[Yandex Cloud documentation](../../index.md) > [Yandex Managed Service for Kubernetes](../index.md) > [Tutorials](index.md) > Setting up time-slicing GPUs # Setting up time-slicing GPUs [Time-slicing GPUs in Kubernetes](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/gpu-sharing.html) enables workloads scheduled on a single oversubscribed [GPU](../../compute/concepts/gpus.md) to interleave with one another. To set up time-slicing GPUs in Managed Service for Kubernetes: 1. [Configure time-slicing GPUs](#configure-time-slicing). 1. [Test time-slicing GPUs](#check-time-slicing). If you no longer need the resources you created, [delete them](#clear-out). ## Required paid resources {#paid-resources} The support cost for this solution includes: * Fee for using the master and outgoing traffic in a Managed Service for Kubernetes cluster (see [Managed Service for Kubernetes pricing](../pricing.md)). * Fee for using computing resources, OS, and storage in cluster nodes (VMs) (see [Compute Cloud pricing](../../compute/pricing.md)). * Fee for a public IP address assigned to cluster nodes (see [Virtual Private Cloud pricing](../../vpc/pricing.md#prices-public-ip)). ## Getting started {#before-you-begin} 1. If you do not have the Yandex Cloud CLI yet, [install and initialize it](../../cli/quickstart.md#install). 1. The folder used by default is the one specified when [creating](../../cli/operations/profile/profile-create.md) the CLI profile. To change the default folder, use the `yc config set folder-id ` command. You can also specify a different folder for any command using `--folder-name` or `--folder-id`. If you access a resource by its name, the search will be limited to the default folder. If you access a resource by its ID, the search will be global, i.e., through all folders based on access permissions. 1. [Create security groups](../operations/connect/security-groups.md) for the Managed Service for Kubernetes cluster and its node groups. {% note warning %} The configuration of security groups determines performance and availability of the cluster and the services and applications running in it. {% endnote %} 1. [Create a Managed Service for Kubernetes cluster](../operations/kubernetes-cluster/kubernetes-cluster-create.md). When creating, specify the preconfigured security groups. 1. [Create a Managed Service for Kubernetes node group](../operations/node-group/node-group-create.md) with [GPU NVIDIA® Tesla® T4](../../compute/concepts/gpus.md#tesla-t4) and preconfigured security groups. 1. [Install kubect](https://kubernetes.io/docs/tasks/tools/install-kubectl) and [configure it to work with the new cluster](../operations/connect/index.md#kubectl-connect). ## Configure time-slicing GPUs {#configure-time-slicing} 1. Create a time-slicing configuration: 1. Create the `time-slicing-config.yaml` file with the following content: ```yaml --- kind: Namespace apiVersion: v1 metadata: name: gpu-operator --- apiVersion: v1 kind: ConfigMap metadata: name: time-slicing-config namespace: gpu-operator data: a100-80gb: |- version: v1 sharing: timeSlicing: resources: - name: nvidia.com/gpu replicas: 5 tesla-t4: |- version: v1 sharing: timeSlicing: resources: - name: nvidia.com/gpu replicas: 5 ``` 1. Run this command: ```bash kubectl create -f time-slicing-config.yaml ``` Result: ```text namespace/gpu-operator created configmap/time-slicing-config created ``` 1. Install [GPU Operator](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/overview.html): ```bash helm repo add nvidia https://helm.ngc.nvidia.com/nvidia && \ helm repo update && \ helm install \ --namespace gpu-operator \ --create-namespace \ --set devicePlugin.config.name=time-slicing-config \ gpu-operator nvidia/gpu-operator ``` 1. Apply the time-slicing configuration to your [Managed Service for Kubernetes cluster](../concepts/index.md#kubernetes-cluster) or [node group](../concepts/index.md#node-group): {% list tabs %} - Managed Service for Kubernetes cluster ```bash kubectl patch clusterpolicies.nvidia.com/cluster-policy \ --namespace gpu-operator \ --type merge \ --patch='{"spec": {"devicePlugin": {"config": {"name": "time-slicing-config", "default": "tesla-t4"}}}}' ``` - Managed Service for Kubernetes node group ```bash yc managed-kubernetes node-group add-labels \ --labels nvidia.com/device-plugin.config=tesla-t4 ``` You can get the Managed Service for Kubernetes node group ID and name with the [list of node groups in the folder](../operations/node-group/node-group-list.md#list). {% endlist %} ## Test time-slicing GPUs {#check-time-slicing} 1. Create a test app: 1. Save the following app specification to a YAML file named `nvidia-plugin-test.yml`. [Deployment](https://kubernetes.io/docs/concepts/workloads/controllers/deployment/) is the Kubernetes API object that manages the replicated application. ```yaml apiVersion: apps/v1 kind: Deployment metadata: name: nvidia-plugin-test labels: app: nvidia-plugin-test spec: replicas: 5 selector: matchLabels: app: nvidia-plugin-test template: metadata: labels: app: nvidia-plugin-test spec: tolerations: - key: nvidia.com/gpu operator: Exists effect: NoSchedule containers: - name: dcgmproftester11 image: nvidia/samples:dcgmproftester-2.0.10-cuda11.0-ubuntu18.04 command: ["/bin/sh", "-c"] args: - while true; do /usr/bin/dcgmproftester11 --no-dcgm-validation -t 1004 -d 300; sleep 30; done resources: limits: nvidia.com/gpu: 1 securityContext: capabilities: add: ["SYS_ADMIN"] ``` 1. Run this command: ```bash kubectl apply -f nvidia-plugin-test.yml ``` Result: ```text deployment.apps/nvidia-plugin-test created ``` 1. Make sure all the app's five [Managed Service for Kubernetes pods](../concepts/index.md#pod) are `Running`: ```bash kubectl get pods | grep nvidia-plugin-test ``` 1. Run the `nvidia-smi` command in the running `nvidia-container-toolkit` Managed Service for Kubernetes pod: ```bash kubectl exec \ --namespace gpu-operator -- nvidia-smi ``` Result: ```text Defaulted container "nvidia-container-toolkit-ctr" out of: nvidia-container-toolkit-ctr, driver-validation (init) Thu Jan 26 09:42:51 2023 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 515.65.01 Driver Version: 515.65.01 CUDA Version: N/A | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 Tesla T4 Off | 00000000:8B:00.0 Off | 0 | | N/A 72C P0 70W / 70W | 1579MiB / 15360MiB | 100% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=============================================================================| | 0 N/A N/A 43108 C /usr/bin/dcgmproftester11 315MiB | | 0 N/A N/A 43211 C /usr/bin/dcgmproftester11 315MiB | | 0 N/A N/A 44583 C /usr/bin/dcgmproftester11 315MiB | | 0 N/A N/A 44589 C /usr/bin/dcgmproftester11 315MiB | | 0 N/A N/A 44595 C /usr/bin/dcgmproftester11 315MiB | +-----------------------------------------------------------------------------+ ``` ## Delete the resources you created {#clear-out} Some resources are not free of charge. Delete the resources you no longer need to avoid paying for them: 1. [Delete the Managed Service for Kubernetes cluster](../operations/kubernetes-cluster/kubernetes-cluster-delete.md). 1. If you created any [service accounts](../../iam/concepts/users/service-accounts.md), [delete them](../../iam/operations/sa/delete.md).