Limits and capacities

DataPrism bounds how much data a visual loads, how large a semantic model can grow, how far an AI turn may go and how big an export may get. This page gathers the numbers you can hit, where each one appears and whether it can be raised. Follow the links for the settings and the steps.

Visual ceilings

Each visual type has a largest value its limits may take. An author sets the value per visual under General > Limits, and the value saved with the visual is the one its picture exports, scheduled emails, the print sheet and version previews use, not a reader's own pick. A chart's CSV and XLSX downloads are not sized by it.

• The ceilings are fixed; an author picks any value from 1 up to the ceiling, and nothing above it is offered.
• A request above the ceiling is refused, for example: invalid chart configuration: limit must be at most 500 for a Pie visual (got 501).
• A rendered view that cannot fit a visual at its size names the lever: Lower the visual's limit in General > Limits to include it.
• The defaults, the limit rows and the reader's side are on Chart limits and Row limits and paging; what a tile shows when it is cut short is on Chart states and rendering rules.

Model capacities

A .semanticModel file and the model it imports into hold a bounded set of objects. An import, update from a file or copy that goes past a cap is refused with a sentence naming it.

• A file over the size cap is refused with The file is larger than 32 MB.; a file carrying too many objects gets a sentence such as the file has 301 BI tables; at most 300 can be imported.
• The caps are fixed; trim the file or the model and run the action again. The flows are on Model files and replacing a table.

Check and test caps

• Check selected takes up to 500 measures and 500 BI tables per check; past the cap the picker disables the rest of the checkboxes and reads At most 500 measures and 500 BI tables per check.
• Data checks cover the first 25 relationships and 25 measures of the scope; the rest are listed under Not checked (N) with the reason data checks cover the first 25 relationships and 25 measures of the scope; narrow the scope.
• A check collects at most 2,000 findings; past that the list shows More than 2000 findings; narrow the scope.
• A measure holds at most 50 tests, and a model at most 2,000.
• Reading a check and running tests is covered on Check, document and trace a model and Measure tests.

What an AI turn may change

• A single Model AI turn stops after 240 changed objects or 80 data calls and hands back, asking you to continue in a new message. One message carries at most 20 calls, and one write changes at most 12 objects.
• The bounds count one turn at a time; a new message continues the work from where it stopped. The panel itself is on Model AI.
• The Report Builder guards its jobs: a job that repeats the same error 5 times, or takes 20 steps with no change to the report, stops and says why instead of looping, and at most 2 custom visual steps run at the same time. See Report Builder AI.

SQL features and newer servers

• rank() with an argument: a model check flags the measure as a warning, because newer server versions refuse that form; the measure keeps rendering. The fix the check gives: Write rank() with no arguments and put the ranked expression in the window ORDER BY, for example rank() OVER (ORDER BY <expression> DESC).
• Pipelined SQL (|>) and FROM-first queries are refused in measures and in raw SQL. A measure's error reads measure validation failed: the expression is not allowed: pipelined SQL (|>) and FROM-first queries are not allowed in measures; raw SQL answers raw SQL is not allowed: pipelined SQL (|>) and FROM-first queries are not allowed in raw SQL; write SELECT ... FROM "<BI table>". A stored measure that uses the syntax keeps running and is flagged uses pipelined SQL (|>) or a FROM-first query until it is rewritten.
• A chart's join may build up to 5.5 GiB in memory. A chart whose join grows past that fails with a join limit error.
• Date-times are exact, with one recorded limitation: a date-time stored at the finest sub-second precision that falls after 8 October 2922 comes back about 2922 years early. Values in the ordinary range are unaffected.
• Which refusals you meet depends on the server version behind your platform. The wording of each message and what to do about it is collected on Troubleshooting; the rules for what a measure may contain are on Measures.

Export sizes

• The picture caps are fixed; a picture over its cap is retried at a lower resolution, and when it still does not fit the export reports it, for example Page 2 exceeds the page picture limit, A tile on page 2 exceeds the tile picture limit or Page 2 exceeds the remaining report size budget, and the export notes what was left out.
• An attachment over 20 MB fails the send with The attachment exceeds the 20MB limit.
• The chart export API: a direct download returns up to 1,000,000 rows and uses up to 600 seconds of query time; an async job covers up to 3,600 seconds and a 5 GB file, and the file is kept for 24 hours after the job finishes.
• Flows and permissions are on Export reports and data; email schedules are on Email subscriptions.

Questions, answered

Can I raise a visual's maximum?

No. Each type's maximum is fixed, and an author sets any value from 1 up to it. The maximums, the defaults and the value rules are on Chart limits.

Why was my .semanticModel file refused?

Most often the file is too large or holds too many objects. Over 32 MB it is refused with The file is larger than 32 MB.; over a count cap with a sentence such as the file has 301 BI tables; at most 300 can be imported. Trim the file, then run the import or update again; both flows are on Model files and replacing a table.

Why did the Model AI stop before it finished?

It reached the working bounds of one turn: at most 240 changed objects or 80 data calls, with at most 20 calls in a message and 12 objects in one write. It hands the work back and asks you to continue in a new message, and the new message picks up where it stopped. See Model AI.

Why does a chart fail with a join limit error?

Its join grew past the chart join limit of 5.5 GiB. Narrow the query with filters or fewer fields, then refresh; Troubleshooting has the message and what to try.

Why does a model check warn about rank()?

Newer server versions refuse rank() with an argument, so the check warns before it reaches a report. The fix it gives: Write rank() with no arguments and put the ranked expression in the window ORDER BY, for example rank() OVER (ORDER BY <expression> DESC). The measure keeps working until you rewrite it; see Check, document and trace a model.

Why are pipelined SQL and FROM-first queries refused?

Measures and raw SQL each accept one SELECT ... FROM form, so |> and FROM-first syntax are refused with a sentence naming the rule. A stored measure that already uses the syntax keeps running and is flagged until you rewrite it; see Measures.

Are very large date-times exact?

Yes, with one recorded exception: a date-time stored at the finest sub-second precision that falls after 8 October 2922 comes back about 2922 years early. Values in the ordinary range are exact.

Why does an export say a picture is over its limit?

The export pictures have caps: 4 MB per tile, 8 MB per page and 64 MB for the whole report. An over-cap picture is retried smaller; when it still does not fit, the export names it, for example Page 2 exceeds the page picture limit. See Export reports and data.