# Parallelism without row groups #reading-list **URL:** https://blog.lancedb.com/file-readers-in-depth-parallelism-without-row-groups/ ## Summary Weston's main point is that row groups create natural boundaries for parallelism, but not the kind you care about. You're better off throwing all of your machine's parallelism at decoding all of the pages of a single "row group" rather than decoding a bunch of row groups one page at a time. You end up having the same bandwidth but the the latency to first row group is much lower.