Results
Each platform ran the same workload: Nav2 driving a material handling AMR through a 200,000 sqft warehouse while a VLM saturates the GPU with situational awareness requests. Results are grouped by TDP power levels and later optimized configurations.
Each platform configured to its maximum rated TDP, so the comparison reflects the most compute each board can deliver.
| Run | TDP | CPU | Missions | Misses/s |
|---|---|---|---|---|
| AMD Strix Halo | 120 W | 18.1% | 10 | 0.46 |
| NVIDIA Jetson Thor | 130 W | 49.8% | 10 | 1.58 |
| NVIDIA Jetson Orin | 65 W | 93.6% | 3 | 5.65 |
Maximum rated TDP with the AI inference stack tuned for the platform according to OEM best practices.
| Run | TDP | CPU | Missions | Misses/s |
|---|---|---|---|---|
| AMD Strix Halo (optimized) | 120 W | 14.4% | 10 | 0.49 |
| NVIDIA Jetson Thor (optimized) | 130 W | 47.2% | 10 | 1.85 |
Each platform set to a balanced power profile (except Orin AGX, max TDP 65W).
| Run | TDP | CPU | Missions | Misses/s |
|---|---|---|---|---|
| AMD Strix Halo | 85 W | 20.2% | 11 | 2.33 |
| NVIDIA Jetson Thor (90 W) | 90 W | 52.0% | 10 | 0.52 |
| NVIDIA Jetson Thor (70 W) | 70 W | 56.4% | 10 | 2.27 |
| NVIDIA Jetson Orin (65 W) | 65 W | 93.6% | 3 | 5.65 |
Application performance comes first on every category page: missions completed, control loop misses, planner cycle time, VLM outcomes. A platform can look healthy on utilization counters and still be failing the robot. The utilization charts explain why a platform performed as it did, but they are not the result.
Every chart carries the run it was measured from and a data table downloadable as CSV.
