How Can I Give Virtual GPU Resources to My End Users Seamlessly?
summary
OpenStack Nova now supports virtual GPU and GPU passthrough and has added support for recent NVIDIA SR-IOV architectures. The talk explains the recent Nova enhancements and shows how to manage and expose GPU resources to end users, including an experimental feature to enforce GPU quotas for users.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:11vGPU in Nova
- 5:25vGPU config demo
- 11:39Caracal improvements
- 12:41SR-IOV GPU support
- 14:50SR-IOV bug
- 18:50vGPU live migration
- 21:18GPU compute demo
- 25:34Live migration demo
- 28:31Constraints and optimizations
- 30:19Unify Limits for vGPU quotas
- 32:52Quota demo
- 35:19Q&A
related talks
18mSmartNIC DPU Enablement in OpenStack with OVN
05mFrom Open Source to Real-World Scale
21mExperience Operating Large GPU Clusters at Organizational Scale
23mEnhancement and optimization of new heterogeneous accelerators based on Cyborg
29mThe Path to GPU as a Service in Kubernetes
32mSmart GPU Management: Dynamic Pooling, Sharing, and Scheduling for AI Workloads in Kubernetes
16mBOSH & GPU - a Match Made in Heaven
36mOpenPOWER and GPU Porting Platform for FOSS Using OpenStack at OSUOSL