Forum Discussion
How do I improve render speed?
I have a report with 7 "Infographic Designer" visualisations, 3 "Tachometer" visualisations, and 9 "Card" visualisations. It takes 22 seconds to render.
When I make a copy of the report and replace all the "Infographic Designer" and "Tachometer" visualisations with "Table" visualisations it renders in under a second.
Obviously I'd like to present the prettier version of the report.
What can I do to improve the performance?
13 Replies
Some things to keep in mind here. With 22 separate visualisations, your report is likely having to make 22 separate sets of calculations. When you brought them all into a singular table, you allowed Power BI to optimise how many calculations it needed to make.
So really you have 2 options:
- You can either optimise how you display the data by reducing the number of visuals
- You can optimise the Measures these visualisations rely upon.
As a quick pointer toward something to optimise. Have a look at your measures and see if, within that same measure, you refer to a set of data more than once. If that is the case, you can front load calculations into a variable and use that result throughout that measure. An example of this:
ExampleA = IF(Sum[Column] < 5, 5, Sum[Column])
Could be optimised to:
ExampleB =
VAR Result = Sum[Column] RETURN IF(Result < 5, 5, Result)Recognizing that none of the values to be displayed were reliant on filters or sliders I had already moved them all to columns. If I understand correctly, they should all be computed when the dataset is imported and not at render time.
That leaves me at option 1 - reduce the number of visualisations by spreading them out across multiple pages.
Quite a few of my reports have 10 to 12 visuals on them and all render fairly fast. The only exception is when i'm displaying graphs that have to calculation on many millions of rows of data. Without knowing your dataset I'm now quite curious what sort of data you are using and what visualisations you are creating using only calculated columns and still getting poor performance. I would have to imagine you either have a tonne of sizable tables with lots of complex relationships? Can you optimise your dataset in anyway instead?
General best practise is that if you have a choice between a calculated column and a measure, the measure should be your first choice. I would be concerned if you feel you need to shy away from them.
