Amazon Redshift rg.large instances, powered by AWS Graviton processors, now support single-node clusters. ^[1] Onur Satici stated that Vortex achieves approximately 30 times faster S3-to-GPU scans than Parquet and approximately 100 times or more faster random access than Parquet, thanks to its encodings and compression. ^[2] Customers can now create a single-node rg.large cluster for smaller workloads that do not require high availability, offering a cost-effective option to conduct proofs of concept and tests quickly. ^[3] RG instances deliver up to 2.4x faster performance running data warehouse and data lake workloads when compared to previous generation RA3 instances, at 30% lower price per vCPU. ^[4] Single-node support for rg.large clusters is available on P204 or later patch versions. ^[5] Onur Satici claimed that Vortex uses lightweight and cascading encodings instead of block compression, which allows compute on compressed data, random access, and aggregations, and that some of these computations can be faster than on decompressed data. ^[6] Onur Satici described a 'decision tax' in ML training iteration, where changing a data mix or filter with current tools requires reprocessing the entire data (re-tokenizing, reshuffling, and reloading), whereas with Vortex, users can simply change the scan, query, or filter without reprocessing. ^[7] Onur Satici explained that Vortex is a columnar file format similar to Parquet but with an extensible spec, a separation of logical types from physical types, and a design that does as little as possible as late as possible, pushing projections, filters, and aggregations from the compute engine into the file format. ^[8]