millions of variables solved near-instantly.
All LP methods run concurrently on GPU and CPU. The fastest solution is returned automatically.
Every industry makes thousands of optimization decisions per day. cuOpt makes them near-instant and near-optimal.
cuOpt is available via pip, conda, Docker, and NVIDIA NGC. Integrates with AMPL, PuLP, CVXPY, GAMSPy, and JuMP.
from cuopt import routing
import cudf
# Define fleet and delivery locations
data_model = routing.DataModel(
n_locations=50,
n_vehicles=5
)
# Set cost matrix (distance/time between locations)
cost_matrix = cudf.DataFrame(distance_data)
data_model.add_cost_matrix(cost_matrix)
# Add vehicle capacity constraints
vehicle_capacity = cudf.Series([100] * 5)
data_model.add_capacity_dimension(
"demand", demand_data, vehicle_capacity
)
# Add delivery time windows
data_model.set_order_time_windows(
earliest, latest
)
# Solve — GPU heuristics find optimal routes
solver_settings = routing.SolverSettings()
solver_settings.set_time_limit(5) # 5 seconds
solution = routing.Solve(
data_model, solver_settings
)
# Get optimized routes
routes = solution.get_route()
print(f"Total cost: {solution.get_cost()}")Own the exact-match domain for NVIDIA's AI-powered decision optimization engine — the most valuable of the four CUDA library domains, sitting at the intersection of logistics, AI, and GPU computing.
2368082693@qq.com
Also listed on Afternic · Sedo · Dan.com