AmazonSageMaker

data-transfer

Hst:Data-Bytes-In

The amount of data (in bytes) transferred into an Amazon SageMaker real-time endpoint.

Hst:Data-Bytes-Out

The amount of data (in bytes) transferred out of an Amazon SageMaker real-time endpoint.

DataTransfer-Regional-Bytes

The amount of data (in bytes) transferred within the same Availability Zone.

DataTransfer-In-Bytes

The amount of data (in bytes) transferred into AWS from the internet. Data transferred into AWS does not incur a fee.

DataTransfer-Out-Bytes

The amount of data (in bytes) transferred out of AWS to the internet.

storage

Host:VolumeUsage

Refers to the underlying hosting instance where the volume is attached. In the context of SageMaker, this could be a notebook instance, training job, or hosting endpoint that uses EBS volumes for storage.

Studio:VolumeUsage

Studio volume charges. Based on the size of the volume and the duration of usage.

Notebk:VolumeUsage

SageMaker Jupyter Notebook volume usage charges. Based on the size of the volume and the duration of usage.

Train:VolumeUsage

Training volume usage charges. Built-in rule storage volumes have no additional charges.

Processing:VolumeUsage

Processing usage charges. For running pre-processing, post-processing, and model evaluation tasks on fully managed infrastructure. Based on the size of the volume and the duration of usage.

Usage

Usage charges based on the size of the volume and the duration of usage.

Processing_DW:VolumeUsage

Data Wrangler processing jobs volume usage charges. Based on the size of the volume and the duration of usage

TrainDebugFreeTier:VolumeUsage

Free Tier volume usage for Training Debugger. 50 hours of m4.xlarge or m5.xlarge instance usage included for first 2 months.

Cluster:VolumeUsage

Cluster volume charges. Based on the size of the volume and the duration of usage.

TrainWarmPool:VolumeUsage

Volume usage charge for training warm pools. A training warm pool allows you to keep trained models and related resources, such as compute instances, ready for use in case you need to retrain or make updates to the model quickly. By maintaining this "warm" state, you can avoid the need to fully launch and initialize resources each time. Based on the size of the volume and the duration of usage.

ml-instance

Host:ml

Specifies that the instance is being used for hosting a deployed machine learning model in Amazon SageMaker. Pricing is based on the selected instance type.

Studio:JupyterLab

Studio IDE for notebooks and code. Pricing is based on the selected instance type.

Notebk:ml

Notebook instances that run in a Jupyter Notebook. Pricing is based on the selected instance type.

Studio:KernelGateway

Studio notebook usage costs. Pricing is based on the selected instance type.

Train:ml

Training models. Built-in rules are free. Custom rules are charged based on the selected instance type.

Processing:ml

For running pre-processing, post-processing, and model evaluation tasks on fully managed infrastructure. Pricing is based on the selected instance type.

Usage

ML instance usage charges.

Studio_DW:KernelGateway

Usage cost for Studio Data Wrangler. Pricing is based on the selected instance type.

Tsform:ml

Charges for transform jobs, indicating a Batch Transform job. Used to apply a trained model to datasets in bulk. Pricing is based on the selected instance type.

Studio:CodeEditor

Code editor for ML code, integrated with Studio. Pricing is based on the selected instance type.

AsyncInf:ml

Solution for handling inference requests by queuing and processing them asynchronously. This option is ideal for use cases involving large data payloads or models with lengthy processing times that do not require immediate response speeds. Pricing is based on the selected instance type.

TensorBoard:TensorBoard

Hosted TensorBoard solution for debugging model convergence issues in SageMaker training jobs. Pricing is based on the selected instance type.

Processing_DW:ml

Data Wrangler processing jobs. Job instance type pricing is calculated per instance hour.

TrainDebugFreeTier:ml

Free Tier charges for SageMaker Debugger. 50 hours of m4.xlarge or m5.xlarge usage per month for the first 2 months included with Free Tier.

TrSpt:ml

Training models that use managed EC2 Spot instances. Savings of up to 90% over on-demand instances. Pricing is based on the selected instance type.

Cluster:ml

Refers to the compute infrastructure or cluster used for a SageMaker job. This typically relates to managed training or inference clusters. Pricing is based on the selected instance type.

Cluster-C14-3Yr-NUP

Specific code for cluster. Pricing is based on the selected instance type.

amazon-sagemaker

MLflow:TrackingServerCompute

Per unit compute charge for server that's required to track ML experiments with SageMaker AI and MLflow. Pay only for what's used. Compute charges based on size and number of running hours.

automl-jobs

Canvas:Session-Hrs

Payment based on the number of hours you're logged in to or use SageMaker Canvas. Session hours are calculated from the time you log in to the time that you log out.

featurestore-storage

FeatureStore:TimedAndPITRStorage

Charges for Feature Store storage based on GB of data stored per month.

ml-serverless

ServerlessInf:Mem

Serverless Inference memory charge. Deploy models without needing to configure any infrastructure. Memory charges depend on size and price per millisecond.

ProvisionedConcurrency:Mem

Provisioned Concurrency on Serverless Inference memory charge. Charges based on memory size.

ProvisionedConcurrency:Usage

Provisioned Concurrency on Serverless Inference usage charge. Charged per second of usage.

featurestore-payperrequestthroughput

FeatureStore:WriteRequestUnits

Write request charges for SageMaker Feature Store. Pricing varies for standard online store vs. in-memory online store. Standard on-demand and Standard in-memory pricing for writes is per million write request units. Standard provisioned is per write capacity unity (WCU) hour.

FeatureStore:ReadRequestUnits

Read request charges for SageMaker Feature Store. Pricing varies for standard online store vs. in-memory online store. Standard on-demand and Standard in-memory pricing for reads is per million read request units. Standard provisioned is per read capacity unit (RCU) hour.

fee

Geospatial:MonthlyUserFee

Monthly fee for Geospatial ML with Amazon SageMaker. Per user fee for access to SageMaker Studio notebook, which includes preloaded geospatial imagery. This fee also provides access to a geospatial data catalog and a dedicated geospatial container running on SageMaker Processing.