---
title: "Discussion on \"4.1 HLOV - Shrinking Ten Million Needles: Quantization, and Why Filters Break Vector Search\" | Hashnode"
description: "Discussion on \"4.1 HLOV - Shrinking Ten Million Needles: Quantization, and Why Filters Break Vector Search\". Priya's HNSW index is fast and accurate, but it has two new problems:\n\nMemory. 10 million chunks × 3,072 bytes is 30.7 GB of RAM, and the archive grows by a million chunks a month. The fix is quantiza"
url: "https://hashnode.com/posts/4-1-hlov-shrinking-ten-million-needles-quantization-and-why-filters-break-vector-search/6afc8d182d431d793ec85ef8"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [4.1 HLOV - Shrinking Ten Million Needles: Quantization, and Why Filters Break Vector Search](https://retrieval-augmented-generation-guide.hashnode.dev/4-1-hlov-shrinking-ten-million-needles-quantization-and-why-filters-break-vector-search?utm_source=hashnode&utm_medium=feed)

Priya's HNSW index is fast and accurate, but it has two new problems: Memory. 10 million chunks × 3,072 bytes is 30.7 GB of RAM, and the archive grows by a million chunks a month. The fix is quantiza

2d ago · 11 min read · retrieval-augmented-generation-guide.hashnode.dev  
Read the article

## Written by

[Harshal Chauhan](https://hashnode.com/@harshal8140)

@harshal8140

  
[View profile](https://hashnode.com/@harshal8140)

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
