Unleashing 'Core' Power: Intel Xeon Accelerates Technology Innovation for AIGC Application Deployment

Recently, Intel and Transwarp Technology jointly unveiled their AIGC vector database solution at the 2023 China International Import Expo. This innovative solution is designed to support vast amounts of vector data generated by diverse machine learning models, addressing the high real-time query, retrieval, and recall needs of enterprises. It aims to bolster a wide range of applications in the era of artificial intelligence and accelerate business development.

As large language models evolve, generative AI is progressively infiltrating various industries. Businesses are increasingly demanding timely and accurate large language model solutions. When a model's parameters reach the tens of billions, it begins to demonstrate initial natural language comprehension capabilities. However, for effective text and code output, the parameter count must scale to between 300 billion and 500 billion, with accuracy hovering around 50%. To achieve superior accuracy, robust reasoning and computational abilities, the parameter size should ideally reach 500 billion.

Furthermore, when large models are applied to specialized domains, they often lack access to essential industry-specific data, leading to uncertainty in the provided information. By utilizing text embeddings to vectorize and store the latest and industry-specific information in a database, the pressure on large models can be alleviated, and timelier insights can be delivered.

In response to these challenges, Transwarp Technology has launched its Transwarp Hippo distributed vector database, powered by the fourth-generation Intel Xeon scalable processors. This solution benefits from high memory bandwidth and multi-core performance, significantly enhancing flexibility. The inclusion of the VNNI instruction set further optimizes vector computation performance. Testing shows that the Transwarp Hippo solution delivers an overall performance boost of 20% to 30%.

With characteristics such as high availability, high performance, and easy scalability, Transwarp Hippo can greatly expand the application boundaries of large models. It enables the models to maintain real-time information and dynamically adjust, providing a form of "long-term memory" to address the issue of "AI hallucinations." Tang Jiong, General Manager of Intel China's Software Technology Cooperation Division, emphasized that the rapid advancement of artificial intelligence not only injects new momentum into the global digital economy but also presents fresh challenges for enterprises across diverse business scenarios. With years of expertise in AI, Intel is committed to collaborating with ecosystem partners to deliver innovative product solutions that meet various business needs, thus accelerating the practical application of large models.