Realize price-performance advantage for AI
summary
Amazon EC2 instances arrive with up to eight Gaudi accelerators and claim 40% lower cost-to-train versus current GPU instances, targeting training for TensorFlow and PyTorch via the Habana SynapseAI stack. An 8‑Gaudi system hits 12,000 images/sec ResNet50 throughput, each Gaudi ships with 32 GB HBM2 and on‑chip 10×100 GbE RoCE for scaling, and instances integrate with SageMaker, ECS/EKS and AWS Elastic Fabric Adapter for distributed training.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:33Agenda
- 1:20AI for your organization
- 3:05Intels AI portfolio
- 7:04AWS and Intel
- 10:19Intel and AWS
- 15:40thorne
- 22:40Havana
- 23:21Introduction
- 23:41Intel and Havana Labs
- 24:20Havana Labs
- 24:45gaudi
- 25:10tensorflow
- 25:29EC2 instances
- 26:21Guard
- 27:38Software experience
- 28:55Summary
- 29:16Outro
related talks
31mRealize price-performance advantage for AI
30mAccelerate deep learning & Habana Gaudi–based Amazon EC2 DL1 instances
50mThe democratization of tomorrow’s technology
24mGenerative AI Chips
50mAI acceleration at the edge
20mAccelerating AI pipelines w/ NVIDIA on AWS: An autonomous driving case
35mReinventing with generative AI on AWS
09mDemocratizing Access to GPUs for Faster MLAI Innovation with Generative AI