I’m trying to enable GPU time slicing on an NVIDIA RTX A5000 in a Kubernetes environment using the NVIDIA GPU Operator and time-slicing configuration as documented here:
GPU Model:** NVIDIA RTX A5000
Driver Version: — 570.86.15
CUDA Version: — 12.8
NVIDIA Container Toolkit: — This is not getting installed at all
GPU Operator Version: — 25.3.2
Kubernetes Version: — On Master 1.32.7, On Worker (GPU) 1.33.3,
OS:Ubuntu - 22.04.4
GPU Sharing Config:
apiVersion: v1
kind: ConfigMap
metadata:
name: time-slicing-config
namespace: gpu-operator
data:
default: |-
version: v1
sharing:
timeSlicing:
resources:
- name: nvidia.com/gpu
replicas: 8
NVIDIA-RTX-A5000: |-
version: v1
sharing:
timeSlicing:
resources:
- name: nvidia.com/gpu
replicas: 8
I had installed the gpu-operator using the below,
helm install gpu-operator nvidia/gpu-operator
–create-namespace
-n gpu-operator
–version=v25.3.2
–set driver.enabled=true
–set toolkit.enabled=true
–set devicePlugin.enabled=true
–set gfd.enabled=true
–wait --debug
Toolkit, driver daemonset and plugins pods are not getting created. So i manually installed the plugin, the plugin detects the gpu and says there 2 gpu (nvidia. com/gpu: 2)
but slicing is not happening. Also when is describe the GPU node it shows 2 gpu(nvidia. com/gpu: 2) but also shows this (nvidia.com/gpu.present: false).
Please help me out to fix the issue. main issue is with the GPU-Time-Slicing. Please give me step by step documentation, so that i can follow it and slicing works for me.