Milvus 2.3 version launched with support for Nvidia GPUs

Zilliz has released the beta version of Milvus 2.3, the latest version of its open source vector database. Milvus 2.3 supports Nvidia GPUs, which Zilliz says provides greater flexibility and improves the performance of real-time workloads. Zilliz claims Milvus 2.3 is 10 times faster than Milvus 2.0 when using GPUs and 4 times faster when using only CPUs.

The Milvus 2.3 update is timely due to the hype surrounding the expansion of generative artificial intelligence applications, which is one of the primary use cases for vector databases due to the size and complexity of the machine learning models on which artificial intelligence is built.

"Vector databases will be indispensable for organizations building their own large language models," Nvidia CEO Jensen Huang said in a keynote at the GTC developer conference this week.

Embedding vectors are numeric representations of unstructured data objects, such as documents, image components, video frames, or geospatial data. They enable fast and scalable similarity search by finding the closest matches between similar objects. Embeddings are generated by AI models, in particular machine learning or deep learning models trained on huge amounts of data, much of which is unstructured, such as the Internet text dataset used to train GPT models in OpenAI. The unstructured data is transformed into lists of floating point values, creating searchable embeddings.

Unlike traditional databases with refined vector functionality, Milvus was specifically designed to support artificial intelligence-based applications. The database stores, indexes and manages billions of embedding vectors generated by machine learning models, including large language models as well as convolutional networks - deep learning algorithms used for purposes such as computer vision.

At its GTC developer conference, Nvidia unveiled a new Milvus integration with the RAFT graph acceleration library, which contains algorithms for data science, graph and machine learning. The library accelerates indexing, data loading, and batch neighbor search in a single query. In addition to Milvus, Nvidia is integrating RAFT into Meta's FAISS (Facebook AI Similarity Search) library, as well as Redis.

Scaling vector search across billions of embeddings can be computationally intensive. Zilliz claims that Milvus is the first vector database to support heterogeneous computing, combining CPUs and GPUs to optimize the performance of real-time recommender systems, question-and-answer systems, anomaly detection, image and video search, and other similarity search applications. The Milvus 2.3 release brings heterogeneous computing capabilities to the reorganized Milvus 2 cloud platform, which the company says delivers hybrid search, customizable consistency and always-on, real-time performance. GPU-accelerated Milvus will continue to improve vector search capabilities for ML and AI-powered applications, Zilliz said.

"Heterogeneous computing is key to achieving the performance required for AI-based applications," said Charles Xie, creator of the Milvus project and CEO of Zilliz. "With Nvidia GPU support in Milvus and RAFT-based integration, these capabilities are now available at massive scale on both CPU and GPU platforms - or both."

In the release, Zilliz outlined other notable features of Milvus 2.3. Change Capture provides a continuous stream of database updates for backup and synchronization without downtime, as well as for rolling updates. In addition, nine different index types are now supported.

Another notable feature is the range search, which allows finding all vectors within a given distance, which can be useful for complex data queries. In addition, an index on disk is provided to optimize memory usage.

"Support for Nvidia GPUs in the latest version of Milvus will provide tremendous heterogeneous computing benefits for real-time applications," said Kari Briski, vice president of software product management at Nvidia. "Milvus is a high-performance vector database, and with the massive parallelism of Nvidia GPUs, users can now accelerate compute pipelines."

Milvus was first released in 2018 and is a Linux Foundation AI & Data graduated-stage project with a large community of contributors and users. You can learn more about Milvus 2.3 from the company's blog.