Kubernetes VMware User Group: Using GPUs with K8s on vSphere
Talkat KubeCon + CloudNativeCon North America 2021 · on Thursday, 2021-10-14· at 11:00· Concourse Hall 150 ABC + Online·
summary
Demonstrates Bitfusion 4.0 pooling four vSphere GPUs (each ~16 GB VRAM, shown as Tesla D4) into server VMs and serving them to pods on GPU-less nodes via a device plugin and mutating webhook that injects an init container and bitfusion client. Shows a TensorFlow benchmark running unmodified containers on Tanzu/Kubernetes, successfully claiming a full GPU over the network, with 177 historical clients recorded and a recommendation of >=10 GbE (40–100 GbE preferred) for acceptable performance.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:22Kubernetes Vmware User Group
- 11:37Create a Client Token
- 13:51Bitfusion with Kubernetes Integration
- 14:01Vmware Bitfusion with Kubernetes Integration
- 17:06Docker File
- 35:07Gpu Profiling
related talks
44mGPU Sharing for Machine Learning Workload on Kubernetes
38mImproving GPU Utilization using Kubernetes
33mKubernetes VMware User Group Intro: Best Practices for Running on VMware
30mExplain How Kubernetes Works With GPU Like I’m 5
10mLightning Talk: K8s Issue #52757: Sharing GPUs Among Multiple Containers
26mState of the GPU(nion)
37mKubeVirt Deep Dive: Virtualized GPU Workloads on KubeVirt
25mMaking the Most Out of Your Hardware Accelerators in a Kubernetes Cluster