Empower Heterogeneous Edge AI Acceleration with Kubernetes

Explore a unified framework for empowering heterogeneous edge AI acceleration with Kubernetes in this conference talk by VMware experts. Delve into the challenges of managing diverse edge AI accelerators from various vendors, including Nvidia edge GPUs, Intel Movidius VPUs, and Google edge TPUs. Learn about key mechanisms such as extending Node Feature Discovery for automatic detection of edge AI accelerators, unifying device plugins for provisioning work nodes, introducing transparent backend accelerations to boost ML upstream frameworks, and attaching remote GPUs to edge devices. Gain insights into how this unified edge AI framework helps Kubernetes empower heterogeneous AI acceleration on the edge, addressing the emerging trends in edge computing and workload management.

Empower Heterogeneous Edge AI Acceleration with K8s - Tiejun Chen & Zitong Xu, VMware