Networking for AI Infrastructure

Configure VPCs, placement groups, Elastic Fabric Adapter (EFA), and security groups for high-performance distributed GPU training with Terraform.

VPC for AI Workloads

resource "aws_vpc" "ai_vpc" {
  cidr_block           = "10.0.0.0/16"
  enable_dns_hostnames = true
  tags = { Name = "ai-training-vpc" }
}

resource "aws_subnet" "gpu_subnet" {
  vpc_id            = aws_vpc.ai_vpc.id
  cidr_block        = "10.0.1.0/24"
  availability_zone = "us-east-1a"
  tags = { Name = "gpu-training-subnet" }
}

# Placement group for low-latency GPU-to-GPU communication
resource "aws_placement_group" "gpu_cluster" {
  name     = "gpu-training-cluster"
  strategy = "cluster"  # Place instances physically close
}

High-Bandwidth Networking (EFA)

resource "aws_instance" "gpu_node" {
  instance_type    = "p4d.24xlarge"
  placement_group  = aws_placement_group.gpu_cluster.id
  subnet_id        = aws_subnet.gpu_subnet.id

  network_interface {
    device_index          = 0
    network_interface_id  = aws_network_interface.efa.id
  }
}

resource "aws_network_interface" "efa" {
  subnet_id       = aws_subnet.gpu_subnet.id
  interface_type  = "efa"  # Elastic Fabric Adapter for RDMA
  security_groups = [aws_security_group.gpu_sg.id]
}