gRPC remote execution

NVIDIA cuOpt gRPC remote execution runs optimization solves on a remote GPU host. Clients can be the Python API, C API, `cuopt_cli`, or a custom program that speaks CuOptRemoteService over gRPC. For Python, the C API, and cuopt_cli, set CUOPT_REMOTE_HOST and CUOPT_REMOTE_PORT to forward solves to cuopt_grpc_server.

Note

Problem types (gRPC remote): LP, MILP, and QP are supported today. Routing (VRP, TSP, PDP, and related APIs) over gRPC remote execution is not available yet; support is planned for an upcoming release. For routing against a remote service today, use the HTTP/JSON REST self-hosted server.

This is not the HTTP/JSON REST self-hosted server (FastAPI). REST is for arbitrary HTTP clients; gRPC is for the bundled remote client in NVIDIA cuOpt’s native APIs.

Start with Quick start (install selector, how remote execution works, Docker, and a minimal LP example). Use Advanced configuration for TLS, tuning, limitations, and troubleshooting; Examples for additional patterns.

In this section

• Quick start
  • How remote execution works
  • Install NVIDIA cuOpt
  • Run the gRPC server (GPU host)
  • Point the client at the server
  • Minimal Python example (LP)
  • Minimal cuopt_cli example (LP)
  • Next steps
• Advanced configuration
  • Configuration parameters
  • Usage
  • Limitations and scope
  • Troubleshooting
  • Further reading
• Examples
  • Where to find examples
  • More samples
  • REST vs gRPC
• gRPC API (reference)
  • Service: CuOptRemoteService
  • Messages and constraints
  • Further reading
• gRPC server behavior
  • Process model
  • Job lifecycle (summary)
  • Job states
  • Logs, capacity, and workers
  • Fault tolerance and cancellation
  • Further reading

See System Requirements for GPU, CUDA, and OS requirements.