Skip to main content

​
Overview

This guide provides a detailed breakdown of AWS infrastructure costs when deploying Poolside. The first step in cost estimation is selecting your deployment model, as this fundamentally affects the cost structure of your implementation.
All services and components described below are required.

​
Deployment options

Amazon EC2

  • Complete control over infrastructure
  • Self-hosted GPU instances via Amazon EC2
  • Full infrastructure cost responsibility
  • Ideal for organizations requiring maximum control

Amazon Bedrock

  • Serverless model inference
  • No GPU infrastructure management
  • Pay-per-request pricing model
  • Reduced operational overhead

​
Cost structure comparison

Amazon EC2

  • EKS cluster management
  • GPU instance costs (p5.48xlarge/p5e.48xlarge)
  • Platform infrastructure (CPU nodes, RDS, etc.)
  • Network and storage costs

Amazon Bedrock

  • Platform infrastructure only (CPU nodes, RDS, etc.)
  • Bedrock inference pricing (see AWS documentation)
  • Reduced operational costs
  • No GPU infrastructure costs

​
Amazon EC2: Deployment cost breakdown

For organizations choosing to provision GPU instances for Poolside via Amazon EC2, the following sections detail the complete infrastructure components and associated costs. These costs can be modeled using the AWS Pricing Calculator.

​
Amazon EKS cluster

EKS Control Plane

  • EKS cluster running version 1.31+
  • Required EKS add-ons:
    • EBS CSI driver
    • Pod Identity Agent

GPU Node Group

  • p5.48xlarge or p5e.48xlarge instances
  • 200 GB gp3 EBS volumes per node (3000 IOPS)
  • Supports On-Demand, Reserved Instances, or Savings Plans
  • Multiple nodes for high availability (configurable)

CPU Node Group

  • m5.4xlarge instances for platform services
  • 200 GB gp3 EBS volumes per node (3000 IOPS)
  • Minimum of 3 nodes recommended for production
  • Runs core-api and supporting services

​
Database and storage

Amazon RDS

  • db.m7g.xlarge PostgreSQL 16.2 instance
  • 64 GB allocated storage
  • Single-AZ deployment by default
  • Optional configurations:
    • Multi-AZ deployment
    • Read replicas
    • Automated backups

Amazon S3

  • Three default buckets:
    • poolside-logs: Platform logging
    • poolside-artifacts: Model storage
  • Standard S3 storage pricing applies

​
Network infrastructure

VPC Components

  • Virtual Private Cloud (VPC)
  • Internet Gateway
  • Optional NAT Gateway
  • Multiple AZ subnet configuration
  • Network ACLs and Security Groups

Data Transfer

  • Inter-AZ data transfer charges
  • Internet ingress/egress charges
  • VPC endpoint charges (if configured)

​
Amazon Bedrock: Deployment cost breakdown

For organizations preferring a serverless approach to model inference, provisioning GPU nodes via Amazon Bedrock provides a more managed experience with different cost considerations.

​
Platform infrastructure

The base infrastructure is similar to Amazon EC2 deployments but without GPU nodes: 
EKS:
  Control Plane: 1 cluster
  CPU Nodes: 3x m5.4xlarge
  EBS Volumes: 200GB gp3 per node

RDS:
  Instance: db.m7g.xlarge
  Storage: 64GB
  Engine: PostgreSQL 16.2+

Network:
  VPC
  Internet Gateway
  NAT Gateway (optional)
  Data Transfer (estimate based on usage)

S3:
  Standard Storage (estimate based on usage)

​
Usage costs

Amazon Bedrock usage costs are based on:
  • Input tokens processed
  • Output tokens generated
  • Model type selected
  • Request volume
Review the Amazon Bedrock pricing page for current rates and to estimate your usage costs.
Deploying Poolside via Amazon Bedrock can significantly reduce operational overhead and provide more predictable scaling costs for organizations that do not want to manage their own GPU infrastructure.

​
Cost estimation process

  1. Choose your deployment modality (Amazon EC2 or Amazon Bedrock)
  2. Work with your Poolside representative to determine:
    • Expected concurrent user load
    • Model inference requirements
    • High availability needs
    • Data retention requirements
  3. Use the AWS Pricing Calculator with the appropriate baseline configuration:
    • For Amazon EC2 deployments reference the below and see this example EC2 calculator for estimation purposes. 
      EKS:
  Control Plane: 1 cluster
  GPU Nodes: 1x p5.48xlarge or p5e.48xlarge (quantity based on workload)
  CPU Nodes: 3x m5.4xlarge
  EBS Volumes: 200GB gp3 per node

RDS:
  Instance: db.m7g.xlarge
  Storage: 64GB
  Engine: PostgreSQL 16.2

Network:
  VPC
  Internet Gateway
  NAT Gateway (optional)
  Data Transfer (estimate based on usage)

S3:
  Standard Storage (estimate based on usage)
    • For Amazon Bedrock deployments, use the platform infrastructure configuration above excluding GPU nodes and add estimated Amazon Bedrock inference costs from the pricing page. See this example Bedrock calculator for estimation purposes.
Key factors that influence total cost:For Amazon EC2 deployments:
  • GPU instance type and quantity
  • Payment model selection (On-Demand vs Reserved)
  • Multi-AZ deployment requirements
  • Data transfer volumes
  • Geographic region selection
For Amazon Bedrock deployments:
  • API request volume
  • Token processing volume
  • Platform infrastructure configuration
  • Data transfer volumes
  • Geographic region selection
Contact your Poolside representative for detailed pricing information and optimization recommendations tailored to your use case. Poolside can design a cost-effective deployment that meets your performance requirements while optimizing AWS infrastructure costs.