Node groups with GPUs

You can create Managed Service for Kubernetes cluster node groups with graphics accelerators (GPUs). A node is created from an image of a GPU-compatible VM with NVIDIA drivers and CUDA libraries installed for GPU acceleration.

Prerequisites

• A cloud must have a non-zero GPU quota.

  By default, a cloud has a zero quota for creating VMs with GPUs. You can request a quota increase in the management console. To do this, you need the quota-manager.requestOperator role or higher.

• A node group must reside in the ru-central1-a or ru-central1-b availability zone, or in both. VMs with GPUs are only available in these zones.

  When requesting a GPU quota, keep in mind which zones you are going to run your Kubernetes clusters in.

Pricing

To run node groups with GPUs, you need a Managed Service for Kubernetes cluster, a VM with a GPU, and traffic. Therefore, billing consists of the following parts:

• Using a Managed Service for Kubernetes master is charged according to the Managed Service for Kubernetes pricing policy.
• VM with a GPU, according to the Yandex Compute Cloud pricing policy.
• Outgoing traffic, according to the Yandex Virtual Private Cloud pricing policy.

Use cases

• Using node groups with GPUs and no pre-installed drivers