The convergence of machine learning and high performance computing
summary
Cloud HPC on AWS claims bare-metal performance via the Nitro system and Elastic Fabric Adapter (EFA) with a new Scalable Reliable Datagrams protocol, delivering 100 Gbps, low-latency RDMA and GPU‑direct networking. Benchmarks show C5n clusters matching or beating an on‑prem Cray for Navy forecasting and P4d UltraClusters cut training time by ~40% for TRI‑AD. ARM Graviton2 instances (C6gn) yield ~40% better price‑performance for CFD and weather workloads. EC2 Inferentia targets cheaper, faster inference at scale.
AI-written summary · about
chapters (9)
- 0:00Start
- 5:10Why Is Machine Learning Relevant to High Performance Computing
- 10:31What Specifically Does It Mean To Do Hpc on Aws
- 11:24Hpc on Aws
- 12:04Amazon Elastic Compute Cloud
- 13:08Amazon Ec2
- 15:55Elastic Fabric Adapter
- 25:00Amazon Ec2 Inferentia Instances
- 25:59Aws Training and Certificates
related talks
42mDifferential Software Solutions on HPC and IBM Power Architecture
40mOptimize Your AI Cloud Infrastructure: A Hardware Perspective
21mDesigning Scalable Networks for Large AI Clusters: Challenges and Key Insights
22mResearch SuperCluster(RSC): Accelerating AI Research at Scale
36mAn almost perfect match: cloud native and scientific computing
09mScaling Simulation & HPC with Agentic Workflows on
30mAccelerate deep learning & Habana Gaudi–based Amazon EC2 DL1 instances
19mFireside Chat