Separate Power BI Query Time From Visual Display and Queue Time
“The report is slow” leaves several different causes unresolved. A visual may be waiting for data, drawing its result or waiting behind other work. Capture one repeatable interaction before changing hardware or rewriting calculations.
Choose a local report copy and a specific action, such as changing one slicer from East to West. Record the report page, starting selection and visible visuals. Use the same starting state for each comparison so a different workload does not masquerade as a performance change.
Record and classify the delay
In Power BI Desktop, choose Optimize, Performance Analyzer, Start recording. Perform the interaction, then Stop and expand its visual entry. DAX query measures the wait for model results; Direct query measures applicable external queries. Visual display covers drawing and related retrieval. Other includes preparation, waiting and background work.
Reported durations are elapsed intervals and can include queued time. Microsoft notes that many canvas and visual operations share a sequential UI thread. Do not add every displayed duration and present the sum as independent processor work, or assume Other identifies one faulty calculation.
Repeat the same selection from the same starting state and retain both observations. If one run differs substantially, record that variation. A single favourable measurement is insufficient to describe a repeatable improvement.
Choose the next investigation from the evidence
If the query portion dominates repeatedly, inspect the relevant calculation and model path. If display time dominates, examine what the visual must render or retrieve. If Other is substantial, consider the neighbouring workload and waiting rather than immediately blaming the underlying data source. These are directions for further investigation, not diagnoses established by a category name alone.
Change one relevant feature in the working copy and repeat the identical action. Compare the same visual and categories while also confirming that the displayed business result still agrees. A quicker visual that omits required records is not an accepted repair.
The pane can export its information as a JSON file; Clear removes the collected information. Save useful evidence before clearing it, and inspect diagnostic contents before sharing because queries and model details may be sensitive. Keep the report state and action with the timings. The handover should identify where the measured delay occurred and what changed, without claiming that one local run predicts every user's experience.
Sources: Microsoft: Performance Analyzer.