Skip to main content
cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 

Join us at FabCon Vienna from September 15-18, 2025, for the ultimate Fabric, Power BI, SQL, and AI community-led learning event. Save €200 with code FABCOMM. Get registered

VivienneVereen
Microsoft Employee
‎02-24-2025 09:00 AM

Fabric Capacity Scaling and Power BI - What happens when Power BI model guardrails are exceeded?

Introduction

 

In Microsoft Fabric, capacities provide the compute power for a comprehensive suite of analytics tools. A single capacity supports all Fabric experiences, including Power BI. Microsoft Fabric offers a robust set of analytics experiences designed to work together seamlessly - it is integrated, secured, and governed. Additionally, there is flexibility with pay-as-you-go or reservation purchase options, and you can get started with a 60-day free trial (Getting Started | Microsoft Fabric).

 

VivienneVereen_0-1739229542611.png

 

Fabric Capacity Scaling and Features

 

There are many features that make capacities flexible for performance and cost management, such as bursting and smoothing:

  • Bursting: Allows Fabric to run operations at top speeds.
  • Smoothing: Adjusts the accounting of spikes to spread over longer periods by using unneeded capacity from other parts of the day. Interactive operations are smoothed over a minimum of 5 minutes or longer, while background operations are smoothed over 24 hours.

Even with these features providing scalability and optimizing performance, there may still be a need to scale a Fabric capacity to avoid throttling.

 

Planning Your Fabric Capacity

 

When planning your Fabric capacity strategy, it is important to understand how consumption is calculated.

There are out-of-the-box monitoring tools available such as the Fabric Capacity Metrics App, Surge Protection (Preview), and Workspace Monitoring (Preview).

One great feature of Fabric SKUs is the ability to seamlessly scale and pause/resume within the Azure Portal. However, it is important to note that there are specific SKU limits and guardrails for Power BI workloads that need to be considered when planning a capacity strategy.

 

Additional Resources

 

My colleague Pat Mahoney has an excellent YouTube playlist about all things Fabric Capacities, including topics on Capacity Metrics App, automating pause/resume, dynamic scaling, and multi-capacity strategies to protect your workloads. I highly recommend checking these out!

 

Power BI Testing and Findings

 

Import Model

 

Testing for import models was performed on F16 and F8 SKUs.

 

Model Information

 

Checked Total Size in Memory with Vertipaq Analyzer in DAX Studio and confirmed it to be 3.04 GB.

 

VivienneVereen_1-1739229565429.png

 

F16 SKU

 

VivienneVereen_2-1739229659904.png

 

F16 SKU Guardrails:

 

The Max Memory limit for F16 SKU is 5 GB.

 

VivienneVereen_3-1739229669156.png

 

The report works:

 

VivienneVereen_4-1739229679446.png

 

F8 Scale Down

 

VivienneVereen_5-1739229689928.png

 

F8 SKU Guardrails:

 

The Max Memory limit for F16 SKU is 3 GB.

 

VivienneVereen_6-1739229696752.png

 

Initially, the report still works:

 

VivienneVereen_7-1739229711312.png

 

However, after a few minutes the report fails to load with the “exceeds the maximum size limit on disk” error:

 

VivienneVereen_8-1739229720222.png

 

Attempting to refresh the model, the refresh fails with the same error:

 

VivienneVereen_9-1739229729688.png

 

F16 Scale Up

 

VivienneVereen_10-1739229738912.png

 

After scaling up to F16 – the report works again, no model refresh required:

 

VivienneVereen_11-1739229745287.png

 

Direct Lake Model

 

Testing for Direct Lake models was performed on F64 and F32 SKUs.

 

Model Information

 

It was confirmed via Lakehouse SQL Endpoint query that the largest table in the model has a row count of 301,337,872:

 

VivienneVereen_12-1739229755924.png

 

Fallback is disabled – model will not fall back to DirectQuery when guardrails are exceeded:

 

VivienneVereen_13-1739229769736.png

 

Automatic Reframing is disabled:

 

VivienneVereen_14-1739229773649.png

 

F64 SKU

 

VivienneVereen_15-1739229780473.png

 

F64 SKU Guardrails:

 

The Direct Lake rows per table limit is 1.5 billion.

 

VivienneVereen_16-1739229791527.png

 

The report works:

 

VivienneVereen_17-1739229801525.png

 

F32 Scale Down

 

VivienneVereen_18-1739229806267.png

 

F32 SKU Guardrails:

 

The Direct Lake rows per table limit is 300 million.

 

VivienneVereen_19-1739229815206.png

 

Report fails to load - if storage Direct Lake Behavior setting had been set to automatic, the report would have loaded but fallen back to DirectQuery:

 

VivienneVereen_20-1739229822961.png

 

F64 Scale Up

 

VivienneVereen_21-1739229829583.png

 

Note, some limitations regarding licensing:

 

VivienneVereen_22-1739229839745.png

 

After scaling up to F64 – the report works again, no reframing required:

 

VivienneVereen_23-1739229845732.png

 

Summary

 

For Fabric and Power BI Premium, each SKU has specific limits and other constraints specific to Power BI workloads. When scaling a Fabric capacity, it is important to keep these guardrails in mind and ensure they are not exceeded so that Power BI reports remain accessible to business users. More details can be found here.

 

The behavior of Import and Direct Lake semantic models and reports during capacity scaling was tested:

 

  • For Import models, if the max memory limitation is exceeded after scaling down, the report will fail to load, and the model cannot be refreshed. However, immediately after scaling up, the report will work again without requiring an additional model refresh.
  • For Direct Lake models, if the direct lake rows per table limitation is exceeded after scaling down (given fallback is disabled), the report will fail to load. If fallback is set to automatic, the report will load but fall back to DirectQuery mode. Upon scaling up, the report will work again without needing reframing.

 

References

 

Fabric Capacities YouTube playlist:

Capacities:

Surge Protection (Preview):

Workspace Monitoring (Preview):

Direct Lake:

Power BI SKU Guardrails:

DAX Studio:

 

Labels: