Wave SIMD Beats FAISS: Complete Benchmark Analysis
Aug 2025 breakthrough: Wave SIMD achieves 0.28ms p50 vs FAISS 0.83ms (≈2.9×), with similar gains at p95/p99 and higher throughput.
Executive Summary
- Query Latency (P50): 0.28ms vs 0.83ms (≈2.9× faster)
- Query Latency (P95/P99): 0.31/0.35ms vs 0.95/1.10ms (≈3.1×)
- Throughput: 3,571 qps vs 1,219 qps (≈2.9× higher)
- Pipeline: ≈30× faster end‑to‑end including encoding; ≈600× vs original Python
Detailed Performance Table
| Metric | Wave SIMD | FAISS | Chroma | Pinecone | Improvement |
|---|---|---|---|---|---|
| Query Latency (P50) | 0.28ms | 0.83ms | 12.5ms | 45ms | 2.9× faster than FAISS |
| Query Latency (P95) | 0.31ms | 0.95ms | 18.2ms | 52ms | 3.1× faster than FAISS |
| Query Latency (P99) | 0.35ms | 1.1ms | 23.7ms | 61ms | 3.1× faster than FAISS |
| Throughput | 3,571 queries/sec | 1,219 queries/sec | 80 queries/sec | 22 queries/sec | 2.9× higher than FAISS |
Comparison Chart
Chart will appear when generated. See Benchmarks report above.
Methodology
See the full methodology and additional analysis in the benchmark report.
Benchmark report will be available soon.