Nvidia launched their Cuopt SDK for GPU optimisation; I modified an existing Winforms tool I had built to connect to the API from the Nvida docker container to test the optimisation solution. I used HERE maps to generate the route matrix and then optimised it using a Cuopt docker container running on my local RTX2080Ti. The optimisation performance was spectacular.
Performance
| Scenario | GPU | CPU |
| Typical fleet route 50-100orders | 10 seconds | 60-120 seconds |
| 700 order scenario | 20 seconds | 9 mins |
| 5000 order scenario | 2 mins | 2 hours, 30 mins |
The GPU optimisation engine is a breakthrough in optimisation performance. It allows for dynamic re-optimisation once the plan has been committed to scheduling new orders and offering tighter time window delivery slots on booking websites by optimising the entire scenario as is, in real-time to generate available time window slots.