KIOXIA AiSAQ uses SSDs instead of RAM, dramatically improving AI performance

KIOXIA AiSAQ (All-in-Storage ANNS with Product Quantization) is a research project created to transition from a "mostly in-storage" solution to an all-in-storage solution for AI. It's a groundbreaking technology because it offloads compressed vectors in large datasets from DRAM to storage. With memory limitations often being the key factor in AI workload speed and efficiency, AiSAQ exponentially improves performance.

4

AI is all about data, lots of it. There is so much data that it's often hard to wrap your head around the complexity of it all. AI models and datasets cover a range of things, including documents, images, music, and more. Retrieving specific data involves Approximate Nearest Neighbor Search (ANNS), vectors, and indices. These are usually stored on fast memory, like DRAM - KIOXIA AiSAQ moves this to the SSD.

KIOXIA's initial research paper on the technology from 2024 calls it a "novel method of index data placement." However, with only around 10MB of DRAM being used in an AiSAQ system regardless of the scale of datasets, with millisecond-order latency, it sounds more like a game changer.

4

The following chart is an eye-opener, showing the dramatic increase in DRAM capacity the larger the dataset or RAG database size. Even compared to the hybrid DiskANN solution, which leverages SSDs to reduce the DRAM footprint, the accuracy of AiSAQ is impressive.

4

KIOXIA is presenting its KIOXIA AiSAQ (All-in-Storage ANNS with Product Quantization) project at CES 2025, so if you're in the business of Large Language Models (LLM) and Retrieval-Augmented Generation (RAG), be sure to check it out.