Benchmarks

These benchmarks show TopK's end-to-end query performance for hybrid vector search across different collection sizes and filter selectivity levels.

The metrics include median (p50), 95th percentile (p95), and 99th percentile (p99) latencies in milliseconds, as well as overall throughput in queries per second (QPS).

Selectivity refers to what fraction of the collection is scanned - from a full scan (100%) down to scanning just 1% of vectors. Lower selectivity generally yields better performance without impacting the quality of results.

1M Document Collection

Dense Vector Search

dim=768, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

1m3s

Ingest + index

Sparse Vector Search

doc_non_zero=512, query_non_zero=32, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

Ingest + index

10M Document Collection

Dense Vector Search

dim=768, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

10m30s

Ingest + index

Sparse Vector Search

doc_non_zero=512, query_non_zero=32, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

15m50s

Ingest + index

100M Document Collection

Dense Vector Search

dim=768, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

1h44m

Ingest + index

Sparse Vector Search

doc_non_zero=512, query_non_zero=32, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

2h52m

Ingest + index

1B Document Collection

Dense Vector Search

dim=768, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

17h12m

Ingest + index

Sparse Vector Search

doc_non_zero=512, query_non_zero=32, k=10

p50 (ms)

p95 (ms)

p99 (ms)

QPS

29h50m

Ingest + index