Which Fabric capacity do I need?

Do you have a Fabric capacity and trying to make sense of the ‘Microsoft Fabric Capacity Metrics’ report or are you wanting to switch over to a Fabric capacity but not sure which one you need? Let’s shed some ligt on that.

I already have a Fabric capacity and want to see my usage

In order to get data on the capacity usage, you need to install the Microsoft Fabric Capacity Metrics report.

1. Navigate to https://app.powerbi.com/groups/me/getapps/apps?experience=power-bi
2. After installing it a new workspace has appeared named ‘Microsoft Fabric Capacity Metrics’.
3. Open it and open the report, the first time you need to enter the Fabric Capacity ID. You can find the ID in the admin portal. (see image below).
4. Enter the capacity ID and sign in with an account that is added as a capacity admin.

How to read the capacity report

After opening the capacity report, you’ll be greeted by something like this:

Now before we can understand this report we need to know a couple of things.

A Microsoft Fabric Capacity runs from a F2 all the way to a F2048 SKU. The value after the F display the CU’s (capacity units). So an F2 capacity has 2 CU’s per second. Capacity units represent a pool of compute power. Compute power is needed to run queries, jobs, tasks, refreshes, report opens, exports, data exports, etc.

A capacity has a certain limit per month (the capacity report calculates everthing to two week periods, so that’s what we’ll be doing as well for clarity purposes. The table below shows the amount of CU’s available per 2 weeks for each capacity.

SKU	Capacity unit (CU)
F2	2.419.200
F4	4.838.400
F8	9.676.800
F16	19.353.600
F32	38.707.200
….	…

If you primarily use the Fabric capacity for Power BI purposes (and don’t have a lot of Fabric pipelines, datawarehouses, etc.) then the following thumb of rule applies;

About ±75 percent of your available CU’s can be used for report refreshes and about 25% should be reserved for report viewers. Viewing a report uses a very small amount of CU’s. Dataset refreshes is where a lot of CU’s are consumed. So keep that in mind.

In our screenshot you can see that we used ‘17.338’ CU’s in the last 14 days. If we were running on an F2 capacity that means that we used a total of 0,71% of our total available CU’s.

If you hover with your mouse on the CU bar in a row, a popup will appear with details on the CU usage.

This means that we uses a total of 102.883 CU’s over the last 14 days. Where 100.341 CU’s were used for dataset refreshes and about 2500 for report interactions (which show up as ‘query’).

I don’t have a capacity yet and trying to determine which one I need

Well, the answer to this question is.. it depends.. Microsoft has recently released a calculator to give you a broad idea of which capacity you’ll need: https://www.microsoft.com/en-us/microsoft-fabric/capacity-estimator

On the estimator page, enable the ‘Power BI embedded’ checkbox and at the bottom of the page enter a value for the amount of daily report viewers you have. Let’s say ‘120’;

As you can see, Microsoft estimates an F2 capacity! Of course it all depends on how heavy your reports are to refresh and use.

Bursting, smoothing, carry forward and overages

A capacity has a predefined amount of CU’s per second (as we just learned) but that doesn’t mean that you can stretch those CU’s temporarely.

Capacity Bursting:

• Fabric allows workloads to exceed their baseline capacity, enabling them to handle sudden increases in demand. 
• This is particularly useful for interactive tasks like loading visuals or background operations like model refreshes. 
• Burstable capacity is tied to the SKU of the Fabric capacity and the specific workload. 
• For example, a workload might use more compute units (CUs) than its baseline for a short period, completing a task faster. 

Capacity Smoothing: 

• Smoothing spreads the evaluation of compute resources over a specific time window to prevent immediate over-consumption.
• For interactive jobs, smoothing distributes the load over 5 minutes.
• For background operations, the smoothing window is 24 hours.
• Smoothing ensures that even if a workload uses bursting, the overall capacity usage remains within manageable limits.

Carry Forward: 

• If a workload exceeds its capacity, it can “borrow” resources from future time windows.
• This allows for efficient utilization of capacity and prevents immediate throttling.
• For example, if a semantic model refresh is performed at 9 am, the CU usage is spread over the next 24 hours, and the system can borrow capacity from that 24-hour window if needed.
• The concept of carry forward is closely related to smoothing and bursting, ensuring that overages are managed effectively.

Once all above fail, it will result in an overage where the refresh or report query is rejected by the capacity. You should try to prevent this because it a pretty negative user experience. The Fabric Capacity Metrics also have a tab that display throttling and overages table.

Optiziming CU usage

Optimizing your CU usage means that you can distribute your reports to more users on the same capacity and thus saving money. There are various ways to reduce the CU usage in reports;

Interactive

• Star schema; use a proper star schema with central fact tables and connected dimension tables. Avoid very complex snowflake schema’s or flat models, which can slow things down.
• Optimize DAX: Poorly written DAX expressions can significantly increase CPU and memory usage.
• Reduce number of elements/visuals: Every visual sends at least one query to the dataset. More visuals = more queries = more CU usage.
• Use aggregations: Aggregations pre-summarize data at a higher level (e.g., monthly totals instead of daily).

Background

• Reduce manual refreshes: refreshes are probably the most CU heavy actions, try to keep the refreshes to a minimum
• Incremental Refresh: Instead of refreshing the full dataset each time, only new or changed data is refreshed.

Did you know that with a Capacity, you can share Power BI reports with anyone in the world without needing any Power BI Pro licenses using DataTako?