Amazon Redshift RG Instances: Graviton-Powered Performance and Unified Data Lake Querying

Amazon Redshift has consistently evolved to meet the growing demands of data analytics, from dense compute nodes to RA3 instances and serverless options. The latest innovation introduces RG instances, powered by AWS Graviton processors, which deliver up to 2.2x faster performance than RA3 at 30% lower cost per vCPU. What truly sets them apart is the integrated data lake query engine, enabling you to query both warehouse tables and Amazon S3 data lakes from a single system. This blend is tailor-made for high-volume workloads, including those from AI agents. Below, we answer key questions about this new instance family.

What are Amazon Redshift RG instances?

Amazon Redshift RG instances are a new family of compute nodes built on AWS Graviton processors. They replace RA3 instances as the recommended option for most workloads, offering better price-performance. The RG family includes sizes like rg.xlarge (4 vCPU, 32 GB memory) for small clusters and rg.4xlarge (16 vCPU, 128 GB memory) for standard production workloads. These instances are designed to handle both traditional data warehouse queries and data lake analytics through an integrated engine that natively queries Apache Iceberg and Parquet formats on Amazon S3. This unification means you no longer need separate systems for structured and semi-structured data, reducing operational complexity.

Amazon Redshift RG Instances: Graviton-Powered Performance and Unified Data Lake Querying — Source: aws.amazon.com

How do Graviton processors improve Redshift performance?

AWS Graviton processors are custom-built by Amazon using 64-bit Arm cores, optimized for cloud workloads. In Redshift RG instances, Graviton delivers up to 2.2x faster query performance compared to equivalent RA3 instances (based on Intel Xeon). This speed boost comes from better instruction efficiency and lower latency memory access. Additionally, Graviton enables a 30% reduction in price per vCPU, making each query cheaper. For data lake queries, the integrated engine leverages Graviton’s capabilities to achieve up to 2.4x faster performance for Apache Iceberg and up to 1.5x for Apache Parquet versus RA3. The combination of faster processing and lower cost directly benefits BI dashboards, ETL pipelines, and AI agent workloads that require many concurrent queries.

What is the integrated data lake query engine?

The integrated data lake query engine is a key feature of RG instances that allows you to run SQL analytics across both your data warehouse tables and your Amazon S3 data lake from a single Redshift endpoint. It uses Amazon Redshift Spectrum technology but is now tightly coupled with the compute layer, eliminating the need to move data between systems. You can query open table formats like Apache Iceberg and Apache Parquet directly, and the engine automatically optimizes query plans to push down processing to S3 where possible. This integration reduces total analytics costs by avoiding duplicate storage and ETL overhead, simplifies operations with a single query interface, and accelerates insights from diverse datasets stored cost-effectively in the data lake.

Why are RG instances ideal for AI agent workloads?

AI agents query data warehouses at scales far beyond typical human usage, often generating thousands of low-latency SQL queries per second. This can drive up operational costs and degrade performance. RG instances are built to handle such demands because they offer 2.2x faster query execution and 30% lower cost per vCPU compared to RA3. The integrated data lake engine further reduces the need to pre-process data before querying, enabling agents to access fresh data from the lake instantly. Moreover, Amazon Redshift’s recent improvements (March 2026) sped up new queries by up to 7 times, which directly benefits autonomous, goal-seeking AI agents that require near-real-time analytics. The overall blend of speed, cost efficiency, and unified querying makes RG instances a strong foundation for agentic AI workloads.

How much can I save by migrating from RA3 to RG instances?

Amazon Redshift RG instances offer a 30% lower price per vCPU compared to RA3 instances. For example, an rg.4xlarge (16 vCPU, 128 GB memory) provides 33% more vCPU and memory than the ra3.4xlarge (12 vCPU, 96 GB) at a lower per-vCPU cost, delivering better value for standard production workloads. Additionally, because queries run up to 2.2x faster, you may need fewer compute resources to achieve the same throughput, further reducing costs. The integrated data lake engine also lowers total analytics spend by eliminating the need for separate query engines or ETL jobs to move data between your warehouse and S3 data lakes. To estimate savings, use the AWS Pricing Calculator with your specific workload patterns—it factors in instance size, data transfer, and storage costs.

How do I get started with Amazon Redshift RG instances?

You can launch new Redshift clusters with RG instances or migrate existing RA3 clusters through the AWS Management Console, AWS CLI, or AWS API. The integrated data lake query engine is enabled by default on RG instances, so you immediately gain the ability to query S3 data lakes without extra configuration. For existing clusters, Amazon Redshift provides a simple resize operation to change instance types—though you should first test in a non-production environment. Refer to the what are RG instances section for instance size recommendations. AWS also recommends using the AWS Pricing Calculator to compare costs before migrating. As always, you can scale storage independently of compute with Redshift managed storage, ensuring flexibility as your workloads evolve.