sites.fmind.dev

Portfolio 17 · byte attribution

Vector index memory and storage sizer

Attribute base vectors, metadata, documented or measured index bytes, replicas, RAM working set, build peak, and usable whole-node capacity without making performance claims.

Decision this answers: How many nodes are required by RAM and disk independently, and which visible byte component causes the next node boundary?

Scenario

Replication never changes this logical count.

measured[1]

Embedding width used in base-vector bytes.

published list[1]

Explicit storage precision.

exact-derived[1]

Measured payload, id, and alignment bytes.

measured[2]

Multiplies physical copies, not logical vectors.

measured[2]

Reserve for system and operational headroom.

model estimate[2]

The complete default scenario is server rendered. Interactions stay in this browser.

Decision summary

Logical index bytes

725.98 GiBexact-derived

Vectors, metadata, and index formula before replicas.

[1]

Steady physical storage

1.42 TiBexact-derived

Logical bytes multiplied by replicas.

[2]

Build-time peak

2.27 TiBmodel estimate

Separate from steady storage and RAM.

[2]

Whole nodes

4exact-derived

RAM binds at 4, disk at 1.

[2]
Direct-labeled result profile

The tables below are the accessible source of truth; bar lengths never carry identity alone.

Index formula ledger

CandidateVector encodingIndex bytesMetadataPerformance evidenceVersion scopeEvidence
Faiss Flat4 × dimensionsDocumentedMeasured separatelyNone claimedEngine-specificexact-derived[1]
Faiss HNSW Flat4 × dimensions4d + xM2×4Measured separatelyNone claimedEngine-specificexact-derived[1]
Qdrant measured rowConfigured precisionMeasured bytes/vectorMeasured payloadImported onlyVersion-specificmeasured[2]

Official surface versus this site

Faiss indexes

Use the official surface to confirm native prices, counting rules, limits, and product eligibility.

This offline workbench

Faiss documents index-specific bytes per vector. This page combines an explicit engine/version formula with corpus metadata, replication, usable node capacity, and a separate build-time peak.

Sources and methodology

Every result-affecting reference is visible here without JavaScript and is retained in the JSON export.

[1] Faiss indexes
Authority
Meta
Native identifier
Faiss index bytes/vector table
Unit
bytes/vector
Evidence
exact-derived
As of
2026-08-01
Retrieved
2026-08-01
Freshness
reference
Normalization
The selected formula is engine-specific; Flat and HNSW terms are not applied to unrelated indexes.

Open the authoritative source

[2] Qdrant capacity planning
Authority
Qdrant
Native identifier
capacity planning guide
Unit
RAM and disk bytes
Evidence
measured
As of
2026-08-01
Retrieved
2026-08-01
Freshness
reference
Normalization
Operational headroom, metadata, and build peak remain explicit measurements or estimates.

Open the authoritative source

Limits and non-claims

  • The selected index-byte formula is measurement/documentation proven: true.
  • No QPS, latency, recall, or quality claim is made without imported measurement evidence.
  • Shard placement, replicas, RAM working set, steady storage, and build peak are validated independently.