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 but requires effective filtering or indexing strategies.

1M Document Collection

Dense Vector Search

1M Documents, 768 Dimensions, k=10
Reducing scan size from 100% to 1% improves median latency by ~28% (6.8ms → 4.9ms) and throughput by ~17% (54 → 63 QPS)
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100% (Full Scan)	6.8	9.1	13	54
10% (100K docs)	5.25	7.3	10	61
1% (10K docs)	4.9	6.7	9	63

p50 (ms)

p95 (ms)

p99 (ms)

Sparse Vector Search

1M Documents, document non-zero dims=512, query non-zero dims=32, k=10
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100%	2.78	2.96	3.20	223
10%	2.70	2.87	3.10	181
1%	2.61	2.87	3.01	182

p50 (ms)

p95 (ms)

p99 (ms)

10M Document Collection

Dense Vector Search

10M Documents, 768 Dimensions, k=10
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100% (Full Scan)	16.5	20	25	36.5
10% (1M docs)	11	15	19	46
1% (100K docs)	9.5	13	16	49

p50 (ms)

p95 (ms)

p99 (ms)

Sparse Vector Search

10M Documents, document non-zero dims=512, query non-zero dims=32, k=10
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100%	12.9	13.5	15.0	63.0
10%	11.77	12.9	14.6	68.5
1%	11.06	12.9	13.3	70.0

p50 (ms)

p95 (ms)

p99 (ms)

100M Document Collection

Dense Vector Search

100M Documents, 768 Dimensions, k=10
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100% (Full Scan)	24	34	41	27
10% (10M docs)	18.5	29	34	34
1% (1M docs)	17	27	33	36

p50 (ms)

p95 (ms)

p99 (ms)

Sparse Vector Search

100M Documents, document non-zero dims=512, query non-zero dims=32, k=10
Selectivity	p50 (ms)	p95 (ms)	p99 (ms)	QPS
100%	13.74	14.85	15.8	56.7
10%	12.33	14.85	15.55	63
1%	11.95	14.17	15.5	67

p50 (ms)

p95 (ms)

p99 (ms)

Latency Scaling - Dense

Summary by Collection Size (100% Selectivity)
Shows how query performance scales with collection size using a single query path replica and single client.
Collection Size	p50 (ms)	p95 (ms)	p99 (ms)	QPS
1M	6.8	9.1	13	54
10M	16.5	20	25	36.5
100M	24	34	41	27

p50 (ms)

p95 (ms)

p99 (ms)

Latency Scaling - Sparse

Summary by Collection Size (100% Selectivity)
Shows how query performance scales with collection size using a single query path replica and single client.
Collection Size	p50 (ms)	p95 (ms)	p99 (ms)	QPS
1M	2.78	2.96	3.2	223
10M	12.9	13.5	15	63
100M	13.74	14.85	15.8	56.7

p50 (ms)

p95 (ms)

p99 (ms)

QPS vs Concurrency

Queries per second (QPS) achieved with different number of concurrent clients and single query path replica.

1% Selectivity

10% Selectivity

100% Selectivity

Achieving higher QPS can be easily achieved by provisioning more query path replicas.