Inside Azure innovations with Mark Russinovich
summary
Azure reports 470 billion inferences/month on its ML serving tier and shows concrete infra wins: hollow core fiber cuts NY–London latency from 55ms to 36ms and expands regional data‑center radius 2.25×; an FPGA offload card firmware upgrade boosts Ebsv5 storage to 400,000 IOPS and 10 GB/s on the same hardware; Project Forge runs 1,024 A100 GPUs for 18 days with transparent GPU+CPU checkpointing and Jupyter kernel restores; Project Silica robotic glass library prototype stores archival data.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 0:40Scale
- 1:12Numbers
- 2:12Hollowcore fiber
- 3:57Server infrastructure
- 5:25Server cards
- 5:56EBS V5
- 7:16Iometer
- 8:54AI infrastructure
- 12:41Transparent Checkpointing
- 13:56ProjectForge
- 14:54Device Proxy
- 15:36Overhead
- 16:49Checkpointing
- 19:45Tensor Query Processing
- 21:53Cloud Native Infrastructure
- 25:27Project Copacetic
- 27:38Dapper
- 34:41Confidential Computing
- 36:01Confidential ACI
- 39:10Storage
related talks
57mInside Microsoft AI innovation with Mark Russinovich
01h 05mKeynote: Building Applications with the Modern SQL Using AI and Microsoft Fabric
50mAzure AI, Power new possibilities for every organization
31mAzure AI Innovations at Microsoft
25mOptimize productivity by meeting next-level demands of the AI era
47mBuilding AI applications that leverage your data in object storage
50mAccelerate development and deployment of Gen AI with NVIDIA on Azure
46mAdvanced RAG with Azure AI Search - top questions answered