In May 2020, National Agriculture and Food Research Organization (NARO) had started operation of the AI supercomputer "Shiho", and "NARO Linked DB" which successfully integrated the diverse research data distributed across its various research fields. Now researchers inside and outside NARO can access and utilize the resource in the database cross-sectionally. NARO, the first national agricultural research institute to install an AI supercomputer to its facility, had strengthened its foundation of agricultural information technology through the establishment of "Shiho" and "NARO Linked DB", and is expected to further promote researches towards the realization of smart agriculture using AI technology and big data.
The shortage of a comprehensive computational resource had been an issue for NARO. Eventually, NARO installed the AI supercomputer and the large-scale database in a laboratory of NARO, Research Center for Agricultural Information Technology (RCAIT), and started operation in May 2020.
The final specification of "Shiho" and "NARO Linked DB" was carefully designed through estimating and analyzing the usage of existing computational resources within NARO and considering the benchmark of other national institutions in Japan, which resulted in the supercomputer with 1 peta FLOPS (floating point operations per second) and the database with a capacity of 3 peta byte. NARO is the first agricultural research institution in Japan to introduce a peta FLOPS class computing system.
The speed of AI processing and high-performance computing of NVIDIA Tesla V100, which is a Graphics Processing Unit (GPU) installed in "Shiho" is worth 100 Central Processing Units (CPU). The fact "Shiho" has a total of 128 GPU reflects its super high performance. An additional highlight of "Shiho" is its user-friendly interface, under the concept of a supercomputer as easy to operate as a personal computer (PC). Through its operability of input/output using web browsers and client PCs, "Shiho" enables users to easily perform interactive analysis and graphical processing. It is also equipped with AI computing program libraries and various data sets, and users can effectively utilize the facility from the very start.
On the other hand, "NARO Linked DB" integrates various research data and assets within NARO from research fields ranging from plant pest management, meteorology, genetic resources to genome information, and so forth, and enables researchers inside and outside NARO from various fields to access those data cross-sectionally. Another feature of the database solves the problem of combining multiple data sources in heterogeneity formats into a unified dataset that can be easily analyzed by "Shiho".
These provisions in agricultural information technology infrastructure will further accelerate the data-driven agricultural research in various stages: breeding, production, processing and distribution, and so forth.
Fig. 1 Comprehensive computational resource of NARO, combining the AI supercomputer "Shiho" and "NARO Linked DB"
Fig.2 The AI supercomputer "Shiho"