Fabric Capacity Calculator
Estimate which Fabric F SKU your Power BI workload needs. Enter your model sizes, refresh schedule and peak users to see compute and per-model memory pressure — and what to fix before paying for the next size up.
Runs entirely in your browser. Nothing you enter is uploaded or stored.
Capacity Configuration & Workload Parameters
Matches P1 Equivalent sizing.
Total semantic models deployed on this capacity.
In-memory VertiPaq footprint per semantic model.
Datasets refreshing simultaneously (e.g. at 8:00 AM).
Longer refreshes burn background CUs for hours.
Simultaneous consumers browsing interactive reports.
This workload does not fit Fabric F64 as configured. Estimated peak demand (442 CUs) is about 6.9× the 64 CUs in Fabric F64; sustained overage leads to delayed and then rejected report queries.
($100,908 / year) by optimizing models & schedules rather than jumping to next tier.
Morning Scheduled Refresh Collision
8 semantic models refresh simultaneously during peak hours, creating massive compute spikes.
Long-Running Unpartitioned Refreshes
Average refresh duration is 45 minutes, burning background CUs for prolonged periods.
Multi-Workload Compute Contention
3 competing workloads (Dataflows, Spark, Semantic Models) consume CUs from the same shared pool.
Validate Your Capacity Before Doubling Your Monthly Azure Bill
Most Fabric and Premium capacity throttling is caused by schedule collisions, VertiPaq high-cardinality bloat, or runaway DAX iterators. I profile your Microsoft Fabric Capacity Metrics app to resolve root causes and avoid unnecessary tier upgrades.
Privacy: the analysis runs in your browser and what you enter never leaves it. I only count that the tool was used — no inputs, results or personal data are recorded.
Get the next Engineering Note
Practical DAX, semantic model and Fabric notes. 1–2 a month, unsubscribe any time.
Understanding Fabric Compute & Memory Throttling
Interactive vs Background Smoothing
Fabric smooths interactive queries over 5 to 64 minutes and background operations such as refreshes over 24 hours. Once more than 10 minutes of future capacity is used up, report queries are delayed; after 60 minutes they are rejected.
Per-Model Memory Limits
Each SKU caps the memory a single semantic model can use, including refresh overhead (25 GB on F64). A full refresh needs roughly twice the model's size, so the largest model — not the average — decides whether refreshes succeed.
Pre-Upgrade Engineering Economics
Upgrading from F32 to F64 or F64 to F128 doubles monthly Azure costs ($50k - $100k/yr). Staggering refreshes and pruning column cardinality routinely eliminates throttling without upgrading.
Frequently Asked Questions About Fabric Capacity Sizing
How does Fabric Capacity Unit (CU) smoothing work?
Fabric lets operations burst above the capacity's size and then smooths their usage over time: interactive operations over a minimum of 5 and up to 64 minutes, background operations (refreshes, dataflows) over 24 hours. Throttling starts when more than 10 minutes of future capacity is already used: first 20-second delays on new report queries, then rejection of interactive work beyond 60 minutes, and rejection of everything beyond 24 hours.
What happens when a model exceeds the SKU's memory limit?
Each Fabric SKU limits how much memory one semantic model can use (for example 10 GB on F32 and 25 GB on F64), including refresh overhead. A refresh that needs more than the limit fails. Inactive models can also be unloaded from memory and reloaded on the next query, which makes the first report view after a quiet period slower.
Does Direct Lake remove capacity memory limits?
No. Direct Lake loads only the columns queries need and avoids long import refreshes, which often lowers memory and refresh load. But the per-model memory limit still applies, and each SKU has Direct Lake guardrails on rows and files per table. Exceeding them causes DirectQuery fallback (on SQL endpoints) or failed refreshes (on OneLake).
How do I get an expert view of my capacity?
The fixed-price audit works from model metadata and, for capacity questions, exports or screenshots from your Capacity Metrics app — no tenant access needed. It identifies the models and refreshes driving CU usage and what to fix before committing to a larger SKU. For hands-on migration or tuning, a fixed-scope sprint follows.
Experiencing Query Timeouts or Capacity Overages?
Before jumping from F32 to F64, let me inspect your workload traces. I pinpoint runaway DAX measures, stagger refresh collisions, and configure incremental refresh partitions to restore capacity headroom.