Software Defined Hardware Accelerated Data-Centric Computing...
summary
Demo shows an x86 OpenShift host composably attach Intel IPU cards and run ARM MicroShift on DPU nodes over a PCIe fabric, with container network functions (firewall, packet filter, compression) offloaded into the IPU packet‑processing engine. A cited IOWN report claims 62% power reduction vs conventional servers, and a composable CDI example drops 25 servers to 5 (claimed 88% cost cut) by pooling GPUs/DPUs. Operators provision DP/IPU operators, traffic dashboards update live, and CPU load is visibly reduced.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:08Ion data-centric infrastructure
- 5:23Kubernetes accelerator chaining challenges
- 12:20DPU and IPU in open PR infrastructure
- 14:58Intel IPU integration with OpenShift
- 26:10DPU and IPU benefits over smart NICs
- 30:08Future of optical computing
- 31:12Summary
related talks
24mUsing Kubernetes with Data Processing Units to Offload Infrastructure
05mTowards a Sustainable World Realized by Kubernetes and Composable...
38mKubernetes Networking Infrastructure Offload
05mProject Lightning Talk: CoHDI: Dynamic Hardware Composition for AI‑Era Kubernetes…
34mDeclarative Device Virtualization: Orchestrating GPUs & Hardware in Cloud Native Environments
32mPack-em or Stack-em: Advantages of Running OpenShift on Power
19mOpenShift for low latency and real time workloads
08mContainerization – It’s Not Just for CPUs Anymore: Introducin...