tmckayus / cuOpt-Resources

A collection of NVIDIA cuOpt samples and other resources

cuOpt Resources

NVIDIA cuOpt is an Operations Research optimization API using AI to help developers create complex, real-time fleet routing workflows on NVIDIA GPUs. This repository contains a collection of resources demonstrating the setup and use of NVIDIA cuOpt via Python and REST based microservice APIs.

The cuOpt-Resources repository is under MIT License

Use of NVIDIA cuOpt is subject to the End User License Agreement

NVIDIA cuOpt

cuOpt uses highly optimized GPU-accelerated solvers relying on heuristics, metaheuristics, and optimization. In addition to providing dramatically accelerated, world class solutions to some of the most difficult optimization problems, NVIDIA cuOpt prioritizes ease of use through high level Python and REST based microservice APIs

cuOpt Docs

cuOpt on NGC container

• Collections
• Containers
• Helm Charts

Contents

• NVIDIA cuOpt Python API example notebooks
  • Routing Optimization
    
    
• NVIDIA cuOpt REST based microservice example notebooks
  • Routing Optimization
    
    
• Cloud deployments scripts for AWS, Azure, and GCP
  
  
• Demo Extension showing cuOpt in Omniverse Isaac Sim

Setup

Whether deploying locally or in the cloud, an NVIDIA GPU Cloud (NGC) account will be required. NGC hosts a catalog of GPU-optimized software for AI developers. Access is free and it takes just a few minutes to register. For information about creating an NGC account and generating the required API key, see the following documentation:

Registering and Activating NGC Account

Local Environment

Prerequisites

• NVIDIA GPU and associated driver 450.80.02+
  • Pascal architecture or better
• CUDA 11.0+
• cuOpt Docker container downloaded from NGC

Instructions

• Clone the cuOpt-Resources repo
• Pull the cuOpt container from NGC
• Follow the instructions provided on NGC to run the cuOpt container and mount the example notebooks relevant to your desired API.

Cloud Environment

Prerequisites

• AWS, Azure, or GCP account (but see Alternative "Cloud Local" for another option)
• Permission to provision compute resources
• NVIDIA NGC account credentials

Alternative "Cloud Local"

If you don't have an AWS, Azure, or GCP account or you want to run cuOpt with NVIDIA Cloud Native Core on a server that you create yourself, read on and look for the Cloud Local section in the cloud-scripts directory README.

Instructions

• Clone the cuOpt-Resources repo
• See detailed instructions provided in the cloud-scripts directory README.