Unverified50% confidenceFactExact time
In Kubernetes, GPUs are exposed as extended resources (such as nvidia.com/gpu) through the Device Plugin mechanism, and the scheduler only performs count-based allocation with almost no awareness of model and generation differences.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedNVIDIA Device Plugin registers the nvidia.com/gpu resource type with the Kubernetes scheduler, enabling Pods to declare GPU requirements via resources.limits.80% similarUnverifiedKubernetes 默认将 GPU 视为不可分割的扩展资源，一个 Pod 请求 1 块 GPU 即独占其全部算力和显存76% similarUnverifiedKubernetes' native GPU scheduling capability is relatively coarse-grained, typically only able to allocate GPUs in whole-GPU units.76% similarUnverifiedNVIDIA的k8s-device-plugin会将每块物理GPU注册为不可分割的资源单元（nvidia.com/gpu: 1），调度器以此为原子单位分配76% similarUnverifiedWhen using GPU sharing on Kubernetes, the scheduler's VRAM quota allocation and the runtime's actual execution must stay consistent, otherwise isolation fails.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1127698API
curl https://kongchang.com/api/v1/knowledge/claims/1127698MCP
get_claim(id=1127698)