Mobileye: Smarter, safer autonomous solutions with Habana & Granulate (PRT279)
summary
Mobileye trains a shared perception backbone on Amazon EC2 DL1 Habana Gaudi instances and runs large Spark map pipelines on AWS S3 with >100k CPU cores to build the Roadbook from ~1.5M daily vehicle uploads; DL1 yields >20% price/perf on small nets and ~40% vs g5 (60% vs p4d) on distillation workloads with near-linear multi-instance scaling, and integrating Granulate cuts CPU hours by ~45%, reducing EC2/EBS costs while preserving output.
AI-written summary · about
chapters (5)
- 0:00Introduction and Intel's role
- 5:53Mobileye's autonomous driving tech
- 12:27Scaling AI training with Habana DL1
- 16:18REM mapping with Granulate optimization
- 38:52Collaboration and future impact
related talks
50mThe democratization of tomorrow’s technology
21mML Infrastructure for Autonomous Vehicles
40mSmart cars of tomorrow: real-time driving patterns
15mScaling the Next Paradigm of Heterogeneous Intelligence
57mBuilding GPU-accelerated multi-agent apps
31mDeep learning for self-driving vehicle technology
19mChallenges & Opportunities for Building Crowdsourced Mapping Services for Autonomous Driving
41mPeer Beneath the Surface of the Earth