Performance — model size, query folding, aggregations

A slow Power BI report has a small set of causes. The diagnostic tools — Performance Analyzer in Desktop and DAX Studio as a separate free download — make every cause visible. Without them you'll fix things at random. With them, optimisation becomes mechanical.

Performance Analyzer

View → Performance Analyzer → Start Recording → interact with the report → Stop. Each visual shows three times: DAX query time, visual display time, and others. Sort by total time. The slowest visual gets investigated first. Click 'Copy query' next to a slow visual to capture its DAX for further analysis in DAX Studio.

The seven most-effective fixes

• Star schema, not flat table — Performance gains 5-10x.
• Disable Auto Date/Time (Options → Data Load → uncheck Auto Date/Time) — this auto-generated date hierarchy bloats models massively.
• Use measures, not calculated columns, wherever possible — calculated columns persist in memory; measures don't.
• Single-direction relationships unless you need bi-directional — fewer hidden filter paths to traverse.
• Remove unused columns from imported tables — every column eats memory.
• Aggregations on top of large fact tables — Power BI auto-routes visuals to the aggregation when possible (Premium feature).
• Variables in DAX — capture once, reuse many times; reduces query plan complexity.

Import vs DirectQuery vs Composite

• Import: data loaded into Power BI's columnar engine (Vertipaq). Fastest queries. Refreshed on schedule. Most analyst reports.
• DirectQuery: every visual interaction issues a live query to the source database. Always fresh, slow on each interaction. Used when data must be real-time or the dataset is too big to import.
• Composite: some tables Import, others DirectQuery. Aggregations on top of DirectQuery tables. Power BI Premium feature for large enterprise scenarios.