ScyllaDB Vector Search Sizing Calculator

Estimate infrastructure requirements for deploying ScyllaDB Vector Similarity Search · Estimating cluster costs? Try the ScyllaDB Cost Calculator.

Cost Calculator

Adjust parameters on the right to see results.

i Total vectors to index, from 10 K to 500 M. The slider moves on a logarithmic scale.
i Embedding dimensionality, 1–4000. Common: 128 (image), 768 (BERT), 1536 (OpenAI), up to 4000 for large models.
i Desired queries per second (10–100,000) at P99 ≤ 15 ms latency.
i Fraction of true nearest neighbours returned (70–99%). Higher recall dramatically reduces throughput per vCPU.
i Nearest-neighbour results per query (1–1000). K=100 halves throughput vs K=10.
i Average payload alongside each embedding (4 B–1 MiB, ScyllaDB-node storage). The slider moves on a logarithmic scale.
i Metadata columns used for filtered search (0–10). Each column adds 30 bytes per vector to Vector Store-node RAM.