ScyllaDB Vector Search Sizing Calculator
Configuration
10 K4 B
Total vectors to index. Logarithmic scale.
116 000
Embedding dimensionality. Common: 128 (image), 768 (BERT), 1536 (OpenAI), 4096 (LLMs).
101 000 000
Desired queries per second at P99 ≤ 15 ms latency.
Advanced Parameters
70 %99 %
Higher recall dramatically reduces throughput per vCPU.
11 000
Nearest-neighbour results per query. K=100 halves throughput vs K=10.
4 B1 MiB
Average payload alongside each embedding (ScyllaDB-node storage). Logarithmic scale.
020
Each column adds 30 bytes per vector to Vector Store-node RAM.
Sizing Recommendation
Adjust parameters on the left to see results.