Accelerating AI pipelines w/ NVIDIA on AWS: An autonomous driving case
summary
NVIDIA and AWS pipeline detail: data collection drives scale — a 50‑car fleet collecting six hours/day produces 2 petabytes/day and can generate up to 1 billion images with only 3 million labeled images used for training. Development uses SageMaker, p3/p3d instances and NVIDIA GPUs (Saturn V supercomputer for internal runs) to train 20+ DNNs; object‑detection models train under 20 minutes, replay runs at super real‑time (3–5× faster) and simulation covers up to 100 million driving scenarios for validation.
AI-written summary · about
chapters (7)
- 0:00Start
- 8:53NVIDIA breaks records in Al perfomance
- 11:59Computational scale required
- 14:02NVIDIA DRIVE validation methodology
- 16:39DRIVE Constellation is an open platform
- 17:38NVIDIA SDK and AWS
- 18:29Acceleration software on NGC
related talks
40mFueling the AI Revolution with Gaming
54mDeep learning on AWS with NVIDIA: From training to deployment
32mInside NVIDIA’s AI infrastructure for self-driving cars
28mScaling agentic AI: From infrastructure to enterprise outcomes
42mAI and Self-Driving Cars
31mRealizing Industrial AI at Scale with
42mFrom Zero to Machine Learning in 45 Minutes
29mGet rolling with ML on AWS DeepRacer