How-To Create SXM/Infiniband VMs using UI, CLI and Terraform with Lifecycle Scripts – Crusoe Support

Introduction

This guide explains how to create SXM/Infiniband Virtual Machines (VMs) using Crusoe Cloud. You will learn three methods: using the User Interface (UI), the Command-Line Interface (CLI), and Terraform scripts, along with incorporating Lifecycle Scripts. These methods allow for flexible deployment of high-performance computing environments utilizing Infiniband networks.

Prerequisites

Crusoe Cloud Account: Ensure you have access with the necessary permissions.
Public SSH Key: A valid SSH key is required for VM access.
CLI Installed: Install the Crusoe CLI and authenticate. Refer: Install and configure the CLI.
Terraform Installed: Install and configure Terraform. Refer: Getting Started with Terraform.
Access Permissions: Verify permissions for VM and Infiniband partition creation.
Startup Script: Ensure you have a startup script available (e.g., ~/Downloads/startup-script.sh).

Step-by-Step Instructions

Using the UI

Navigate to Crusoe Cloud Console:
- Log in to the Crusoe Cloud Console.
- Click the Compute tab in the left navigation panel.
Create a New Instance:
- Click the Instances tab.
- Click Create Instance.
Configure Instance Details:
- Instance Name: Provide a name.
- Instance Type: Select h100-80gb-sxm-ib.8xor a100-80gb-sxm-ib.8x. To learn more about instance types, please refer our public document.
- Location: Choose us-southcentral1-aor any other locations such aseu-iceland1-a, us-northcentral1-a, us-east1-a.
- Image: Select ubuntu22.04:latest or any other images listed in our public document.
Network Settings:
- Select VPC Network, VPC Subnet, IB Network, and IB Partition.
- Note: IB partitions are region-specific.
Add Lifecycle Scripts:
- Upload or specify your startup script (startup-script.sh).
Finalize and Create:
- Add SSH Key.
- Click Create.

Using the CLI

Run the following command:

crusoe compute vms create \
  --name my-vm \
  --type h100-80gb-sxm-ib.8x \
  --location us-east1-a \
  --image ubuntu22.04:latest \
  --keyfile ~/.ssh/id_ed25519.pub \
  --ib-partition-id 7c8eb124-44c3-4897-88b5-841442446c9b \
  --startup-script ~/Downloads/startup-script.sh

Replace parameters as needed.

Using Terraform

// Crusoe Provider
terraform {
  required_providers {
    crusoe = {
      source = "registry.terraform.io/crusoecloud/crusoe"
    }
  }
}

// Local files
locals {
  ssh_key        = file("~/.ssh/id_ed25519.pub")  // Replace with the correct path to your public SSH key
  startup_script = file("~/Downloads/startup-script.sh")  // Ensure this file exists
}

// New VM with InfiniBand and Lifecycle Script
resource "crusoe_compute_instance" "infiniband_vm" {
  name     = "infiniband-vm"
  type     = "h100-80gb-sxm-ib.8x"
  location = "us-southcentral1-a"
  image    = "ubuntu22.04:latest"
  ssh_key  = local.ssh_key

  host_channel_adapters = [{
    ib_partition_id = "ce608b10-6a3e-40aa-a391-aa4c39929ccf"  // Replace with your actual InfiniBand partition ID
  }]

  network_interfaces = [{
    subnet = "2c7b6b4f-e856-4eca-8f5a-ff594f76d90a"  // Replace with the correct subnet ID for the location
    public_ipv4 = {
      type = "static"
    }
  }]

  startup_script = local.startup_script
}

Terraform Deployment Steps:

Initialize Terraform: terraform init
Validate the configuration: terraform plan
Apply configuration: terraform apply

Example Use Case

A user deploying an h100-80gb-sxm-ib.8x instance in us-southcentral1-a for HPC workloads benefits from high-speed networking using Infiniband, while leveraging a startup script for automation.

Common Issues & Resolutions

Invalid IB Partition:
- Ensure the IB partition ID matches the VM region.
SSH Key Issues:
- Verify key format and upload correctly.
CLI Authentication Errors:
- Check .crusoe/config parameters. Please refer: Install and configure the CLI.
Startup Script Not Executing:
- Verify the script has execution permissions and correct syntax.

Additional Resources