# 4.5× faster than Databricks

Source: https://www.hotdata.dev/compare/databricks
Site index: https://www.hotdata.dev/llms.txt

On half the hardware.

## At a glance

| | Hotdata | Databricks |
| --- | --- | --- |
| What it is | A Rust query engine built for low latency at high concurrency. | A lakehouse platform. |
| Where data lives | Queries Delta Lake and Unity Catalog tables in place. | Delta Lake and Iceberg tables in your storage. |
| Deployment | In your VPC, run by us. Or fully managed. | Managed, with compute in your cloud or theirs. |
| Pricing | Compute runs only while queries run. | DBUs, plus your cloud bill. |
| Concurrency | 4.5× faster than Databricks with 10 concurrent clients. | Adds clusters as queries grow, at extra cost. |
| Snapshots and copies | Snapshots, time travel, and forks. | Delta clones and time travel. |
| Search | Vector, full-text, and geospatial in one SQL query. | Vector search as a separate service. |
| SQL | Postgres-familiar HotSQL. PostgreSQL, DuckDB, Snowflake, and ClickHouse SQL run as written. | Spark SQL. |

## Performance

TPC-H SF1 · 22 queries · 10 concurrent clients.

| System | Hardware | p50 | p95 | Queries/s |
| --- | --- | --- | --- | --- |
| Hotdata | 4 vCPU · 16 GiB | 0.17 s | 0.38 s | 51.1 |
| Databricks | ~8 vCPU · ~64 GiB · Pro | 0.76 s | 1.18 s | 12.4 |

## When to choose Hotdata

- You need answers 4.5× faster, on half the hardware.
- Apps send bursts of short queries, and paying for a warehouse to idle between them adds up.
- You want your lakehouse queryable in milliseconds, without risk of changing it.
- You want vector, full-text, and SQL filters in one query.

## When Databricks still fits

- You run Spark ETL, streaming, and machine learning training on one platform.
- Your workloads are large batch jobs rather than many short queries.

## Using Hotdata with Databricks

Connect Databricks and Unity Catalog as read-only sources. Your lakehouse stays the source of truth. See [Connecting a source](/docs/pull-data).

## Get started

The [quick start](/docs/quick-start) gets a table into Hotdata in a few minutes.