> For the complete documentation index, see [llms.txt](https://docs.zingg.ai/latest/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zingg.ai/latest/stepbystep/hardwaresizing.md).

# Hardware Sizing

Hardware required for different sizes of data

Zingg has been built to scale. Performance is dependent on:

* The **number of records** to be matched.
* The **number of fields** to be compared against each other.
* The **actual number** of duplicates.

Here are some performance numbers you can use to determine the appropriate hardware for your data.

* 120k records of **examples/febrl120k/test.csv** take 5 minutes to run on a 4 core, 10 GB RAM local Spark cluster.
* 5m records of [North Carolina Voters](https://github.com/zinggAI/zingg/tree/main/examples/ncVoters5M) take \~4 hours on a 4 core, 10 GB RAM local Spark cluster.
* 9m records with 3 fields - first name, last name, email take 45 minutes to run on AWS m5.24xlarge instance with 96 cores, 384 GB RAM
* 80m records with 8-10 fields took less than 2 hours on 1 driver (128 GB RAM, 32 cores), 8 workers (224 GB RAM, 64 cores). This is a user-reported stat without any optimization.
*

```
<img src="https://github.com/user-attachments/assets/4dadeb56-9d66-4ed2-be6e-e0f0b9ab68c6" alt="image" height="1091" width="2271">
```

If you have up to a few million records, it may be easier to run Zingg on a single machine in Spark local mode.