Instance Types & Configuration

HPC Instance Types

Common HPC Instances

Instance Cores Memory (GiB) Network (Gbps) Best For
c6a.xlarge (AMD) 2 (4 vCPUs) 8 12.5 Head node
hpc6a.48xlarge (AMD) 96 368 100 (EFA) HPC workloads (CFD, MD, WRF)
c6i.32xlarge (Intel) 64 (128 vCPUs) 256 50 (EFA) General HPC, multi-core
hpc7g.16xlarge (Graviton) 64 128 200 (EFA) ARM-based HPC

GPU Instances

Instance vCPUs Memory (GiB) GPU Best For
g5.xlarge 4 16 1x A10G ML inference, visualization
p4d.24xlarge 96 1152 8x A100 Large-scale ML training

Storage Options

Type Performance Persistence Notes
FSx for Lustre (SCRATCH_2) Very high throughput Deleted with cluster Best for HPC I/O
FSx for Lustre (PERSISTENT) Very high throughput Survives cluster deletion Higher cost
EBS (gp3) Standard SSD Deleted with cluster Head node root volume
S3 High throughput for bulk Persistent Use for long-term data storage

Config File Reference

The full ParallelCluster configuration specification is available at: AWS ParallelCluster Configuration Reference

Key Config Sections

Section Purpose
HeadNode Instance type, networking, DCV, storage
Scheduling Scheduler type, queues, compute resources
SharedStorage FSx for Lustre, EBS, EFS configuration
Image Operating system selection
Region AWS region for the cluster