AWS Parallel Computing Service features

AWS PCS gives you one set of APIs and a console to build and operate clusters, bringing together compute, storage, networking, login nodes, and visualization. Manage clusters interactively or automate operations with the AWS SDK and CLI.

Provide login nodes for interactive work and remote visualization for graphics-heavy analysis. Connect your own identity system so users sign in with the credentials they already have.

Update a running cluster when your needs change: adjust compute and capacity, change scheduler and accounting settings, and apply scheduler version upgrades in place. Clusters adapt as your requirements evolve, so work keeps moving.

AWS PCS runs your configuration when nodes launch, sourced from Amazon S3 or HTTPS. Nodes mount file systems, install software, and apply settings automatically as the cluster scales, so newly launched capacity comes up ready to run work.

Launch nodes from an AWS-provided, production-ready AMI, including a Deep Learning AMI tuned for GPU workloads, or bring your own custom AMI to set the exact operating system and preinstalled software your workloads need. Make applications available through shared file systems so users run the tools they depend on.

Manage the full cluster lifecycle through infrastructure as code, including AWS CloudFormation and Terraform. Define clusters declaratively, version them alongside the rest of your stack, reproduce environments on demand, and drive create, update, and teardown through the automation your organization already runs.

AWS PCS operates the Slurm controller and cluster control plane for you, applying software and security patches and keeping them healthy. Your teams run their workloads while the service handles the upkeep that keeps the scheduler current and available.

Manage and submit jobs programmatically with the Slurm REST API. Connect PCS to your own portals, pipelines, and automation so users launch and control work the way that fits their environment.

AWS PCS supports Slurm plugins, including SPANK, so you can validate and adjust job submissions and control what happens at job launch. Configure the cluster to run jobs the way your workloads need.

Managed Slurm accounting records job history across your teams and lets you set limits and fair-share policies. Use it for chargeback, reporting, capacity planning, and keeping shared clusters fair.

AWS PCS scales compute up as jobs arrive and back down as they finish, with configurable idle timeouts and per-job scaling. Clusters match spend to real demand so you pay for capacity while your work runs.

Reserve GPU capacity for a future window with Amazon EC2 Capacity Blocks for ML, or hold capacity with On-Demand Capacity Reservations. Plan large training and simulation runs with the accelerated instances in hand when the work starts.

AWS PCS supports current-generation accelerated instances as they become available, so you can scale demanding AI and HPC workloads across large fleets of GPU-based instances.

AWS PCS surfaces cluster health, performance, and job telemetry so you can see how clusters are running. Stream job completion data and metrics to your own monitoring so teams track activity and troubleshoot quickly.