“NVIDIA's current-generation end-to-end autonomous driving model can achieve an inference latency of under 100 milliseconds.”