nvidia.com/gpu

kubectl get nodes -o json | jq -r '.items[] | select(.status.capacity."nvidia.com/gpu" != null) | {name: .metadata.name, capacity: .status.capacity}'

 tolerations:
   - key: nvidia.com/gpu
     operator: Exists
     effect: NoSchedule

See also

• Nvidia tools: CUDA, NVIDIA Data Center GPU Manager (DCGM), nvcc, Nvidia Management Library (NVML), nvidia-smi, nvidia-device-plugin, CUDA Time-Slicing