AWS Parallel Computing Service FAQs

AWS PCS is a managed service for running high performance computing (HPC) workloads with Slurm. It operates the Slurm controller and scaling logic for you and launches compute in your own AWS account, so your teams run simulations and models using standard Slurm workflows while AWS manages the cluster control plane.

AWS PCS is available in multiple AWS Regions, including AWS GovCloud (US). For the current list of Regions where AWS PCS is available, see the AWS Regional Services List.

AWS PCS supports Slurm, a widely used open source scheduler and workload manager for distributed HPC. You interact with your clusters using standard Slurm commands and workflows, so your users work in a familiar environment.

Slurm is a popular open source scheduler for managing distributed HPC workloads.

AWS PCS applies minor Slurm patches automatically as part of the managed control plane's ongoing maintenance. Major version upgrades run in place with a single API call. Long-running jobs continue through the transition.

You define your queues and compute configurations, and AWS PCS provisions a managed, highly available Slurm controller and runs the scaling logic to match the demand in your job queues. You manage clusters through the AWS PCS API, CLI, the AWS Management Console, AWS CloudFormation, or Terraform.

AWS PCS is built for scientific and engineering modeling, simulation, and high performance data analytics. Common workloads include computational fluid dynamics, weather modeling, finite element analysis, electronic design automation, and reservoir simulation, across industries such as automotive, aerospace, energy, semiconductors, life sciences, and weather.

AWS provides a Service Level Agreement for AWS PCS. If service availability falls below the committed levels, you may be eligible for service credits. For the current commitment and terms, see the AWS PCS Service Level Agreement.

AWS PCS supports nearly all of the Amazon EC2 instance types available in the Region where you use AWS PCS.

Yes. You can run workloads on AWS PCS using GPU, AWS Trainium, and AWS Inferentia instance types, including Amazon EC2 UltraServers.

AWS PCS launches and manages Amazon EC2 instances in your account through compute node groups. You set each node group's instance types and a minimum and maximum instance count, and map node groups to Slurm queues. As jobs queue, AWS PCS scales a group up toward its maximum; as jobs finish, it scales back down after a configurable idle timeout, so capacity follows demand.

Yes. AWS provides production-ready AMIs to build on, including a PCS-ready Deep Learning AMI tuned for GPU workloads, or you can bring your own custom AMI. For the AMI specification and sample AMIs, see the AWS PCS documentation.

AWS PCS supports several Linux distributions and versions. For the current list of supported operating systems, see the AWS PCS documentation.

With AWS PCS, you can create compute and login node groups that launch Amazon EC2 instances either in a single Availability Zone or across multiple Availability Zones. When you configure a node group across multiple Availability Zones, PCS looks across all of them for available capacity as it launches instances.

AWS PCS supports Amazon EFS, Amazon EBS, Amazon FSx for Lustre, Amazon FSx for NetApp ONTAP, Amazon FSx for OpenZFS, Amazon S3, Mountpoint for Amazon S3, and Amazon File Cache. You can also connect to self-managed and third-party file systems. For details, see the AWS PCS documentation.

Yes. AWS PCS supports a wide range of Amazon EC2 instances with advanced networking, including Elastic Fabric Adapter (EFA). It also supports AWS PrivateLink and Amazon VPC endpoints.

Yes. You can configure your AWS PCS compute node groups to work with directory services such as Microsoft Active Directory, Microsoft Entra ID, and OpenLDAP.

AWS PCS charges have a few components. You pay an hourly controller fee based on the size of the managed Slurm controller you choose for a cluster, and a node management fee for each compute instance AWS PCS manages, with separate rates for standard instances and for accelerated instance families. If you enable the optional Slurm accounting feature, you also pay an accounting usage fee and an accounting storage fee. These AWS PCS charges are in addition to the Amazon EC2, storage, and other AWS resources you use, which are billed at their standard rates. There are no upfront commitments. For current rates and worked examples, see the AWS PCS pricing page.

Savings Plans apply automatically to the Amazon EC2 instances that AWS PCS launches in your account. You can also target reserved capacity by setting a compute node group's purchase option: On-Demand Capacity Reservations, Amazon EC2 Capacity Blocks for ML, or interruptible capacity reservations. You configure this per node group through the AWS PCS API, CLI, the AWS Management Console, AWS CloudFormation, or Terraform.

Yes. AWS PCS applies AWS tags at both the cluster and compute node group level, so you can track your Amazon EC2 spend at those granularities.

AWS PCS runs compute in your own AWS account and VPC. It uses AWS Identity and Access Management (IAM), including service-linked roles and AWS managed policies, for access control, and API activity is recorded to AWS CloudTrail for auditing. You control network access using your VPC configuration and security groups, and you can access the AWS PCS API privately using AWS PrivateLink.

Yes. AWS PCS encrypts your data at rest by default using an AWS owned AWS Key Management Service (KMS) key, and you can choose a customer managed key instead. Connections to the AWS PCS API use TLS, and communication between the managed controller and your cluster nodes is encrypted. Traffic among your cluster nodes stays within your VPC and is protected by Amazon EC2 encryption in transit on the instance types that support it. For details, see the AWS PCS documentation.

Third-party auditors assess AWS PCS as part of the AWS compliance programs. For the current list of compliance programs that include AWS PCS, see AWS Services in Scope by Compliance Program, and download audit reports from AWS Artifact. Your own compliance responsibility depends on the sensitivity of your data and the laws and regulations that apply to you.

AWS PCS publishes cluster capacity and utilization metrics to Amazon CloudWatch in the AWS/PCS namespace, including desired, actual, idle, and utilized capacity per cluster, at one-minute intervals retained for 15 months. Use them to see how your cluster scales and how fully it is used.

AWS PCS can expose a Prometheus-compatible OpenMetrics endpoint for real-time visibility into jobs, nodes, and scheduling. You turn it on for a cluster and scrape it with the Prometheus-based tools and dashboards you already use, alongside the metrics AWS PCS publishes to CloudWatch.

Yes. AWS PCS can deliver control-plane and audit logs to Amazon CloudWatch Logs, Amazon S3, or Amazon Data Firehose. You can also configure the Amazon CloudWatch agent on your compute nodes to send node-level logs, such as Slurm daemon and PCS bootstrap logs, to CloudWatch Logs.

If a compute node has a problem while a job is running, AWS PCS can automatically reschedule the affected job at the highest priority. The job restarts as soon as healthy capacity is available, so a single node issue does not send the work to the back of the queue.

To get started, visit the AWS PCS console. You must have an AWS account to access this service. If you do not have an account, you will be prompted to create one. After signing in, visit the AWS PCS documentation page to access the getting started guide.