Summary:
- Kioxia Corporation updates KIOXIA AiSAQ software to optimize SSD use for AI vector database searches.
- New release allows system architects to balance search speed and vector count without hardware changes.
- KIOXIA AiSAQ technology enables efficient vector searches directly on SSDs, reducing DRAM limitations.
—
Article:
Kioxia Corporation Enhances KIOXIA AiSAQ Software for Improved SSD Optimization
Kioxia Corporation, a renowned leader in memory solutions, has recently introduced an update to its KIOXIA AiSAQ software. This update aims to enhance the optimization of solid-state drives (SSDs) for AI vector database searches, particularly within retrieval-augmented generation (RAG) systems. The new release, known as KIOXIA AiSAQ (All-in-Storage ANNS with Product Quantization), offers system architects the flexibility to fine-tune the balance between search speed and vector count, crucial parameters that were previously constrained by the system’s set SSD storage capacity.
One of the key features of the updated KIOXIA AiSAQ software is its utilization of an approximate closest neighbor search (ANNS) algorithm, specifically designed to maximize the efficiency of SSDs. Unlike traditional methods, this algorithm does not rely on storing index data in dynamic random-access memory (DRAM), thereby eliminating the limitations imposed by restricted DRAM capacity. By enabling direct vector searches on SSDs and reducing host memory requirements, KIOXIA AiSAQ technology revolutionizes the efficiency of AI vector database searches.
In the realm of RAG systems, balancing search performance with SSD capacity usage per vector is a critical challenge. The new KIOXIA AiSAQ software addresses this dilemma by offering a wide range of setup options to help system administrators find the optimal balance for diverse workloads. With this latest upgrade, KIOXIA AiSAQ technology transitions into an SSD-based ANNS that can cater to offline semantic searches and other vector-intensive applications, in addition to its primary function in RAG systems.
As the demand for scalable AI services continues to rise, the significance of SSDs as a viable alternative to DRAM becomes increasingly evident. KIOXIA AiSAQ software plays a pivotal role in meeting the high throughput and low latency requirements of RAG systems, enabling large-scale generative AI without being hampered by memory constraints. By promoting SSD-centric designs for scalable AI through the open-source release of KIOXIA AiSAQ software, Kioxia Corporation reaffirms its commitment to empowering the AI community with innovative solutions.
