Configure settings for scale

The model is designed. Now configure the settings that control how it handles large datasets, concurrent queries, and external tool access. These settings determine whether the model can keep up as data volumes grow and more users and tools consume it.

Large semantic model storage format

The large semantic model storage format changes how the model stores and compresses data. By default, Power BI limits model uploads to 10 GB. With this setting enabled, models can grow beyond that limit on refresh. The maximum size equals the Fabric capacity size or the limit set by the capacity administrator.

Direct Lake models automatically enable this setting, so you don't need to configure it manually for those models. For Import mode models, you need to enable it explicitly.

This setting is also a prerequisite for both XMLA endpoint read/write access and query scaleout. You need to enable it first when using Import or DirectQuery storage modes.

Enable large semantic model storage format when:

• Your data volumes require models that grow beyond the 10-GB upload limit.
• You need XMLA endpoint access for external tools.
• You plan to use query scaleout for high concurrency.
• You plan to use incremental refresh with partitioned tables.

XMLA endpoint read/write access

The XMLA endpoint lets external tools connect to your semantic model. Tools like Tabular Editor, DAX Studio, and ALM Toolkit use this endpoint for development, debugging, and deployment operations that aren't available in the Fabric service interface.

XMLA endpoint read/write access requires the large semantic model storage format as a prerequisite. Once both are enabled, you can:

• Use Tabular Editor for model development and source control integration.
• Use DAX Studio for query analysis and performance tuning.
• Deploy models through CI/CD pipelines using the Analysis Services client libraries.

At scale, these external tools become essential. Manual edits through the service interface don't support the level of model development that large, team-maintained models require.

Query scaleout

Query scaleout distributes read queries across read-only replicas of your semantic model. When hundreds of users access the same model simultaneously, a single instance can become a bottleneck. Query scaleout adds replicas that share the query load.

When you enable query scaleout, the read replicas use a separate copy of the model. This copy synchronizes after each refresh. There can be a brief delay between the primary model finishing a refresh and the replicas reflecting the updated data.

Query scaleout requires large semantic model storage format as a prerequisite.

Enable query scaleout when:

• The model serves hundreds of concurrent users.
• Query performance degrades during peak usage periods.
• The model is backed by a Fabric capacity that supports replicas.

Direct Lake fallback configuration

Direct Lake reads Delta tables directly from OneLake into memory. Some queries can cause the model to fall back to DirectQuery mode, which changes performance characteristics. The fallback setting controls how the model handles these situations:

• Allow fallback (default): Queries that can't run in Direct Lake mode fall back to DirectQuery automatically. Users get results, but performance might decrease.
• Disallow fallback: Queries that can't run in Direct Lake mode return an error. This enforces consistent performance but requires that all queries stay within Direct Lake capabilities.

For models at scale, start with fallback allowed. Monitor which queries trigger it, then optimize those queries or data structures to reduce fallback frequency. Disallow fallback only when all query patterns stay within Direct Lake limits and you need guaranteed performance consistency.

OneLake integration

OneLake integration makes your semantic model data accessible as Delta tables in OneLake. When enabled, downstream Fabric items like notebooks, pipelines, and other services can read data directly from the semantic model without rebuilding it from source.

This extends the model's reach beyond reports. A semantic model with well-structured star schema and calculation logic becomes a curated data source for the broader analytics platform.

Enable OneLake integration when:

• Data engineers or data scientists need to consume semantic model data in notebooks or other Fabric items.
• You want to use the semantic model as a shared data source across Fabric.
• Downstream consumers need access to curated, business-logic-enriched data without rebuilding it from raw sources.

Settings decision framework

The following table summarizes the key settings decisions for scale: