# gpu-infer 命名空间配额：最多 8 张卡（案例约定）
apiVersion: v1
kind: ResourceQuota
metadata:
  name: gpu-infer-quota
  namespace: gpu-infer
spec:
  hard:
    requests.nvidia.com/gpu: "8"
    limits.nvidia.com/gpu: "8"
    requests.cpu: "64"
    requests.memory: 256Gi
    pods: "40"
---
# GPU 节点污点：只跑推理（控制面 / 普通业务需容忍或避开）
# 实际执行：kubectl taint nodes gpu-w01 nvidia.com/gpu=true:NoSchedule
# 下列为文档对照用示意；污点通常用 kubectl / 节点 bootstrap，不靠本文件 apply
apiVersion: v1
kind: ConfigMap
metadata:
  name: gpu-node-ops-notes
  namespace: gpu-infer
data:
  taint: "nvidia.com/gpu=true:NoSchedule"
  label_present: "nvidia.com/gpu.present=true"
  label_product: "nvidia.com/gpu.product=NVIDIA-L40S"