GBase and Intel jointly launch a vector database solution for the era of large models.
“The innovation and application of large model technologies have expanded the use cases for vector databases and highlighted the necessity of vector database deployment. By partnering with Intel, we offer a software-hardware co-optimized vector database solution that helps users effectively address the performance pressure from massive vector data processing, building high-performance database infrastructure for the era of large models.”
— Guan Lianpo
General Manager of GBase 8a Product Operations
“Innovative technologies such as vector databases and multimodal data processing signify more changes in the database landscape, bringing greater complexity and stringent demands for diverse computing power. Intel is leveraging an innovative combination of hardware and software technologies to reconstruct and optimize infrastructure, build a new generation of database solutions, and help users fully unleash hardware potential to accelerate digital and intelligent transformation.”
— Tang Jiong
General Manager of Software Technology Collaboration, Intel China
Artificial intelligence (AI), deep learning, and other technologies have spurred the rise of vector databases. With vector databases, users can efficiently process massive volumes of unstructured data just as they handle structured data, empowering applications such as search and recommendation. At the same time, because vector databases involve indexing, retrieval, and generation of vector data, they require extensive matrix computations, placing considerable demands on database performance.
Against this backdrop, GBase launched a vector database system (GBase 8a MPP Cluster, GCVD) based on the 5th Gen Intel® Xeon® Scalable processors. This solution provides high-dimensional support, similarity search, multiple vector fields, fast insertion and updates, multi-model joint analysis, integration with large models, easy scalability, and high stability and reliability. Leveraging the built-in Intel® Advanced Matrix Extensions (Intel® AMX) accelerator engine, it boosts matrix computation performance in vector search, achieving over 2x overall database performance improvement.
A vector database is a database system specifically designed for managing vector data, with the core capability of understanding and processing high-dimensional data similarity. Through vector databases, users can better analyze unstructured data such as images and videos, empowering applications like search and content recommendation. With the rise of AI technologies such as large models, enterprise demand for vector databases has increased significantly; as a critical component of typical AI systems, they are being deployed in real-world scenarios. Key vector database use cases include:
Retrieval-Augmented Generation (RAG)
Performs similarity searches in the vector database and returns the top-K results most similar to the user's query. The results are then merged with the original question so that the large model can provide a more accurate answer.
Recommendation Systems
Uses vector similarity search to compare and compute distances between user vectors and product vectors, retrieving the top-K most relevant results and recommending products with higher matching scores to users.
Multimodal Search
Enables joint similarity searches across multiple data modalities (e.g., text, video, audio, images) using a vector database.
Additionally, while the development of vector databases drives innovation in AI and other applications, it also poses significant challenges to database infrastructure. These challenges include:
Database Performance Bottlenecks
Due to the rapid growth in vector data volume and the increasing emphasis on real-time processing, enterprises have rising expectations for vector database performance. However, performance is often severely constrained by CPU capabilities, database architecture, and other factors.
Total Cost of Ownership (TCO) Pressure
To cope with the mounting data pressure, enterprises often need to invest heavily in vector database infrastructure, deployment, operations, and implementation, leading to significant TCO pressure.
GCVD: Vector Database Based on 5th Gen Intel® Xeon® Scalable Processors
GBase's vector database system GCVD implements vector database capabilities on top of the GBase 8a MPP Cluster architecture, inheriting its high availability, high scalability, high security, and management capabilities in a distributed vector database. GCVD stores data vectors and uses vector similarity metrics to achieve efficient and accurate data search and analysis. It is suitable for a variety of AI-driven application scenarios, including image retrieval, video analysis, natural language processing, recommendation systems, targeted advertising, personalized search, intelligent customer service, fraud prevention, and genetic testing.
GCVD Architecture Diagram
GCVD adopts a compute-storage separation architecture, with each tier independently scalable. The entire distributed cluster system can simultaneously support traditional analytical workloads such as business intelligence, report analysis, and decision support, as well as vector data workloads like image processing, recommendation systems, natural language processing, and machine learning. It features high-dimensional support, similarity search, support for multiple vector fields, fast insert and update, multi-model joint analysis, large model tuning, easy scalability, and stable performance.
Boosting Database Performance with 5th Gen Intel® Xeon® Scalable Processors
5th Gen Intel® Xeon® Scalable Processors
To address performance bottlenecks, GCVD adopts the 5th Gen Intel® Xeon® Scalable processors, which deliver more reliable performance and better energy efficiency. While achieving significant per-watt performance gains for workloads, they offer higher compute power and faster memory, and are fully compatible with previous-generation hardware and software, greatly reducing testing and validation efforts.
Furthermore, GCVD fully utilizes the built-in Intel® AMX accelerator to accelerate matrix computation, efficiently handling the massive matrix multiplication operations required by various AI tasks, and improving work efficiency during training and inference.
Intel® AMX Architecture and GCVD Performance Before and After Optimization
GBase and Intel jointly verified the performance of GCVD with and without Intel® AMX optimization on a three-node cluster. Test data shows that with Intel® AMX optimization, GCVD's retrieval performance improved by up to 2.44x.
Customer Benefits
GBase's vector database GCVD, powered by 5th Gen Intel® Xeon® Scalable processors, delivers the following benefits:
Accelerate database operations to empower applications such as recommendation systems, large models, and multimodal search: effectively resolves the CPU performance bottleneck of high-performance vector databases. Combined with software optimizations, it delivers outstanding performance and can handle high-volume workloads.
Reduce TCO and improve ROI of vector database systems: enables users to achieve target performance with a smaller server footprint, helping lower costs related to server scaling, energy consumption, and data center space, thus increasing ROI.
Future Outlook
The application of AI technologies such as large models will continue to drive the adoption and growth of vector databases, underscoring the importance of addressing performance bottlenecks. GBase and Intel have jointly delivered a high-performance vector database solution that helps users protect infrastructure investments while better leveraging the advantages of vector databases in processing high-dimensional data, expanding into new business scenarios such as large models. Looking ahead, the two parties will deepen their collaboration, exploring more co-engineered software-hardware solutions to support enterprise digital transformation.