Assaf Sella | VP & CTO
KIOXIA Israel, Ltd

Future of Memory and Storage - Day 2 @ 10:35

AiSAQ: Enabling Ultra-High-Scale RAG with Cost-Efficient All-in-Storage ANNS and GPU-Accelerated Indexing

As large language models are increasingly used for knowledge-intensive tasks, RAG has become an important mechanism for ensuring factual accuracy and relevance. The demand for high-scale RAG continues to grow as enterprises seek to index, search, and reason over ever-larger volumes of information.  Two barriers dominate the path to high-scale RAG: scaling cost and index build time. In this session, we will discuss how an all-in-storage ANNS solution cost-effectively enables RAG vector search at ultra-high, 10B scale, with near-zero DRAM requirements and a minimal number of query servers. We will also demonstrate how integration with NVIDIA cuVS significantly accelerates GPU-based AiSAQ index build, making index generation practical at scale. Finally, we will share benchmark results demonstrating performance at ultra-high scale, integrated with Milvus, a leading open-source vector database.

Assaf Sella, VP & CTO, KIOXIA Israel, Ltd

×

Assaf Sella / AiSAQ: Enabling Ultra-High-Scale RAG with Cost-Efficient All-in-Storage ANNS and GPU-Accelerated Indexing

Assaf Sella, VP & CTO,
KIOXIA Israel, Ltd

Assaf Sella is VP & CTO of the KIOXIA Israel Development Center, where he leads research in generative AI, AI infrastructure, and deep neural network technologies for enhancing Flash memory reliability. Before joining KIOXIA, Assaf served as CTO of Texas Instruments Israel and held senior...

CloseMore