2,070 TFLOPS on a Robot's Back

Intermediate · 3 lessons · 0 complete

NVIDIA's Jetson Thor delivers 2,070 FP4 TFLOPS with 128 GB of memory in a 40 to 130 watt envelope — roughly 7.5 times the AI performance of AGX Orin at 3.5 times the efficiency. This course is about what that unlocks for onboard robot autonomy, and the constraints that don't go away: power budgets, heat, latency, and deciding what genuinely has to run on the robot.