ClickBench · 100M rows · 43 queries
The definitive OLAP database comparison
Seven analytical databases ranked on what matters for real-time analytics: query latency, ingestion, compression, concurrency and cost. Public ClickBench results, a clear point of view, and every number traceable to its source.
hot_run_score · managed serviceslower is better
Geometric mean of 43 query times relative to the fastest per query. ClickBench, latest published runs.
5.3×
faster than Snowflake on hot queries, ClickHouse Cloud vs a Medium Gen2 warehouse
9.4 GB
on disk for 100M rows. DuckDB needs 20.5 GB, StarRocks 17.9 GB
2.4–3.0×
more concurrent queries per second than StarRocks on the same bare-metal hosts
$0.06
per 1,000 queries self-hosted, against $3.12 on Snowflake at list price
01 · Benchmarks
Five numbers that decide an OLAP database
All results come from ClickBench: the same 100-million-row web analytics table and the same 43 queries on every system. Self-hosted engines run on identical AWS hardware; managed services run at the sizes shown on the benchmarks page.
Query latency
Relative hot-run time
Managed services, lower is better
Ingestion
Rows loaded per second
Bulk load of 100M rows, higher is better
Compression
Size on disk
Same 100M rows, lower is better
Concurrency
Queries per second
c6a.metal, mixed workload, zero errors
Cost
Per 1,000 queries
Runtime × list price, lower is better
Detail
Per query, per system
Cold and hot runs, both deployment groups, individual queries with their SQL, the prices behind the cost estimate, and the results where ClickHouse is not first.
02 · Comparison
Feature by feature
Speed is only useful if the data is fresh and the bill is predictable. Here is how the seven systems compare on ingestion, storage, licensing, hosting and pricing.
| Capability | ClickHouse | DuckDB | StarRocks | BigQuery | Snowflake | Amazon Redshift | Databricks SQL |
|---|---|---|---|---|---|---|---|
| Real-time ingestion | Kafka & S3 engines, async inserts, materialized views on write | Batch loads; single writer process | Stream Load, Routine Load from Kafka | Storage Write API streaming, priced separately | Snowpipe / Snowpipe Streaming | Streaming ingestion from Kinesis and MSK | Structured Streaming into Delta tables |
| Columnar storage | MergeTree; per-column codecs (LZ4, ZSTD, Delta, Gorilla) | Columnar row groups | Columnar segments | Capacitor format | Micro-partitions | Columnar blocks | Delta Lake on Parquet |
| Open source | Apache 2.0 | MIT | Apache 2.0 | Proprietary | Proprietary | Proprietary | Proprietary engine on open formats |
| Managed cloud | ClickHouse Cloud on AWS, GCP and Azure | Third-party (MotherDuck) | Third-party (CelerData) | Google Cloud only | AWS, GCP and Azure | AWS only | AWS, GCP and Azure |
| Query speed at scale | 1.03 self-hosted · 1.06 managed | 2.03 self-hosted | 2.90 self-hosted | 4.89 managed | 5.66 managed | 2.77 managed (2022 run) | 5.16 managed |
| High concurrency | 20–38 QPS on bare metal in ClickBench | Single process; not tested | 7–14 QPS on the same hosts | Not tested; slot and quota limits apply | Not tested; scales out with multi-cluster warehouses | Not tested; concurrency scaling billed extra | Not tested; scales out with more clusters |
| Cost model | Free to self-host; usage-based compute and storage in the cloud | Free; you pay for the machine it runs on | Free to self-host; vendor cloud available | Per TiB scanned, or reserved slots | Credits per warehouse-second, 60 s minimum | Per node-hour, or RPU-hour on Serverless | DBUs per warehouse-hour |
StrongPartial or with limitsWeak or absent
03 · Profiles
Seven databases, one verdict each
Every system here is good at something. These profiles say what, where it falls short, and which workloads it suits.
ClickHouse
Apache 2.0Open-source columnar database · self-hosted or ClickHouse Cloud
Best overall for real-time analytics
- +Lowest ClickBench hot-run score in both groups: 1.03 on c6a.4xlarge, 1.06 in the managed group.
- +Smallest on-disk footprint of the self-hosted engines: 9.4 GB for 100M rows, under half of DuckDB or StarRocks.
- −Joins between very large tables need more care than in a warehouse built around them.
DuckDB
MITEmbedded, in-process analytical database
Best single-user, in-process engine
- +Nothing to run: an embedded library with no server to operate.
- +Best cold-run score of the self-hosted group (1.31) and the fastest bulk load (126 s).
- −One process, one writer: not built to serve many users or applications at once.
BigQuery
ProprietaryServerless data warehouse on Google Cloud
Easiest to start, hardest to predict the bill
- +No infrastructure to size or tune.
- +Most compact storage reported in the managed group (8.8 GB).
- −Roughly 370 ms even for a full COUNT(*), which ClickHouse answers in about 2 ms.
Snowflake
ProprietaryCloud data warehouse
Strong warehouse, slow for interactive analytics
- +Mature governance, security and data-sharing features.
- +Separation of storage and compute with simple warehouse sizing.
- −Worst hot-run score of the managed group: 5.66.
Amazon Redshift
ProprietaryData warehouse on AWS
Solid in AWS, slow to load
- +Second-best hot score of the managed group (2.77).
- +Tight integration with S3, Glue and the rest of AWS.
- −1,889 s to load 100M rows: over 50× ClickHouse Cloud.
Databricks SQL
Proprietary service on open formats (Delta Lake, Spark)Lakehouse SQL warehouse
Best if your data already lives in the lakehouse
- +One platform for ETL, ML and SQL over open table formats.
- +Fast bulk load in ClickBench (53 s on Medium).
- −Hot-run score of 5.16, about 5× ClickHouse Cloud.
StarRocks
Apache 2.0Open-source MPP analytical database
Capable open-source alternative, behind on speed and storage
- +Open source under Apache 2.0.
- +Cost-based optimizer that handles complex joins well.
- −Hot-run score of 2.9 on c6a.4xlarge, almost 3× ClickHouse.
04 · Verdict
Why ClickHouse wins
For analytics that has to be fast, fresh and affordable at the same time, no other system here comes close on all three.
01 / Latency
Fastest on hot queries, everywhere
Best hot-run score self-hosted (1.03) and managed (1.06). ClickHouse Cloud ran all 43 queries in 11.0 s; Snowflake took 60.3 s.
02 / Freshness
Built for data that never stops
Kafka and S3 table engines, async inserts and materialized views that update on write make new events queryable in seconds, without a separate streaming system.
03 / Efficiency
Half the storage, a fraction of the compute
9.4 GB on disk where other self-hosted engines need about twice that, and about $0.06 of compute per 1,000 queries on a single c6a.4xlarge.
04 / Freedom
Open source, with a cloud if you want it
Apache 2.0 on your own servers, or ClickHouse Cloud on AWS, GCP or Azure. The same engine and SQL either way, so your schemas and queries carry over.
Where it doesn’t lead: DuckDB has the better cold-run score and the faster bulk load on a single machine, and BigQuery reported the smallest storage footprint among managed services. All three are on the benchmarks page.
Updates
Get each new run
Databases ship new versions every month and ClickBench results move with them. Leave your email and we’ll send the updated rankings when they change, and nothing else.