Direct Lake is cheaper than Import in my tests, and the reason is Import's refresh. But Direct Lake's DAX queries cost more, and it adds an overhead on the Lakehouse that you won't see if you only watch the semantic model. Measured across nine replications of an 8-hour workday, on F64 and F2.

Tento článek se zaměřuje na to, jak Direct Lake funguje v praxi, především ve variantě Direct Lake on OneLake. Věnuje se framingu Delta tabulek, Parquet row groups, transcodingu do VertiPaq, residency a eviction segmentů, prewarmingu a spotřebě CU ve srovnání s Import režimem.

This article examines how Direct Lake works in practice, with a focus on Direct Lake on OneLake. It covers Delta-table framing, Parquet row groups, transcoding into VertiPaq, segment residency and eviction, prewarming, and CU consumption compared with Import.