pbPassingBI
/
Data sources & datasets intermediate 6 min

SPICE capacity and refresh

Managing the in-memory engine, refresh schedules and incremental refresh.

What you'll be able to do
  • Manage SPICE capacity
  • Schedule refreshes
  • Use incremental refresh appropriately

What SPICE is

SPICE is QuickSight's in-memory columnar engine. A dataset imported into SPICE is queried from memory rather than from the source, which is much faster and takes load off the source system.

The alternative is direct query, which passes every request through to the source. Direct query stays current but is only as fast as the underlying database, and every reader interaction hits it.

Capacity

SPICE capacity is purchased per account and shown in Manage QuickSight → SPICE capacity.

Running out means refreshes fail — often quietly, leaving a dashboard showing stale data with no obvious warning. Worth monitoring rather than discovering.

The ways to use less: exclude unneeded fields, filter rows at dataset level, and aggregate before import rather than loading raw detail.

Refresh schedules

Set a schedule per dataset — hourly, daily, weekly, or on a custom cadence.

Match it to how often the data actually changes. An hourly refresh of a daily-loaded table is wasted capacity and wasted queries against the source.

Incremental refresh

For large datasets, incremental refresh processes only rows within a recent window based on a date column, rather than rebuilding everything.

Much faster, but it will not capture rows updated or deleted outside that window. The usual pattern is incremental frequently with a periodic full refresh.

It requires a date field suitable as the lookback boundary, and a SQL-based source.

SPICE or direct query

NeedChoose
Fast dashboards, many readersSPICE
Data must be current to the minuteDirect query
Source system under loadSPICE
Dataset larger than your capacityDirect query, or aggregate first
Key points
  • Running out of SPICE capacity makes refreshes fail, often quietly
  • Match refresh frequency to how often the data actually changes
  • Incremental refresh misses rows updated outside the lookback window
Check yourself