Go to contents

THE DONG-A ILBO Logo

Open Menu Close Menu Open Search Bar
검색창 닫기

Startup

Dinotisia Unveils Key Paper at VLDB 2025

Dong-A Ilbo | Updated 2025.09.05
Dnotitia presented a major paper on the processing flow of vector databases (hereinafter referred to as vector DB) at the 51st VLDB 2025 (Very Large Data Bases International Conference). Dnotitia previously announced in July that the paper titled "Turbocharging Vector Databases using Modern SSDs" was officially accepted by VLDB 2025. With this presentation, information related to Dnotitia's vector DB has officially been recognized by the conference.


The Approach of Korea's First Vector DB Startup 'Dnotitia'


Dnotitia is a specialized company in long-term memory artificial intelligence (AI) and semiconductor integrated solutions, founded by CEO Jung Moo-kyung in June 2023. Dnotitia offers AI-related software services such as the vector DB system Seahorse and the Korean logical inference large language model DNA-R1, and is developing a dedicated semiconductor, VDPU, aiming for release next year.


The newly released paper, "Turbocharging Vector Databases using Modern SSDs," is a paper that presents the flow of processing vector data using SSDs, without mentioning VDPU architecture and detailed technology.

Instead, when it was accepted by VLDB 2025 last July, it was stated that "it is closely connected with the vector data operation accelerator and large-scale RAG system being developed by Dnotitia, and it can be directly linked to actual productization, not just theory," indirectly allowing for an inference of VDPU. Through this, it is faintly possible to understand how Dnotitia will handle vector DBs and how VDPU will process data in the future.


However, as it operates randomly, the processing speed of SSDs slows down, and the next access data cannot be known, resulting in a lower cache hit rate. Although modern SSDs are designed to make this possible, the existing input/output method is sequential processing, making it inefficient for processing vector data.


Before the advent of AI, existing x86 CPUs were sufficient for utilizing vector DBs. However, as AI emerged as a catalyst across industries and efforts to improve performance and power efficiency continued, it led to the concept of VDPU. When vector DBs are introduced to AI, search performance improves dramatically through similarity searches, but existing CPU technology cannot handle the vector DB search load at the data center level.

Since there were no previous attempts like Dnotitia's VDPU, it is significant in terms of pioneering a new market and bringing this technology to Korea. Dnotitia plans to unveil VDPU as early as this year and start services across the vector DB spectrum next year. It is hoped that Dnotitia's challenge will lead to pioneering a new market beyond efforts.

IT Donga Reporter Nam Si-hyun (sh@itdonga.com)
AI-translated with ChatGPT. Provided as is; original Korean text prevails.
Popular News