[BLOG] Boost OpenSearch Vector Search performance with Intel® AVX-512

[naveentatikonda]

Describe the blog post

With the rise in usage of Large Language Models (LLMs) and Generative AI (GenAI), the workloads have increased from millions to billions of vectors. Ingesting and querying performance is impacted on such huge workloads.

Intel® Advanced Vector Extensions 512 (Intel® AVX-512) is a set of new instructions that can accelerate performance of these workloads. With ultra-wide 512-bit vector operations capabilities, Intel® AVX-512 can handle most demanding computational tasks.

In this blog, we will focus on Intel® AVX2/Intel® AVX512 comparison and benchmark OpenSearch using OpenSearch-benchmark for indexing and search and showcare how Intel® AVX-512 provides a performance boost, up to 15% for indexing and up to 18% for search, over AVX2 for both FP32 and FP16 encoding across multiple vector dimensions.

Expected Title

Boost OpenSearch Vector Search performance with Intel® AVX-512

Authors Name

Mulugeta Mammo, Assane Diop, Noah Staveley, Akash Shankaran, Naveen Tatikonda, Vamshi Vijay Nakkirtha, Dylan Tong

Authors Email

[email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]

Target Draft Date

03/13/2025

Blog Post Category

technical, partners, community

Target Publication Date

03/20/2025

Additional Info

No response

[pajuric]

Thanks, @naveentatikonda for opening the issue.