--- sidebar_position: 3 title: Duckling Sizes description: Learn about MotherDuck Duckling (compute instance) sizes and their optimal use cases. --- MotherDuck implements a distinct tenancy architecture that diverges from traditional database systems. The platform utilizes a [hypertenancy](/concepts/hypertenancy) model, which provisions isolated read-write Ducklings (compute instances) for each Organization member. This architecture ensures dedicated compute resources and Duckling-level configuration at the individual user level, allowing users to independently optimize performance parameters according to their specific workload requirements. Each Duckling size has different performance characteristics and [billing implications](/about-motherduck/billing/pricing/#compute-pricing). MotherDuck uses fast SSDs for spill space, so queries can exceed their memory limits with minimal performance impact. DuckDB caches data in memory, and MotherDuck uses fast local disks for storage, which improves cold start times. ## Duckling sizes | Duckling Size | Plans | Use Case | Default Cooldown | Configurable Cooldown Period | Startup Time | Read-Write Duckling Enabled? | Read Scaling Duckling Enabled? | |---------------|------------|----------|------------------|------------------------------|--------------------|-----------------------------|-------------------------------| | Pulse | Lite, Business | Good for small workloads| 1 second | N/A | ~100ms | Yes | Yes | | Standard | Business | Good for most data loading workloads | 1 minute | 1 min – 24 hours | ~100ms | Yes | Yes | | Jumbo | Business | Better for large, complex transformations during loading | 1 minute | 1 min – 24 hours | ~100ms | Yes | Yes | | Mega | Business | Optimal for demanding jobs with even larger scale and volumes than a Jumbo can handle | 5 minutes | 1 min – 24 hours | ~a few minutes | Yes | Yes | | Giga | Business, and in [Free Trial on request](https://motherduck.com/contact-us/product-expert/) | Best used for your largest and toughest workloads like batch jobs that run overnight or on weekends | 10 minutes | 1 min – 24 hours | ~a few minutes | Yes | No | - The cooldown period is [configurable](#configuring-the-cooldown-period) for Standard, Jumbo, Mega, and Giga Ducklings - We recommend keeping the cooldown periods in mind when planning batch sizes - To shut down a Duckling without waiting for cooldown, use [`SHUTDOWN` or `SHUTDOWN TERMINATE`](/sql-reference/motherduck-sql-reference/shutdown-terminate/) ### PULSE **Optimized for ad-hoc analytics and read-only workloads** Pulse Ducklings are auto-scaling and designed for efficiency, making them ideal for: - Running ad-hoc queries (**Note** complex queries involving [spatial analysis](https://duckdb.org/docs/stable/extensions/spatial/functions.html) or regex-like functions may perform better on larger Duckling sizes) - Read-optimized workflows with high concurrent user access, such as those in customer-facing analytics. - Powering data apps and embedded analytics where quick, short queries are common. - High-concurrency, read-optimized workflows [Learn how Pulse Ducklings are billed.](/about-motherduck/billing/pricing/#compute-pricing) ### STANDARD **Production-grade Duckling designed for analytical processing and reporting** Standard Ducklings offer a balance of resources for consistent performance, suited for: - Core analytical workflows requiring balanced performance metrics. - Development and validation environments for production workflows. - Standard ETL/ELT pipeline implementation, including: - Parallel execution of incremental ingestion jobs. - Multi-threaded transformation processing. [Learn how Standard Ducklings are billed.](/about-motherduck/billing/pricing/#compute-pricing) ### JUMBO **A larger Duckling built for high-throughput processing and faster performance** Jumbo Ducklings provide resources for heavy workloads, including: - Large-scale batch processing and ingestion operations. - Complex query execution on high-volume datasets. - Advanced join operations and aggregations. - RAM-intensive processing of deeply-nested JSON structures or other large data objects. [Learn how Jumbo Ducklings are billed.](/about-motherduck/billing/pricing/#compute-pricing) ### MEGA **Built for high-throughput processing on demanding jobs at even larger scale than a Jumbo's capacity** Mega Ducklings provide compute resources to help expedite large-scale transformations and complex operations, perfect for: - Batch processing and high-volume ingestion operations. - Running a weekly job that rebuilds all of your tables that needs to run quickly, in minutes - not hours. - Complex query execution on high-volume datasets that a Jumbo Duckling won't be able to handle in a time crunch. - Advanced operations for users with 10x the data volume as other users who require low-latency, swift performance. [Learn how Mega Ducklings are billed.](/about-motherduck/billing/pricing/#compute-pricing) ### GIGA **Our largest Duckling, built for the toughest workloads with massive scale and complexity** Giga Ducklings provide compute resources for the most demanding tasks, perfect for: - Complex, large-scale workloads and jobs that won't run on any other Duckling size. - Running one-time jobs that need to complete overnight or over the weekend, like restating revenue actuals for 10 years's worth of high-volume data. - Huge volumes of advanced join operations and aggregations. - Very large amounts of RAM-intensive processing of deeply-nested JSON structures or other large data objects. [Learn how Giga Ducklings are billed.](/about-motherduck/billing/pricing/#compute-pricing) ## Configuring the cooldown period The **cooldown period** is the duration an idle Duckling stays running after the last query completes. During cooldown, the Duckling remains warm — cached data stays in memory, so follow-up queries start faster. You are billed for the cooldown period, since the Duckling is still running. You can configure the cooldown period per user or service account through the MotherDuck UI (under **Settings > Ducklings**) or through the [`Set user Ducklings` REST API](/sql-reference/rest-api/ducklings-set-duckling-config-for-user/). ### Configurable cooldown period by Duckling type | Duckling type | Default Cooldown | Configurable Cooldown Period | |---------------|-----------------|------------------------------| | Pulse | 1 second | N/A | | Standard | 1 min | 1 min – 24 hours | | Jumbo | 1 min | 1 min – 24 hours | | Mega | 5 min | 1 min – 24 hours | | Giga | 10 min | 1 min – 24 hours | Pulse Ducklings are meant for 'bursty' workloads - as a result, they are on-demand and auto-scaling. Because they are metered on a per-query basis, with a minimum of 1 Compute Unit (CU)\* second, they do not have a configurable cooldown. \***We define and measure the amount of CPU and memory usage over time as a Compute Unit (CU).** ### When to adjust the cooldown period **Shorter cooldown** — reduces idle billing when queries are infrequent or spread out over long intervals. Good for batch jobs or scheduled pipelines where you know the Duckling won't be needed again immediately. **Longer cooldown** — keeps the Duckling warm between queries, avoiding cold-start latency. Good for interactive analytics sessions, dashboards with periodic refreshes, or workloads where cache hits improve performance. ### Example: reducing costs for a nightly batch job A Giga Duckling has a default cooldown of 10 minutes. If you run a batch job that takes 5 minutes and know there's no follow-up query, the Duckling stays idle (and billable) for 10 minutes after the job completes. By reducing the cooldown to 5 minutes, you save 5 minutes of idle billing per run. For a daily job, that's over 30 hours of saved compute per year. To eliminate idle billing entirely, use [`SHUTDOWN`](/sql-reference/motherduck-sql-reference/shutdown-terminate/) at the end of your job to shut down the Duckling gracefully, or [`SHUTDOWN TERMINATE`](/sql-reference/motherduck-sql-reference/shutdown-terminate/) to force-terminate it immediately. Note that you will always be billed for the minimum cooldown time of 1 minute. ::::info MotherDuck meters compute per-second and bills for a 1-minute minimum. While Standard, Jumbo, Mega, and Giga Ducklings are billed for *wall clock time*, Pulse Ducklings are metered on a per-query basis to support 'bursty' workloads. As a result, they are on-demand and auto-scaling. Because Pulse Ducklings are metered on a *per-query basis, with a minimum of 1 Compute Unit (CU)\* second*, they do not have a configurable cooldown. \***Compute Unit (CU): The amount of CPU and memory usage over time.** :::: ### Important notes - Cooldown is **best effort** — Ducklings may be shut down before the configured cooldown expires due to lifetime limits, background operations, and maintenance upgrades. - You are only billed for the time a Duckling is actually running. If a Duckling shuts down early, billing stops at that point. - The UI validates the min/max bounds and shows an error if the configured value is out of range. ## Changing Duckling sizes Duckling sizes can be changed in MotherDuck UI by clicking on the icon in the top right, or under "Settings > Ducklings". Here you can choose the desired Read/Write and Read Scaling size. Changing Duckling size can take up to a few minutes while your new Duckling wakes up. ![Duckling Selector](img/duckling_selector.png) The Duckling size for a user or service account can also be set using the [`Set user Ducklings` REST API](/sql-reference/rest-api/ducklings-set-duckling-config-for-user/). **Note:** Changing Duckling size in the UI or through our [REST API](/sql-reference/rest-api/motherduck-rest-api/) takes * **2 minutes** for Pulse, Standard and Jumbo * **5 minutes** for Mega * **10 minutes** for Giga