Pricing overview
AWS Parallel Computing Service (AWS PCS) is a managed service for running high performance computing (HPC) and scientific and engineering modeling workloads on AWS. Usage of AWS PCS will incur two separate charges—an hourly cluster controller fee and an hourly node management fee for every Amazon Elastic Compute Cloud (Amazon EC2) instance launched within your AWS PCS compute node groups. The controller fee will vary by the cluster controller size chosen to manage the cluster. The node management fee is tiered—an advanced tier for Amazon EC2 UltraCluster instance types (P and TRN instance families) and a standard tier for all other Amazon EC2 instance types. Both fees are rounded up to the nearest minute. You only pay for what you use, as you use it.
Slurm Accounting is an optional feature in PCS. If you decide to enable this feature, you will incur two additional charges – an hourly accounting usage fee that will vary by the controller size chosen, and an accounting storage fee that is billed in per GB-month increments.
In addition to the AWS PCS service fees, you will be billed separately for all the AWS resources used by your cluster. For example, you will be charged for Amazon EC2 instances used across your cluster as well as for applicable data transfer charges. For more information, visit the Amazon EC2 pricing page.
AWS PCS Cluster Controller Sizes for Slurm
When creating an AWS PCS cluster, you will be asked to select a cluster controller size and can choose based on the following cluster-level requirements. If your workload requirements are greater than what the Large size offers, contact us. To learn more about choosing the right cluster size, refer to the AWS PCS User Guide.
| Slurm Cluster Controller Size | Number of Instances Orchestrated | Number of Active & Queued Jobs |
Small |
Up to 32 |
Up to 256 |
Medium |
Up to 512 |
Up to 8,192 |
Large |
Up to 2,048 |
Up to 16,384 |
AWS PCS fees – Pricing
Pricing examples
Example 1: Running an Always On cluster in US East (N. Virginia)
Let’s say you run a 500-instance HPC cluster 24/7 for a month, provision 2 compute node groups to scale up instances across the C, HPC, and R instance families, and plan to run and queue a maximum of 800 jobs at any point in time to meet your workload needs in the US East (N. Virginia) AWS Region. You choose to deploy a Medium size AWS PCS controller, opt-in to the Slurm Accounting feature, and store an average 10GB of accounting data. In one month, your bill will be the following:
Controller fee charges: $3.2579/hour * 24 hours/day * 30 days/month = $2,345.688
Node management fee charges: 500 instances * $0.08/instance/hour * 24 hours/day * 30 days/month = $28,800.00
Accounting usage fee charges: $0.98/hour * 24 hours/day * 30 days/month = $705.60
Accounting storage fee charges: $0.81/GB/month * 10 GB = $8.10
Total monthly bill: controller fee charges + node management fee charges + accounting usage fee charges + accounting storage fee charges = $31,859.388
Note: AWS PCS charges are in addition to compute, storage, and other underlying resources used.
Example 2: Running a spiky daily workload in US East (N. Virginia)
Let’s say you run an HPC workload spanning 5 jobs that run for 2 hours a day, 5 days a week, every week in the US East (N. Virginia) Region. The workloads are spiky and run a maximum of 400 instances but on average run 200 instances of which 10 are in the P or TRN families. You choose to deploy a Medium size AWS PCS cluster, don’t enable the Slurm Accounting feature, and the cluster controller runs the whole month. In one month, your bill will be the following:
Controller fee charges: $3.2579/hour * 24 hours/day * 30 days = $2,345.688
(Standard) Node management fee charges: 190 instances * $0.08/instance/hour * 2 hours/day * 5 days/week * 4 weeks/month = $608.00
(Advanced) Node management fee charges: 10 P instances * $0.64/instance/hour * 2 hours/day * 5 days/week * 4 weeks/month = $256.00
Total monthly bill: controller fee charges + node management fee charges = $3,209.688
Note: AWS PCS charges are in addition to compute, storage, and other underlying resources used.