Generative AI systems require significant computing, memory, and storage resources. While they have the potential to drive transformative advancements across various sectors, their implementation often comes at a high cost. Regenerative AI (RAG) is a critical phase of AI that refines large language models (LLMs) with enterprise- or application-specific data.
A core component of RAG is a vector database that accumulates and converts specific data into feature vectors within the database. RAG also utilizes an Alternative Number Scanning (ANNS) algorithm, which identifies vectors that improve the model based on the similarity between the accumulated vectors and the target vectors.
For RAG to be effective, it must quickly retrieve the most relevant information for a given query. Traditionally, ANNS algorithms were deployed in DRAM to achieve the high-speed performance required for these searches.
KIOXIA's AiSAQ™ technology provides a scalable and efficient ANNS solution for large-scale datasets with negligible memory usage and very fast index changes.
A key feature of KIOXIA's AiSAQ™ technology
is that it allows large-scale databases to operate without relying on limited DRAM resources, thus improving the performance of RAG systems.
It eliminates the need to load index data into DRAM, allowing the vector database to start instantly. This enables seamless switching between user- or application-specific databases on the same server for efficient RAG service delivery.
Optimized for cloud systems by storing indexes in disaggregated storage for sharing across multiple servers. This approach dynamically adjusts search performance in vector databases for specific users or applications and facilitates rapid migration of search instances between physical servers.
“KIOXIA’s AiSAQ™ solution paves the way for near-infinite scalability of RAG applications on generative AI systems based on SSD flash drives in the core,” said Axel Stoermann, Vice President and Chief Technology Officer of KIOXIA Europe GmbH. “By utilizing SSD-based ANNS technology, we reduce reliance on expensive DRAM while meeting the performance requirements of leading in-memory solutions, significantly improving the performance range of large-scale RAG applications.”
