A 50ms query can still be slow and expensive to serve
Before data can be queried, it must be ingested, organized, and kept query-ready.
Sorting, indexing, materialized view maintenance, and freshness all take time and resources.
Traditional benchmarks usually measure only the read step.
CostBench measures the whole path.
Real-time exposes architectural tradeoffs
Most cloud data warehouses were built for batch analytics. Real-time workloads are different: data is constantly arriving and must be immediately available for analysis. CostBench measures how those architectural tradeoffs impact cost and performance from ingestion to insight.
Learn more about selecting a real-time analytical database ->->
Traditional warehouses
- Designed for batch processing
- Delayed optimization
- Scheduled refreshes
- Additional compute to maintain freshness
ClickHouse
- Query-ready on ingest
- Incremental materialized views: fresh on write
- Continuous optimization
- No separate freshness layer
Read-side cost-performance results
ClickHouse delivers the best query cost-performance of any major cloud warehouse, and the gap widens as data scales.
Lower is better. Relative to ClickHouse (1x).
- Ingest costs up to 82% less
than Snowflake and Databricks. - Preparation costs up to 65% less
than Snowflake and Databricks. - Read costs up to 83% less
than Snowflake and Databricks.
Here's what continuous ingest actually costs
Using a continuous ingest workload of one million rows per second, CostBench measures the cost of keeping data fresh and query-ready at scale.
ClickHouse
$130
per 100B rows
$41k
Annually
Snowflake
$2,800
per 100B rows
~$1M
Annually
This has been tested with the ClickBench dataset.
What systems are truly real-time?
In ClickHouse, data is query-ready as it arrives. Materialized views update incrementally in real time, with no delayed catch-up process required before data becomes usable.
With Snowflake, ordering, indexing, and optimization happen later. Fresh data may be visible quickly, but fully query-ready data can take hours.
Instant. Query-ready as data arrives. No delayed catch-up.
More results coming soon
Same full-path methodology coming for Databricks, BigQuery, and Redshift.
Real-time isn't just fast reads. ClickHouse leads on speed and cost-performance across writes and reads.
Against Snowflake, ClickHouse reaches query-ready data at 22x lower cost, with 28x better write-side cost-performance. On the read side, it leads every warehouse we tested. More write-side comparisons coming next.
Hardware-grounded billing, no opaque credits or DBUs. For agentic workloads that pressure reads and writes continuously, that adds up fast.
Open methodology
Workloads, pricing assumptions, configurations, and benchmark code are all available publicly, making it possible to inspect, validate, and reproduce every result.
Open source
Real pricing
Reproducible
Transparent
All proven by benchmarks that can be reproduced by anyone.
At ClickHouse, we measure performance relentlessly - believing every millisecond matters. We continually push to make queries faster.
We back up our claims with public, reproducible benchmarks like ClickBench and JSONBench. You can explore our results directly and compare ClickHouse against other technologies before deciding how to power your analytics.