Fireside Chat
summary
AWS describes four custom silicon lines—Nitro, Graviton (Graviton3 in 2022), Trainium and Inferentia (Inferentia2)—and reports over 20 million chips deployed since 2012. Graviton instances give up to 40% better price‑performance and up to 60% less energy versus comparable x86 sizes; one partner migration shows spot savings ~10% and overall Gravton-driven savings near 60%. Hugging Face on Inf1 sees up to 70% lower cost and Inf2 yields ~8× lower latency for BERT‑like models.
AI-written summary · about
chapters (7)
- 0:00Introduction to AWS silicon
- 1:56History of AWS hardware
- 4:06Performance and security
- 6:15Partner-driven security
- 8:33Purpose-built silicon
- 11:36Real-world partner success
- 14:30Future of AWS innovation
related talks
01h 05mKeynote: Builders Day
27mThe convergence of machine learning and high performance computing
32mOptimizing AI/ML workloads for sustainability
24mFIRESIDE CHAT: The Future of AI Infra: The Opportunities and Challenges That Await Us On Our Journey
15mScaling the Next Paradigm of Heterogeneous Intelligence
20mFireworks AI: Go from Hype to High - Value AI
20mBuilding GenAI That Doesn't Cost the Earth
35mAccelerate your ML innovation with AWS ML infrastructure services