# Performance

We regularly test our architecture, infrastructure, and query engine at scale to ensure Altertable delivers consistent performance as your data grows.

## Benchmarking

One benchmark we use is TPC-H at scale factor 100. It provides a repeatable workload for query performance comparisons.

Benchmark numbers are directional: real-world performance depends on data shape, storage layout, query mix, and cache warmth.

### Query Time

Average TPCH Benchmark (sf=100) results - lower is better

| Database   | Size    | Query time |
| ---------- | ------- | ---------- |
| Altertable |         | <1s        |
| Snowflake  | X-Large | 1s         |
| ClickHouse |         | 2s         |
| Databricks |         | 3s         |
| BigQuery   |         | 3s         |
| Fabric     |         | 3s         |
| Snowflake  | X-Small | 4s         |
| Redshift   |         | 4s         |

## Cold and cached queries

A cold query may need to download data from object storage to the [Worker](/docs/lakehouse/compute/workers.md) before executing it. Later queries can reuse the Worker’s local cache, reducing storage reads and latency. Compare queries with similar cache conditions when measuring performance.

## Choose the next optimization

Start with the slow query and its scan size:

- [Inspect the query](/docs/api-references/lakehouse.md#post-explain) to estimate rows, files, and bytes scanned.
- [Choose compute](/docs/lakehouse/compute/sizing.md) for the workload.
- [Optimize table layout](/docs/lakehouse/features/optimize-table.md) when file layout or compaction limits performance.