Kubernetes DRA Reaches GA: Native GPU Scheduling and Slicing
sloppenheimer · x · 2026-08-13
Kubernetes' Dynamic Resource Allocation (DRA) reached general availability in v1.34 and is enabled by default since v1.35, natively supporting GPU scheduling.
A maintainer from CNCF incubating project HAMi breaks down DRA's impact on existing GPU sharing solutions. Previously, due to the narrow device plugin API, HAMi built a complex pipeline to achieve GPU memory and compute slicing. Now, DRA's consumable capacity feature allows Pods to request fractions of a device directly from the scheduler, absorbing HAMi's resource allocation role.
However, DRA does not make HAMi obsolete. While DRA handles scheduling, HAMi continues to provide hard in-container enforcement at the CUDA-call granularity. HAMi has decoupled its enforcement layer from DRA-based allocation to adapt to the new ecosystem.
More from Infra
- New hardware day: local LLM deployment the moment it arrives — lxfater · 2026-10-16
- Together AI's score centering fixes RL training collapse under quantized inference — nrehiew_ · 2026-10-16
- Inference is splitting prefill and decode across specialized chips, not just H100s — AccBalanced · 2026-10-16
- Rebasely launches in beta: Git hosting built for humans and coding agents — handotdev · 2026-10-16
- Drama3 TTS lands in audio.cpp: up to 5x real-time on RTX 5090, <8GB VRAM mode too — Acceptable-Cycle4645 · 2026-10-16
- Three CoreWeave co-founders have sold $3.3B in shares since IPO amid credit worries — SumitGup · 2026-10-16