Interconnecting GPUs in Managed Service for Kubernetes® clusters using InfiniBand™

To accelerate ML, AI and high-performance computing (HPC) workloads that you run in your Managed Service for Kubernetes clusters with GPUs, you can interconnect the GPUs using InfiniBand, a high-throughput, low-latency networking standard. For more details about InfiniBand in Nebius AI Cloud, see the Compute documentation. In this article, you will learn how to set up InfiniBand in a Managed Kubernetes cluster.

How to enable InfiniBand for a node group

Web console
CLI

In the node group creation form (

Compute → Kubernetes → your cluster → Node groups → Create node group), under Computing resources:

Select With GPU.

Select a platform and a preset compatible with GPU clusters. The compatible platforms and presets:

Platform	Presets	Regions
NVIDIA® B300 NVLink with Intel Granite Rapids (`gpu-b300-sxm`)	`8gpu-192vcpu-2768gb`	`uk-south1`
NVIDIA® B200 NVLink with Intel Emerald Rapids (`gpu-b200-sxm`)	`8gpu-160vcpu-1792gb`	`us-central1`
NVIDIA® B200 NVLink with Intel Emerald Rapids (`gpu-b200-sxm-a`)	`8gpu-160vcpu-1792gb`	`me-west1`
NVIDIA® H200 NVLink with Intel Sapphire Rapids (`gpu-h200-sxm`)	`8gpu-128vcpu-1600gb`	`eu-north1`, `eu-north2`, `eu-west1`, `us-central1`
NVIDIA® H100 NVLink with Intel Sapphire Rapids (`gpu-h100-sxm`)	`8gpu-128vcpu-1600gb`	`eu-north1`

Select a GPU cluster or create one. If the field is inactive, make sure you have selected a compatible platform and preset.
Under GPU settings, keep the Install NVIDIA GPU drivers and other components option enabled. If you want to install the drivers manually, disable this option.

Depending on your project’s region, select an InfiniBand fabric and save it to an environment variable:
```
export INFINIBAND_FABRIC=<fabric>
```

Create a GPU cluster and save its ID:

export NB_GPU_CLUSTER_ID=$(nebius compute gpu-cluster create \
  --name gpu-cluster-name \
  --infiniband-fabric $INFINIBAND_FABRIC \
  --format json \
  | jq -r ".metadata.id")

Create a node group with GPUs and specify the GPU cluster ID in its parameters by using the nebius mk8s node-group create command:

nebius mk8s node-group create \
  --template-resources-platform gpu-h100-sxm \
  --template-resources-preset 8gpu-128vcpu-1600gb \
  --template-gpu-cluster-id $NB_GPU_CLUSTER_ID \
  --template-gpu-settings-drivers-preset cuda12 \
  ...

In --template.gpu-cluster-id, specify the GPU cluster ID.

In --template-resources-platform, specify a platform with GPUs. In --template-resources-preset, specify a compatible preset (number of GPUs and vCPUs, RAM size). The compatible platforms and presets are:

Platform	Presets	Regions
NVIDIA® B300 NVLink with Intel Granite Rapids (`gpu-b300-sxm`)	`8gpu-192vcpu-2768gb`	`uk-south1`
NVIDIA® B200 NVLink with Intel Emerald Rapids (`gpu-b200-sxm`)	`8gpu-160vcpu-1792gb`	`us-central1`
NVIDIA® B200 NVLink with Intel Emerald Rapids (`gpu-b200-sxm-a`)	`8gpu-160vcpu-1792gb`	`me-west1`
NVIDIA® H200 NVLink with Intel Sapphire Rapids (`gpu-h200-sxm`)	`8gpu-128vcpu-1600gb`	`eu-north1`, `eu-north2`, `eu-west1`, `us-central1`
NVIDIA® H100 NVLink with Intel Sapphire Rapids (`gpu-h100-sxm`)	`8gpu-128vcpu-1600gb`	`eu-north1`

In --template-gpu-settings-drivers-preset, specify a supported preset to use a boot disk image that contains drivers and other components for GPUs. For more details, see GPU drivers and other components. If you want to install the drivers manually, omit the --template-gpu-settings-drivers-preset parameter.

Example: NCCL tests

To test InfiniBand performance in a Managed Service for Kubernetes cluster, you can run the NVIDIA NCCL test in it. For instructions, see our tutorial.

Virtual machines

Slurm and Soperator in Nebius AI Cloud

Managed Service for Kubernetes®

Interconnecting GPUs in Managed Service for Kubernetes® clusters using InfiniBand™

How to enable InfiniBand for a node group

Example: NCCL tests

See also

Virtual machines

Slurm and Soperator in Nebius AI Cloud

Managed Service for Kubernetes®

​How to enable InfiniBand for a node group

​Example: NCCL tests

​See also

How to enable InfiniBand for a node group

Example: NCCL tests

See also