Vector memory sizer

Reference scenarios

Calculated using the default assumptions above, varying the first input only.

How this estimate works

Raw vector bytes equal vector count times dimensions times bytes per component. The simplified HNSW graph estimate adds M times two times eight bytes per vector, then applies the overhead percentage to vectors and graph together. Outputs use GiB (2^30 bytes). This is a planning approximation, not a database vendor’s sizing guarantee. Metadata, allocator behavior, upper graph layers, deletions, replicas, disk indexes and quantization differ by implementation. Set bytes per component to match your storage format and measure a representative sample.

Start with your measured workload, adjust one assumption at a time, and compare the resulting capacity. Save the scenario link to return to these inputs or share it with your team. Reference scenarios remain fixed so you can compare against a consistent baseline.