Qdrant's official guide: tuning optimizers across 13 configs on 1.76M vectors for stable latency
qdrant_engine · x · 2026-10-13
Qdrant published a guide plus benchmark quantifying how its optimizers (indexing, merge, vacuum) affect search latency:
- Data: 1.76M Cohere-embedded MS MARCO passages; hardware: single node with 32GB RAM, 14 Intel cores, Ubuntu 26.04.
- Right after bulk load, optimizers work through a backlog and queries can take hundreds of ms to seconds (780ms measured); once settled, latency drops to 4.3ms.
- Upload took 70–316s depending on whether indexing ran concurrently.
- Three-stage methodology: Upload (no queries, latency not measured), Draining (one query at a time with optimizers active, p50/p95), Steady (5,000 fixed queries baseline, p50/p95).
- Client colocated with instance to remove network latency; cloud deployments would see higher latencies.
Related event: Qdrant Benchmarks Show Optimizer Config Drastically Affects Search Latency(2 posts)→
More from Infra
- Modal publishes the manual for serving trillions of tokens to coding agents — charles_irl · 2026-10-13
- Basalt: A fast Blackwell inference engine for Qwen 3.8 Flash Next with dynamic vision — jesdga95 · 2026-10-13
- 80B MoE Model Runs in a Chrome Tab: llama.cpp WebGPU Backed Pages Experts From Disk — naklitechie · 2026-10-13
- The End of the Best-Model-Wins Era: AI Value Is Rotating Into Data Centers and Apps — RihardJarc · 2026-10-13
- Market rotating out of semis into hyperscalers and application layer, analyst says trend will continue — RihardJarc · 2026-10-13
- Intel hikes CPU prices 15-20% across Panther Lake and older lines; AMD expected to follow — TheZachMueller · 2026-10-13