① According to media reports, at the GTC2024 conference, the world's first GPU acceleration vector database was launched, and Zilliz and Nvidia joined hands to release Milvus 2.4. ② As a “memory” for large models, the importance of vector databases is self-evident. The application of vector databases is expected to grow rapidly in the future as the development and usage of large generative AI models increases.
According to media reports, at the GTC2024 conference, the world's first GPU-accelerated vector database was launched, and Zilliz and Nvidia joined hands to release Milvus 2.4. It is reported that this is a revolutionary vector database system. For the first time in the industry, it uses the efficient parallel processing capabilities of Nvidia GPUs and CAGRA (CUDA-Accelerated Graph Index for Vector Retrieval) technology newly introduced in the RAPIDS CUVs library to provide GPU-based vector indexing and search acceleration capabilities. Benchmarks showed that compared with the most advanced CPU-based indexing technology currently on the market, the new GPU-accelerated Milvus can provide up to 50 times better vector search performance.
As the “memory” for large models, the importance of vector databases is self-evident, and they will help the development of big AI models in the future. The GF Securities computer team said that in the past, when the amount of data for AI model training was small and the data type was single, there were few scenarios where vector databases could be applied. Since the launch of the Transformer model in 2017, various technology vendors began exploring big language models, and demand for vector databases only began to take shape. In the future, vector database applications are expected to grow rapidly as large-scale generative AI models are developed and used.
According to the Finance Federation's theme library, among the relevant listed companies:
Starlink Technology's vector database Hippo can store, index, and manage vector data sets, expanding the time and spatial dimensions of large models. The company and Intel also jointly released an AIGC vector database solution, which can achieve high-real-time query, retrieval, and recall functions of massive vector data.
Daily interactive vectorization technology has been implemented in the company's business. Through machine learning and deep learning models and technology, a user vector library has been established, and data vectorization technology has been accumulated to help improve model capabilities.