FROM S3 TO GPU IN ONE COPY

Summary from the TLDR Dev newsletter

Vortex streams compressed columnar data from S3 through host memory to the GPU with a single copy, avoiding disk staging and CPU decompression bottlenecks. Its layouts prune data before transfer, and its encodings support GPU-side decompression and computation, producing scans around 30 times faster than Parquet in the demo.