Unverified50% confidenceTradeoffExact time
Kubernetes' native GPU scheduling capability is relatively coarse-grained, typically only able to allocate GPUs in whole-GPU units.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/4/2026
First Seen
Valid until: 12/3/2026
Sources
Related Entities
Related Claims
UnverifiedKubernetes 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77% similarUnverified只要GPU上有任何一个计算内核在运行，即使只占用极小部分计算单元，nvidia-smi的GPU利用率指标依然会显示100%70% 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69% similarUnverifiedGPU utilization (DCGM_FI_DEV_GPU_UTIL) measures the proportion of time during a sampling period in which at least one kernel is running on the GPU.67% similarUnverified在NVIDIA GPU架构中，线程以32个为一组形成一个warp，这是GPU调度和执行的基本单位66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/855563API
curl https://kongchang.com/api/v1/knowledge/claims/855563MCP
get_claim(id=855563)