27. Performance Tips

27.3 Reduce Model Size and Cardinality

By Ravindra Bagale · Last updated: 25 September 2026

Power BI's VertiPaq engine stores data by column and compresses it. Compression (संक्षेपण) is best when a column has few distinct values (low cardinality). So:

• Remove unused columns – the biggest win. Every column costs memory, even if nobody uses it.
• Remove unnecessary rows – e.g. load only the last 3–5 years if older history is not needed.
• Split Date/Time columns into a Date column and a Time column (or drop the time if not needed). A timestamp with seconds (like Order DateTime) can have millions of unique values; a date column has a few thousand, and an hour column only 24.
• Round decimals to the precision you need (e.g. 2 decimal places).
• Avoid high-cardinality text columns such as comments, GUIDs or unique transaction descriptions unless really needed.
• Use integer keys instead of long text keys where possible.
• Turn off Auto date/time.
• Prefer calculated columns in Power Query or the source rather than DAX calculated columns (they compress better and keep the model simpler).
• Use summarised tables or aggregations (एकत्रीकरण) for very large fact tables.