01-cluster.yaml

# Cost-Optimized EKS cluster for mining with spot GPU instances 
# Built in efforts to reducing  costs of ML workloads and increase hash rate.
# This spec creates a cluster on EKS with the following active nodes 
# +---------------+-------------------------------------+-------+--------+-----------+-------------------+--------------------+
# | Instance Name | GPUs series                         | GPUs  | vCPUs  | RAM (GiB) | GPU Memory (GiB)  | Spot Price         | 
# +-----------+-----------------------------------------+-------+--------+-----------+-------------------+--------------------+
# | g4dn.xlarge   | NVIDIA T4 Tensor                    | 1     |   4    | 16        |                   | $0.157 per Hour    |     
# | g2.2xlarge    | NVIDIA GRID K520 (Kepler)           | 1     |   8    | 64        |                   | $0.195 per Hour    |     
# | g3s.xlarge    | NVIDIA Tesla M60                    | 1     |   4    | 30.5      |        8          | $0.225 per Hour    |     
# | p2.xlarge     | NVIDIA K80                          | 1     |   4    | 61        |        12         | $0.270 per Hour    |     
# | p3.2xlarge    | NVIDIA Tesla V100                   | 1     |   8    | 61        |        16         | $0.918 per Hour    |     
# +---------------+-------------------------------------+-------+---------+----------+-------------------+----------------- --+

apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: cluster-1    # Name of your cluster, change to whatever you find fit.
  region: us-east-1  # choose your region wisely, this will significantly impact the cost incurred
  version: "1.18"
availabilityZones: ["us-east-1a", "us-east-1b",  "us-east-1c",  "us-east-1d"]   
vpc:
  clusterEndpoints:
    privateAccess: false # Nothing is secret on iur cluster
  nat:
    # NatGateway is an AWS managed instance that permits Internet traffic from instances sitting in a private subnet inside your VPC. 
    # NAT Gateway resource and that’s billed 24/7
    # Disable for reducing costs
    gateway: Disable # other options: Disable, Single (default)

managedNodeGroups:
  - name: gpu-spot-g4dn-xlarge
    spot: true
    minSize: 2
    maxSize: 3
    instanceTypes: ["g4dn.xlarge"]
    amiFamily: AmazonLinux2    
    labels:
      workgroup: g4dn-xlarge
      lifecycle: Ec2Spot
      aws.amazon.com/spot: "true"
      gpu-count: "1" 
    tags:
      k8s.io/cluster-autoscaler/node-template/label/lifecycle: Ec2Spot
      k8s.io/cluster-autoscaler/node-template/label/aws.amazon.com/spot: "true"
      k8s.io/cluster-autoscaler/node-template/label/gpu-count: "1"
      k8s.io/cluster-autoscaler/node-template/label/workgroup: g4dn-xlarge
      k8s.io/cluster-autoscaler/enabled: "true"   
    iam:
      withAddonPolicies:
        autoScaler: true