Weaviate 1.39 adds 4-bit Rotational Quantization, cutting heap memory 45%
CShorten30 · x · 2026-09-17
Weaviate 1.39 extends Rotational Quantization with 4-bit support: roughly the same recall as 8-bit RQ while reducing heap usage by 45%, with faster imports and search gains across the RQ family.
Key technical details:
- Built on Extended-RaBitQ with a Fast Walsh-Hadamard Transform for the random rotation; 1.39 adds SIMD support for FWHT, bit-identical to the Go reference but much faster.
- Rotations, distance kernels, encoding, and the memory path were all overhauled; 8-bit RQ now imports 16% faster and remains the default in Weaviate Cloud.
- The post also covers how RQ recall holds as datasets scale and how it compares with TurboQuant.
4-bit RQ is available now for memory/cost-sensitive deployments, with a detailed engineering blog.
More from Infra
- Robotics startup Watney raises $80M to expand data-center-building robot fleet — Polymarket · 2026-09-18
- OpenAI Says AI Can Automate 80% of GDP; Swyx Argues Chips and Memory Supply Are the Bottleneck — iamrobotbear · 2026-09-18
- Oracle: renewed GPUs fetched 20% premium in Q1, most 4+ years old — Beth_Kindig · 2026-09-18
- LinkedIn to present a PyTorch-native GPU retrieval engine powering feed and search — PyTorch · 2026-09-18
- vLLM boosts Kimi K3 serving throughput 2.2-2.8x with scheduler, KDA and MoE kernel optimizations — vllm_project · 2026-09-17
- AWS compares Bedrock RAG vector stores: OpenSearch vs pgvector vs S3 Vectors — AWS ML Blog · 2026-09-17