Forum Discussion
Would Enabling Large Semantic Model Storage Format Improve Refresh Performance?
Hi manoj_0911 ,
Yes, you can enable it as its recommended by Msft and the semantic model can grow beyond the size during refresh and it handles the XMLA operarion in a better way . While it will not reduce the load on your SQL server, it provides significant backend benefits for your Fabric/Premium capacity:
- SQL Server Load: Zero impact. It does not change how data is queried or extracted from the source.
- Refresh Efficiency: Much faster and safer "commit" phase. It writes to disk incrementally during the refresh, significantly reducing memory spikes and Out-Of-Memory failures.
- Capacity Health: Datasets load back into memory faster for end-users, improving overall Premium capacity performance.
- The Only Downside: Once enabled, you cannot download the dataset back to a local `.pbix` file. (Always maintain the latest copy of the PBIX in sharepoint or a common drive)- Bottom line: Enable it to future-proof your dataset's reliability and protect your capacity's memory as your 3-year archive grows.
https://learn.microsoft.com/en-us/fabric/enterprise/powerbi/service-premium-large-models?source=docsThanks!
Natarajan Manivasagan
If you found this helpful, please consider giving it a Kudos and marking it as the accepted solution — it goes a long way in helping others facing the same issue.🏆 Best Solution for Enterprise BI — 2026 Microsoft Fabric Semantic Link Developer Experience Challenge
👉 Microsoft announcement · View the winning notebookFor more Power BI tips and discussions, let's connect on LinkedIn.
Cheers!
Hi,
Enabling **Large Semantic Model Storage Format** does not automatically improve report or query performance for every semantic model. Its main purpose is to support larger models, improve scalability, and optimize memory management for enterprise-scale datasets.
If your semantic model is relatively small and performs well, enabling this setting may provide little or no noticeable performance improvement. It is generally more beneficial for large Import models, high-capacity workloads, and scenarios that require features such as query scale-out or improved XMLA write performance.
Could you also share:
* The current size of your semantic model.
* Whether you're using Import, Direct Lake, or DirectQuery mode.
* The specific performance issue you're trying to improve (refresh time, query speed, or report loading).
This information will help determine whether enabling the large storage format is likely to benefit your scenario.