Reitse Eskens performs a comparison:
This blog will be a quite short one compared to the other blogs as it’s more of an overview to show you the capacity of Fabric ingesting CSV files in their native format into a Lakehouse and ingesting SQL data into a table structure inside the Lakehouse. Simple, straightforward stuff without any form of modification. You could call it bronze, raw, ingestion, temp or whatever your preferred naming convention is.
Why is this important? Well, we still have source systems that can only output to files. Just as we still have customers running on SQL Server 2000, legacy or even antique systems are still running. And it’s important to know how much capacity you use when just ingesting data without any modification.
Read on for the two scenarios, giving you an idea of which one is faster. I’d be interested in a third option, which is reading from Parquet files. My initial expectation would be that it would be even faster and more efficient, depending on the structure of the data.