10GB Semantic Model limit on F64 for incremental refresh

[rvp_ordix]

We are currently trying to make a Semantic Model usable in a MS Fabric Workspace.

• The WS is configured with an F64 capacity
• The Semantic Model is in Import Mode
  • It was uploaded while empty, which worked
  • The initial refresh happened on Premium-Per-User capacity and worked there
• The Model is set to Large Semantic Model Storage Format
• There is no artifical model limit configured in our Capacity Settings
• The dataset is ~16GB in size
• The dataset has partitions configured: Yearly for all prior years; Monthly for the current year

A full refresh of that model doesnt work with the memory limit on an F64 (we'd need >32GB memory presumably), but we figured that an incremental refresh (started via XMLA endpoint, which is enabled in the capacity settings) for only the current month would do the trick. For some reason it does not. Stranger still is the error message that appeared some 40min into running the refresh:

Resource Governing: This operation was canceled because there wasn't enough memory to finish running it. Either reduce the memory footprint of your dataset by doing things such as limiting the amount of imported data, or if using Power BI Premium, increase the memory of the Premium capacity where this dataset is hosted. More details: consumed memory 10221 MB, memory limit 10204 MB, database size before command execution 15395 MB. See https://go.microsoft.com/fwlink/?linkid=2159753 to learn more.

Why would there be a limit of 10GB on an F64? Where does this come from?

The only entry we found even mentioning such a limit was this question stating that individual partions may not be larger than 10GB, but that shouldn't apply for us since even the largest table is only 3GB in total according to Dax Studio (there was also no official documentation supporting that, but regardless).

We are somewhat stumped.

[rvp_ordix · Accepted solution]

The answer was simpler than expected, though no less frustrating to pin down for it. Our memory limit was at ~10GB because the DB took up ~15GB already. If we add up 10204 MB and 15395 MB we get 24,999GB, the limit of our F64.

It was merely unfortunate that the two values looked like somewhat round numbers on their own.

[v-kpoloju-msft]

Hi @rvp_ordix,

Thanks for reaching out to the Microsoft fabric community forum.

After thoroughly reviewing the details you provided, here are few alternative workarounds that might help resolve the issue. Please follow the steps below:
Thank you, @lbendlin for your valuable input regarding the issue.

• F64 capacity has 64GB total memory shared across workloads, with a max of 25GB per dataset. Power BI enforces per-operation memory limits during refreshes, and the error indicates query execution exceeded available memory (10,221 MB > 10,204 MB). Memory spikes can cause refresh failures even if the dataset fits in capacity.
• Power BI has a 10GB limit per partition during processing. Ensure your partitions, especially the current months, are under this limit after compression. Check partition sizes using DAX Studio or Fabric Capacity Metrics. Test a smaller timeframe refresh to identify excessive memory usage.
• Ensure XMLA is performing a true incremental refresh, processing only new data and leaving existing partitions unchanged. Verify the partitioning strategy in Power BI Service for correct incremental refresh settings.
• Reduce dataset size by removing unused columns and high-cardinality fields. Avoid complex calculated columns that increase memory usage. Use Fabric Capacity Metrics to monitor memory usage during refresh operations.
• If incremental refresh continues to fail despite optimizations, you may want to consider upgrading to F128, as it offers more memory for handling large refresh operations.
  
  

Please go through the below following links for more information:
What is Power BI Premium? - Power BI | Microsoft Learn
Capacity and SKUs in Power BI embedded analytics - Power BI | Microsoft Learn
How to configure workloads in Power BI Premium - Power BI | Microsoft Learn

If this post helps, then please give us ‘Kudos’ and consider Accept it as a solution to help the other members find it more quickly.

Best Regards.

[v-kpoloju-msft]

Hi @rvp_ordix,

May I ask if you have resolved this issue? If so, please mark the helpful reply and accept it as the solution. This will be helpful for other community members who have similar problems to solve it faster.

Thank you.

[v-kpoloju-msft]

Hi @rvp_ordix,

I wanted to check if you had the opportunity to review the information provided. Please feel free to contact us if you have any further questions. If my response has addressed your query, please accept it as a solution and give a 'Kudos' so other members can easily find it.

Thank you.

[v-kpoloju-msft]

Hi @rvp_ordix,

I hope this information is helpful. Please let me know if you have any further questions or if you'd like to discuss this further. If this answers your question, please Accept it as a solution and give it a 'Kudos' so others can find it easily.

Thank you.

[lbendlin]

There is a hard limit of 10GB (on any capacity size), but for a partition, not a semantic model.  Use partition bootstrapping and fill the partitions manually via XMLA.