Power BI Performance Analyzer: Separate a slow visual from its DAX query
A repeatable measurement plan to distinguish DAX time, rendering and waiting in a slow Power BI report page.
Define the interaction that feels slow
Record the page, role, filter and action before changing the report. Power BI Performance Analyzer reports how individual visuals load during a recording. A total can include time for DAX, visual display and other waiting. Treat the first visit, a filter change and a repeated warm interaction as separate cases. That makes a performance claim reproducible rather than anecdotal.
Repeat a controlled measurement
In Desktop, open Performance Analyzer from Optimize, start recording and perform the same filter change. Repeat under comparable data and permissions. Compare the components as well as total duration. Long DAX time prompts an examination of model and measure behavior. A short query with heavy visual display points toward rendering or the number of categories. Record the circumstances alongside the numbers.
Change one variable without changing the answer
Try reducing rows in an overly detailed table while keeping the business total. Compare group values and totals under the same filters. Copy a visual’s DAX query when appropriate and reduce it to a test case. Do not count a speedup if it removes data readers require. Compare any revised calculation against a small, understandable set of expected values.
Write down a release decision
Store the action, model version, role, filters, several durations and a correctness check. Test Desktop and service results separately because run conditions differ. The improvement should persist for the measured interaction while preserving its meaning. Check key alternative filters and roles. This gives maintainers a traceable regression case rather than an isolated baseline with no context.
Validate a measure before publishing it
A plausible number is not enough for a DAX measure. Build a small test matrix across dates, products and user filters. Check row context and filter context carefully, including grand totals. State the business question before selecting CALCULATE, ALL or other filter-changing functions. Repeatable tests detect subtle mistakes long before the model becomes widely used.
Complete the decision with performance and access tests
A measure should be correct, fast enough under production data volumes and consistent with row-level security. Test concurrent queries, realistic filter combinations and different identities. Compare results with an independent source query or control total. Preserve the previous version and verify numerical differences rather than relying on screenshots alone.
Comments
No comments yet. Be the first to contribute.
Discussion question from the editor In your slowest report interaction, is DAX or rendering the larger share?
Reading “Power BI Performance Analyzer: Separate a slow visual from its DAX query” raised this question for me: results should be tested under different filters.
Editor note: Results can vary by scale. Cost, latency and user experience should be compared together.
The learning curve and accessibility also matter for actual users.
A practical case study with load testing or a pilot would help establish the trade-offs.
✎ Write a comment