vroom

Benchmarks show vroom reads at 1.23 GB/sec, up to 53x faster than base R, by indexing files and using multi-threading for parsing non-character columns.

Uses the Altrep framework to load data only when accessed, minimizing memory usage for large datasets without code changes.

Adds capabilities over readr like column selection (similar to dplyr::select()), big integer support, and native reading from multiple files or connections.

Leverages multiple threads for indexing, writing, and parsing, with environment variables to control thread count for optimization.