A benchmark comparison between SingleStore and ClickHouse for vector search workloads using a 45,466-movie dataset with 768-dimensional embeddings. SingleStore ran on half the CPU resources (2 vCPUs vs 4 vCPUs) yet matched or outperformed ClickHouse across most scenarios. Key findings: SingleStore's DOT_PRODUCT index reduced query times to ~70ms vs ClickHouse's 2+ seconds; L2 distance queries completed in ~75ms vs 241-366ms for ClickHouse; concurrency tests showed 50-260ms improvements per query. Cold-cache performance was 3-7x faster on first run. SingleStore's native VECTOR type, ANN index support (IVF_PQFS, HNSW), and unified architecture are cited as key differentiators for production AI applications.
Table of contents
IntroductionKey Performance FindingsPortal-Based Query ResultsWhy SingleStore Excels for Vector WorkloadsThe Bottom LineSort: