Silicon Development
summary
Annapurna Labs explains how cloud-scale design and manufacturing yield cheaper, more efficient processors: 20 million chips deployed since 2014, custom CPUs (Graviton), Trainium and Inferentia for AI, and five Nitro generations. Design specifics include ~50 billion transistors, ~35 km of on-chip wiring and ~80 metal layers and power delivery that can spike to ~500 A in microseconds. Tooling and orchestration use the AWS cloud, Neuron SDK, EFA/ENA networking and analytics (Redshift, Glue, QuickSight) to cut cost, improve reliability and speed time to volume.
AI-written summary · about
chapters (12)
- 0:00Introduction and team background
- 1:44Annapurna Labs founding and mission
- 5:16Complexity of chip design and manufacturing
- 8:44End-to-end chip development at AWS
- 10:57Integrating chips with AWS infrastructure
- 13:27Predicting AI workloads for future chips
- 14:30Developing Graviton CPUs challenges
- 15:52Advantages of cloud-based chip development
- 17:59Energy efficiency in silicon design
- 19:22Neuron SDK for AI optimization
- 21:30Future of silicon development at AWS
- 22:55Closing and team achievements
related talks
36mThe journey of silicon innovation at
32mBuild without limits with AWS Cloud infrastructure
25mSustainability at the Core with Cloud Native Processors
16mKeynote: Designing Processors in the Cloud: How Advanced Emulation and AWS Cloud Infrast...
42mHow Intel & AWS solve some of the world’s most urgent issues
42mEmerging Computing Trends in the Datacenter
01h 00mfor virtually every application
12mAI-powered infrastructure: