GPU Resources

Updated at: 2025-11-18 14:21:25

The Virtual Kurbernets Services provides the following GPU resource types. When applying to Activate Virtual Kurbernets Services, please select the required models as needed.

Model Type	GPU Label	GPU Memory	Resource Request / Limits
NVIDIA-H800A-NV-96G	Contact customer service	96G	18c : 200G : 1GPU
NVIDIA-H800A-NV-80G	Contact customer service	80G	13c : 200G : 1GPU
NVIDIA-H800-NV-80G	nvidia.com/gpu-h800	80G	18c : 200G : 1GPU
NVIDIA-L40S-PCIE-48G	nvidia.com/gpu-l40s	48G	10c : 100G : 1GPU

GPU Usage

In a [Container], you can define GPU resource requests and limits. The key is the GPU label, and the value is the number of GPUs requested or limited.
Example configuration:
Resource Requests: GPU Label: nvidia.com/gpu-h800
GPU Count: 1
cpu: 18
memory: 200G

Resource Limits: GPU Label: nvidia.com/gpu-h800
GPU Count: 1
cpu: 18
memory: 200G

tip

The CPU and memory assigned per GPU must not exceed the resource request/limit values in the table above. For multiple GPUs, CPU and memory should increase proportionally. When specifying CPU, write only the numeric value without the letter "c".
For example: cpu: 18

Example:

spec:
  imagePullSecrets:
    - name: your-harbor-secret
  containers:
    - name: your-container-name
      image: your-image:tag   # Replace with your image name
      command: your-command
      resources:
        requests:
          memory: "200G"
          cpu: "18"
          nvidia.com/gpu-h800: 1 # Request 1 GPU
        limits:
          memory: "200G"
          cpu: "18"
          nvidia.com/gpu-h800: 1 # Limit 1 GPU