A more efficient search engine — included on Serverless
Elastic published benchmarks (Apr 2025). Elastic Cloud Serverless already runs this current engine — no self-managed Lucene / BBQ / logsdb upgrade project.
~95%
Less vector memory
BBQ vs float32
(Search Labs)
8×–30×
Higher vector throughput
Updated BBQ with SIMD
(Search 9.0 / 8.18)
~20%
Leaner log storage
Logsdb sort routing
vs 8.17 baseline
Also in the current engine
- Lucene 10 — better multi-core search parallelism and object-store I/O (Elasticsearch 9.0)
- BBQ GA — up to ~20% higher recall alongside the throughput gains above
- TSDS downsample — ~63% less time to roll metrics to a 1-hour interval
Why it matters: you get these efficiency gains automatically on Serverless — with managed HA and a 99.95% monthly uptime SLA — instead of operating upgrades on a large node estate.
Figures are from Elastic published benchmarks. Results vary by workload and data shape. Charts below unpack the same numbers.
Vector search uses far less memory
Better Binary Quantization (BBQ) compresses vectors aggressively while preserving ranking quality — so large semantic or hybrid search workloads need less memory for the same corpus.
Relative vector memory footprint
Elastic published benchmarks · float32 baseline = 100
BBQ search throughput improvement
Elastic Search 9.0 / 8.18 · SIMD path · range from published benchmarks
Faster core engine and leaner storage
Elasticsearch 9.0 is built on Lucene 10 for better parallelism and object-store I/O. Logs and metrics modes continue to cut storage and processing cost at scale.
Storage and operations improvements
Elastic Platform 9.0 · logsdb sort-field routing ~20% vs 8.17; TSDS downsample to 1h ~63% less time
What improved
Summary of Elastic published results — open each source for detail.
- Elasticsearch Labs
Vector search memory
Better Binary Quantization (BBQ) can reduce vector memory by about 95% versus float32 while keeping high ranking quality.
- Elastic Search 9.0
Vector search speed
Updated BBQ (generally available in 9.0 / 8.18) delivers up to about 20% higher recall and 8×–30× higher throughput with SIMD.
- Elastic Platform 9.0
Search engine core
Elasticsearch 9.0 runs on Lucene 10 — improved multi-core parallelism, better I/O for object storage, and more efficient sparse indexing.
- Elastic Platform 9.0
Log storage efficiency
Logsdb with sort-field routing showed about 20% additional storage reduction in benchmarks versus 8.17.
- Elastic Platform 9.0
Metrics downsampling
Time-series downsampling to a one-hour interval took about 63% less time in Elastic benchmarks — with less CPU in downsample tiers.