Difference between the weights of the semantic models in Power BI Desktop and the weights in Power B

[MiguelGaoCa]

The company has generated through Power BI Embedded a platform to be able to visualize the reports created in Power BI Service, but also gives the possibility to download reports AS LONG AS they weigh less than 300MB.

However, our model on the desktop weighs 282MB,

but when we upload the model to the service, and we validate it in the Configuration part of the work area and go to the System Storage option we see that the same model weighs 757MB.

We would like to know what is the difference of the weights and which are the characteristics that explain the weight in the service.

To give context, the model has several reports attached to it and has been published for a long time, in case it has something to do with saving previous versions.

Thank you very much

[v-pgoloju]

Hi @MiguelGaoCa,

Just following up to see if the solutions provided by community members were helpful in addressing the issue.

If one of the responses helped resolve your query, please consider marking it as the Accepted Solution. Feel free to reach out if you need any further clarification or assistance.

Best regards,
Prasanna Kumar

[v-pgoloju]

Hi @MiguelGaoCa,

Just following up to see if the solution provided was helpful in resolving your issue. Please feel free to let us know if you need any further assistance.

Best regards,

Prasanna Kumar

[v-pgoloju]

Hi @MiguelGaoCa,

Thank you for reaching out to the Microsoft Fabric Forum Community. Thanks to @GilbertQ  and @MasonMA for prompt and helpful responses.

Just checking in to see if the explanation and suggested steps helped clarify the difference in model size between Power BI Desktop and the Service. Please let me know if you need any further details or assistance in optimizing your model.

If any of the responses addressed your concern, please consider marking one as the Accepted Solution. Feel free to reach out if you need further assistance or clarification.

Best regards,
Prasanna Kumar

[GilbertQ]

Hi @MiguelGaoCa 

The reason this happens is because the PBIX version is the version where it is compressed and stored to disk. When this gets opened up, it expands into memory, which uses more memory, which is typically 3 to 4 times the size of the PBIX file size. And that is why you're seeing the larger file size that is consistent in how it works.

[MiguelGaoCa]

Hi @GilbertQ 
First of all, thank you very much for your response.
On the other hand, is it possible to know in some way how that weight that you see in the service is composed. In order to be able to minimize that weight or to be able to optimize it.

[MasonMA · Accepted solution]

@MiguelGaoCa 

The major reasons are, 

1. Columnar Storage Expansion. Power BI Service converts your model into a highly optimized VertiPaq columnar database. While this format is compressed for query performance, metadata (dictionaries, hierarchies, relationships) can inflate storage by 2–3× compared to the desktop file.
2. Data Partitioning. The service splits data into segments/partitions for incremental refresh and parallelism, adding management overhead.

As per this logic, you would have the expection that your semantic model can grow by 2-3 times after they are published in Power BI service. And regarding how to reduce the size of your model, you would need to optimize them in desktop tool with a couple of measures below: 

- Remove unused columns, hide unused tables.
- Use integer/date keys instead of strings.
- Avoid high-cardinality columns (e.g., free-text).
- Switch to DirectQuery for large tables if feasible.

these are very general and I'd recommend you refer to SQLBI for detailed guidance and best practices. 

https://www.sqlbi.com/articles/optimizing-high-cardinality-columns-in-vertipaq/