Fireside Chat
summary
AWS describes moving virtualization into hardware starting in 2013 and shipping a full Nitro system in 2017 that cuts hypervisor overhead and yields up to ~27% real‑world performance gains. Custom CPU work begins with Graviton (2018): ~20% lower list price, ~40% better price‑performance, ~60% less energy per unit of work, 50,000 customers and 48 of the top 50 adopting it; interpreted workloads port in weeks. Custom ML silicon includes Inferentia 1 (≈40% lower inference cost), Inferentia 2 (10x latency, 4x throughput) and TRN1 clusters with petabit non‑blocking networking for large‑scale training.
AI-written summary · about
chapters (7)
- 0:00Introduction and welcome
- 0:45Journey at Amazon EC2
- 3:35Why AWS builds ARM processors
- 8:50Migrating to Graviton challenges
- 12:38Sustainability in AWS data centers
- 15:05AI accelerators and Inferentia
- 21:09Future innovations and closing
related talks
36mThe journey of silicon innovation at
01h 00mfor virtually every application
32mBuild without limits with AWS Cloud infrastructure
46mEnhance security with the AWS Nitro System
01h 15mAWS Summit Tel Aviv Keynote
20mAWS On Air ft. AWS Graviton3 Processors and Amazon EC2 C7g Instances
25mAdopt AWS Graviton2 instances for best price-performance in EC2
28mWhat's new in Amazon EC2