# Quantization

> A GPU Node runs INT8 or INT4 weights: faster and smaller, and its answers are worth 5% or 15% less.

## Prerequisites

Research these first: [Model serving](https://wiki.playquorum.dev/research/model-serving.md).

## Engineering terms

- [Quantization](https://wiki.playquorum.dev/glossary/quantization.md)
- [Model inference](https://wiki.playquorum.dev/glossary/model-inference.md)

## Facts

Quantization
- **Track:** Infrastructure
- **Era:** [Series B](https://wiki.playquorum.dev/eras/series-b.md)
- **Cost:** 18.0k
- **Requires:** [Model serving](https://wiki.playquorum.dev/research/model-serving.md)

## See also

- [Research tree](https://wiki.playquorum.dev/research.md)

Source: https://wiki.playquorum.dev/research/quantization