Amazon Redshift Unveils AWS Graviton-based RG Instances and Integrated Data Lake Query Engine

Amazon Redshift has unveiled AWS Graviton-based RG instances. Compared to RA3, data warehouse workloads are up to 2.2x faster, and price per vCPU is 30% lower. The integrated data lake query engine is enabled by default, querying the warehouse and Amazon S3 data lake together within a single engine.

Representative mappings are as follows.

• ra3.xlplus → rg.xlarge: 4 vCPU, 32GB, small departmental analytics
• ra3.4xlarge → rg.4xlarge: 12 → 16 vCPU, 96GB → 128GB, standard production/mid-size data

In terms of performance, Apache Iceberg queries are up to 2.4x faster than RA3, and Apache Parquet queries are up to 1.5x faster. Creating new clusters and migrating existing clusters are supported via the console, AWS CLI, and API, and existing external tables, schemas, and Spectrum query syntax remain unchanged.

There are two migration paths.

• Elastic Resize: In-place migration with 10-15 minutes of downtime for compatible configurations
• Snapshot and Restore: Creates a new RG cluster from an RA3 snapshot, suitable when configuration changes are needed

Since data lake queries run on cluster nodes, Amazon Redshift Spectrum is no longer required. Queries are processed within the VPC boundary and use existing IAM roles, and the $5 per terabyte scan charge is also eliminated.

RG instances are available in multiple regions including the US, Asia Pacific (including Seoul), Canada, Europe, and South America. However, in the May 12, 2026 update, Middle East (UAE) was removed from the supported list.