Loading page
About this tool
Estimate total vector index size from vector count, dimensions and data type, including HNSW or IVF overhead and replicas.
This calculator estimates the total storage a vector index needs from vector count, dimensions and data type: raw data is count × dimensions × bytes per dimension, and index overhead follows the Faiss sizing rules (HNSW adds M × 2 × 4 bytes of graph links per vector; IVF adds 8-byte ids plus a √N centroid table). It is for engineers capacity-planning a Pinecone, Qdrant, Milvus, pgvector or Faiss deployment who need a defensible RAM and disk figure before provisioning.
Open Vector Storage Size Calculator on AltFTool — it loads instantly in your browser.
Enter the values you already know.
Fine-tune the options to match your scenario.
Read the result and use it in your planning or reporting.
HNSW graph links (M × 2 × 4 bytes per vector) and IVF ids plus centroids are added on top of the raw data, not ignored.
float64 down to 1-bit binary quantisation, with metadata bytes and replica count folded into the total.
The arithmetic follows the published Faiss index sizing guidelines rather than a vendor's marketing calculator.
Multiply vector count × dimensions × bytes per dimension for the raw data (4 bytes per dimension for float32), then add index overhead: an HNSW graph adds roughly M × 2 × 4 bytes per vector (128 bytes at the default M=16), and IVF adds an 8-byte id per vector plus about √N float32 centroids. Finally multiply by your replica count.
About 5.7 GiB of raw data for 1 million text-embedding-3-small vectors (1,536 dims × 4 bytes × 1M), plus roughly 122 MiB of HNSW graph overhead at M=16 and whatever metadata you store per vector. Most engines also want 20% or more free headroom, so plan around 8 GiB of RAM per replica.
Roughly M × 2 × 4 bytes per vector — 128 bytes at the common default M=16 — following the Faiss guideline that HNSW memory per vector is d × 4 + M × 2 × 4 bytes. For a 1536-dimension float32 vector (6,144 bytes) that is only about 2% extra, but for short or quantised vectors the graph can exceed the data itself.
Yes — int8 scalar quantisation cuts raw vector data to a quarter of float32 and binary quantisation to one thirty-second, but graph overhead and metadata do not shrink with it. That is why a binary-quantised HNSW index is dominated by its links: 128 bytes of graph per vector versus 192 bytes of data for a 1536-dimension binary vector.
Add the Vector Storage Size Calculatorwidget to your blog or website — free, responsive, no signup. Just keep the “Widget by AltFTool” credit link visible.
<iframe src="https://www.altftool.com/embed/widget/vector-storage-size-calculator"
title="Vector Storage Size Calculator — free AltFTool widget"
width="100%" height="640" style="border:0;border-radius:12px;overflow:hidden"
loading="lazy" referrerpolicy="no-referrer-when-downgrade"></iframe>
<p style="font-size:12px;margin:4px 0 0">Widget by <a href="https://www.altftool.com/tools/all/vector-storage-size-calculator?utm_source=embed&utm_medium=widget">AltFTool — free online tools</a></p>