KubeFATE deploys GPU-enabled FATE

Prerequisites

KubeFATE v1.11.1+
Kubenertes support GPU (https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/)

The algorithm and device in cluster.yaml must be changed to NN and GPU. (Currently FATE only supports the use of GPU for NN algorithms)

algorithm: NN
device: GPU

Then the resource of the python pod is allocated at least 1 GPU resource. (GPU computing is only in the pod of fateflow)

python:
  resources:
    requests:
      nvidia.com/gpu: 1
    limits:
      nvidia.com/gpu: 1

Here is an example cluster-gpu.yaml.

Then deploy the cluster defined by cluster.yaml, and you can use FATE to run GPU tasks.