AI BriefingKubernetes 1.34: DRA reaches GA, enabling dynamic GPU allocation and default CDI activation
·2026.11.04 11:19
Key point
In Kubernetes 1.34, DRA transitioned to GA, allowing flexible, attribute-based requests for GPUs.
Details
- DRA (Dynamic Resource Allocation) was promoted to GA in Kubernetes 1.34, supporting dynamic allocation of GPU resources
- ResourceClaim enables condition-based GPU requests, such as memory capacity and NVLink configuration, as well as exclusive allocation per Pod
- CDI (Container Device Interface) is enabled by default, applying a standardized GPU device injection method instead of runtime hooks
- MIG Manager detects node GPUs and automatically generates MIG profiles, reducing the burden of manual configuration
- NVIDIA donated the DRA driver for GPUs to the CNCF at KubeCon Europe in March 2026, formalizing the transition to the open-source ecosystem
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Single-VM OpenStack Sandbox: Cinder NFS Storage Reboot Verification CompletedAnalysis of Power Density and Liquid Cooling Infrastructure Changes in North American AIDC Driven by NVIDIA GPU Evolution