Knowhere, the core component of the vector database Milvus Cloud, has been upgraded to support GPU indexing and Cosine similarity types.

Friends who are familiar with us know that there is a crucial component in Milvus Cloud and Zilliz Cloud - Knowhere.


 

What is Knowhere ? If the vector database as a whole is regarded as the Marvel Guardians of the Galaxy universe, then Knowhere is a veritable headquarters. Its main function is to accurately search its nearest neighbors for vectors or perform low-latency, approximate nearest neighbor searches (ANNS) by building indexes .


 

Knowhere 2.x version has been refactored since July 2022. After many iterations of solution discussion, design, development and testing, it finally meets you with Milvus Cloud2.3. For users , compared to version 1.x, Knowhere 2.x version provides a more standardized interface and richer features, such as supporting GPU index, Cosine similarity type, ScaNN index and ARM architecture, etc. For developers , the upgraded Knowhere can more easily add new indexing algorithms, which is beneficial to later maintenance.


 

Next, I will introduce you to the new features, optimization and design concepts of Knowhere 2.x in detail .


 

Guess you like

Origin blog.csdn.net/qinglingye/article/details/132764279