AI functionality using Power BI Dataset

[visheshjain]

Hello everyone,

I am trying to introduce AI funcationality to our users, which will be based on the datasets in PBI service.

I got a F2 capacity and enabled copilot which worked a little and then the stopped since the compute units were consumed.

Once the units were consumed, the reports in that workspace would not load since the compute units were consumed.

I talked to the PBI support team and the Azure team as well and they said that the capacity is thorttling and that it will take time to normalize again and if I turn the capacity on and off, it will work but I will have to pay for it etc. etc.

I understand that the F2 capacity is small but I do not mind if my prompts take longer time to return an answer, cause even if I increase the capacity size, eventually it is going to hit that threshold of consuming compute units.

Is there any way that I can enable use of AI capabilities based on the datasets in PBI Service, in a safe and secure way with RLS enabled, without hitting any thresholds and without providing user access to the desktop files?

Again, I do not mind if the prompts take a little longer for returning results, due to a small capacity.

Thank you,

Vishesh Jain

[v-aatheeque]

Hi @visheshjain 
Following up to confirm if the earlier responses addressed your query. If not, please share your questions and we’ll assist further.

[visheshjain]

Hi,

Yes I went through the solutions provided and none of them really work.

I turned on surge protection but in that case copilot usage is very limited and if I use the entire capacity as a Copilot capacity then the report refreshes stop completeley.

I spoke to Microsoft over multiple tickets and they said that I have depleted my 'compute units' which even they cannot tell that how long will it take to normalize. I did deploy the Fabric Metrics Capacity app but still no answer to this questions.

Upgrading it to a F4 capacity will increase the number of 'compute units' but it will also have the same limitations with higher thresholds, not to mention the cost associated with the F4 capacity will be double.

Is there some way to use an MCP server with Power BI service since users will not have access to the desktop files.

Or we can create a custom visual, which will be connected to a LLM in the backend...I'm just spitballing here.

Thank you,

Vishesh Jain

[lbendlin]

In my experience you need at least a F128 for a meaningful Copilot experience. Even a F64 can be brought to its knees by a handful of concurrent Copilot users. Ask me how I know.

CU smoothing happens over a 24 hour period, mostly. 

[visheshjain]

Hi @lbendlin,

F128 is going to be super expensive (not to mention Microsoft's billing isn't as transparent...my F2 capacity has different daily cost, even when in paused state) and as you correctly pointed out if a F64 capacity can be brought to its knees, then even F128 might fall short at some point of time. Eventually, it is going to spoil our user's experiences, cause Copilot goes through CU units in a snap.

I am trying to avoid going the Copilot route and probably figure something out using Model Context Protocols (MCP) instead.

P.S. - CU smoothening happens MAY happen over 24 hour period, but I might have consumed some 2000% of my daily units or something or thats what the Microsoft advisor told me. LOL!

[v-aatheeque]

Hi @visheshjain 

Your concern is valid. Copilot workloads can spike CU consumption quickly, especially on smaller SKUs like F2. While CU smoothing averages usage over 24 hours, throttling can still occur during short-term bursts if the instantaneous consumption exceeds the allocated capacity.

[v-aatheeque]

Hi @visheshjain 

Have you had a chance to look through the responses shared earlier? If anything is still unclear, we’ll be happy to provide additional support.

[v-aatheeque]

Hi @visheshjain 

We wanted to follow up to check if you’ve had an opportunity to review the previous responses. If you require further assistance, please don’t hesitate to let us know.

[lbendlin]

1. Set up Surge Protection for that capacity.  Setting it somewhere to 40% will leave enough room for interactive queries while blocking background (which includes CoPilot)

2 Get a (much) bigger capacity.  F2 is too small for CoPilot.

[ibarrau]

Hi. F2 is really small to use both the interactivity of reports and backgroup data operations + copilot. F2 will help you to understand how it's used, but it won't work with a productive environment using both. As an alternative you could purchase another F2. Keeping two F2. You can change at admin portal settings that you want one F2 to be exclusively for Copilot. That way it will never use the reports capacity. It will take time to answer the AI but it won't make trouble with reports. The setting is:

You can find more info here: https://learn.microsoft.com/en-us/fabric/enterprise/fabric-copilot-capacity

I hope that helps,