LLMs to turn Ad Hoc Questions into Real-Time Answers

Watch this webinar with MotherDuck and Artefact for a discussion about how to achieve exceptional performance for gold layer analytics using DuckDB .

Mehdi joins us as a Data Engineering and Analytics Advocate after a career as a Data Engineer and a nighttime hobby of creating YouTube videos and other content for data professionals.

Mehdi Ouazza

Maël Deschamps is Director of Data Engineering at Artefact Northern Europe. He leads a team of data engineers, AI engineers, and analytics engineers, and is a speaker at conferences including PyData Amsterdam.

Maël Deschamps

large_Boosting Data Performance.png

medium_Boosting Data Performance.png

small_Boosting Data Performance.png

thumbnail_Boosting Data Performance.png

Boosting Data Performance.png

Boosting Data Performance: Unlocking the Power of DuckDB in your Gold Layer

Data teams have become remarkably efficient at building modern lakehouse architectures. Data flows through bronze and silver layers, where it is cleansed, transformed, and prepared for analysis. Yet, a common bottleneck persists at the final mile: the gold, or serving, layer. This is where data powers business intelligence (BI) dashboards and analytical applications. Traditional cloud data warehouses can be slow for interactive analytics, costly for exploratory queries, and introduce significant developer friction, undermining the agility of the entire stack. A recent benchmark by the data and AI consultancy, Artefact, [explores how MotherDuck tackles this "final mile" problem](https://www.artefact.com/blog/motherduck-explained-how-it-fits-into-your-data-stack-with-benchmarks/), proving it is a production-ready solution for the serving layer.

## Finding the Sweet Spot: Why MotherDuck Excels in the Gold Layer

For many organizations, the concept of a serving layer is the sweet spot for MotherDuck. According to Maël, a Director of Data Engineering at Artefact, this is where the platform’s strengths align perfectly with typical data volumes. While bronze and silver layers can contain massive, multi-terabyte datasets, the gold layer often consists of aggregated, curated datasets measured in gigabytes.

This is precisely the scale where the DuckDB engine excels. Unlike distributed engines designed for petabyte-scale processing, which often incur significant overhead from cold starts and network communication, DuckDB's architecture is optimized for analytical performance on single-node, "scale-up" infrastructure. This focus makes it exceptionally fast for the interactive query patterns common in BI and data applications. Artefact’s evaluation positions MotherDuck as an ideal component to either replace an existing gold layer or sit between it and the BI tool to accelerate performance and reduce costs.

## The Architecture Behind Sub-Second Analytics

This exceptional performance at the gigabyte scale is not accidental; it stems from a fundamentally different architectural approach. The core of this is MotherDuck's hybrid, or dual execution, model. Powered by DuckDB, an in-process analytical database, MotherDuck can process data both locally on a client machine and in the cloud. This flexibility is a key differentiator from traditional cloud data warehouses that operate exclusively in the cloud.

A powerful demonstration of this is through WebAssembly (WASM), which allows DuckDB to run directly inside a web browser. In a custom data application, an initial query can be sent to MotherDuck to load an aggregated dataset into the browser. From that point on, all subsequent filtering, slicing, and dicing operations trigger SQL queries that execute instantly within the browser itself. There is no network latency, no round trip to the cloud, and no waiting. The result is a truly interactive experience that feels more like a local application than a web-based dashboard.

This architecture contrasts sharply with the scale-out model of traditional warehouses. By leveraging serverless "ducklings" (dedicated, single-node compute instances), MotherDuck avoids the overhead of distributed computing for workloads that do not require it. This scale-up approach eliminates cold starts and provides consistent, low-latency performance for the rapid-fire queries typical of a serving layer.

## Putting Performance to the Test: A TPC-H Benchmark

Theoretical advantages only matter if they translate to real-world performance. To validate this, the team at Artefact conducted a TPC-H benchmark, a widely recognized industry standard for evaluating the performance of analytical query engines. They compared MotherDuck against Google BigQuery, a well-established leader, and Postgres, a common operational database often pressed into analytical service.

### Analyzing the Results

The benchmark delivered definitive results. Across the 22 TPC-H queries, executed from both the MotherDuck UI and an open-source BI tool (Apache Superset), MotherDuck was consistently the fastest engine.

On average, **MotherDuck was 3 to 4 times faster than BigQuery**. As the Artefact team noted, BigQuery has long been considered the "Olympics champion" of data warehousing. Seeing a new platform decisively outperform it was a powerful validation of MotherDuck’s architectural efficiency for this class of workload.

## Beyond Speed: The ROI of Developer Experience and Cost

While raw speed is impressive, the true value of a tool is measured by its overall impact. Two often-overlooked factors are developer experience and cost, and here again, MotherDuck presented a compelling case.

The biggest, though often unseen, return on investment comes from an improved developer experience. A tool that enables developers to move faster delivers more value to the business. The team at Artefact noted that setting up the TPC-H benchmark in MotherDuck was remarkably simple. The video demonstration shows that generating the entire 10-gigabyte test dataset and all 22 associated queries required running a single built-in function directly within the MotherDuck user interface. The entire process completed in seconds, a stark contrast to the manual data downloading, uploading, and schema creation often required for other platforms. This local-first workflow, where an engineer can start with a simple `pip install duckdb` and seamlessly transition to cloud execution with a one-line change, accelerates development cycles and simplifies debugging.

On the cost side, the results were just as striking. For the interactive workload simulated by the benchmark, **MotherDuck was approximately 20 times less expensive than BigQuery**. This is largely due to the different pricing models. MotherDuck's session-based pricing is highly advantageous for the bursty, interactive nature of BI, where a user may run dozens of queries in a short period. In contrast, BigQuery's scan-based pricing can become expensive for exploratory workflows that repeatedly query large tables.

## Future-Proofing the Serving Layer

As MotherDuck matures, the community is looking toward features that will further solidify its place in the enterprise, such as more granular permissions like Row-Level Security (RLS). This is part of a natural evolution for any data platform and is on the product roadmap.

Beyond specific features, a fast, open, SQL-based serving layer like MotherDuck helps address a broader industry challenge: the semantic layer. Too often, critical business logic becomes trapped in proprietary BI tools, such as in DAX for Power BI. This creates lock-in and makes it difficult to maintain consistency across different tools. By providing a performant SQL interface, MotherDuck encourages teams to define this logic in the data warehouse, where it can be version-controlled, tested, and reused, creating a more robust and future-proof data stack.

## Your Next Step in Modernizing Analytics

The evaluation conducted by Artefact offers a clear conclusion: MotherDuck is ready for the gold layer of the modern data stack. It provides superior performance for interactive analytics, a more cost-effective model for BI workloads, and a developer experience that accelerates productivity.

The evidence suggests MotherDuck is ready for your gold layer. The only remaining question is whether you are ready to test it. By leveraging the free tier, teams can upload their own data, connect their favorite BI tool, and validate the performance and developer experience for themselves. The results may be just as surprising as beating an Olympics champion.

Maël Deschamps

1 BLOG POSTS AND VIDEOS

Boosting Data Performance: Unlocking the Power of DuckDB in your Gold Layer

SUBSCRIBE

Subscribe to MotherDuck Blog