31 years ago, E United Group founder Yi-Shou Lin established the “Kaohsiung Institute of Technology” with the vision of cultivating high-level professional talent for the country. In 1997, it was restructured into “I-Shou University”. Today, the university has 9 colleges and 43 departments, and is one of the few comprehensive universities in Taiwan with a medical department, giving it a strong advantage in health medicine development.
With Taiwan facing a significant trend of declining birth rates and an aging population, the demand for healthcare services has surged, highlighting the urgent need for the integration of healthcare and artificial intelligence (AI). In response, I-Shou University has made “cross-domain innovation” a central development strategy, promoting interdisciplinary research and development between “health medicine” and “smart technology.”
In 2020, the university established the “School of Smart Technology” and, following the “Health Medicine Code” course, introduced the “Smart Technology Code” general education course, demonstrating its commitment to driving cross-disciplinary research. Additionally, the Department of Computer Science and Engineering helped establish the “AI High-Performance Computing Laboratory,” deploying the latest NVIDIA DGX A100 AI supercomputing system and INFINITIX’s AI-Stack machine learning/deep learning collaboration management platform. This system ensures the fair and equitable distribution of valuable AI computing resources to faculty, doctoral students, and doctors from I-Shou Hospital or I-Shou Cancer Hospital involved in academic or industry-academia research projects.
Selecting an AI Platform: Key Focus on Portal, Resource Allocation, and Pricing
Chih-Chang Chen, Director and Professor of the Department of Computer Science and Information Engineering at I-Shou University, noted that last year, the university decided to establish an “AI High-Performance Computing Lab.” In addition to purchasing AI supercomputing systems, Vice President Ji-Yen Shen advocated for a centralized portal and management system to ensure shared resources and a pricing mechanism for effective resource utilization.
The Department of Computer Science included both the supercomputer and AI collaboration management platform in the project scope. The supercomputer selection targeted DGX A100, and the AI platform allowed various vendors to submit proposals, but with mandatory support for portal, resource allocation, and pricing functions.
Chen explained that DGX A100 was chosen due to its NVIDIA A100 GPUs with “MIG” (Multi-instance GPU) functionality, allowing each GPU to split into seven independent instances for multiple workloads. With DGX A100, training, inference, and analysis tasks are all possible on a single infrastructure, saving costs on inference servers.
To verify the DGX A100’s capabilities, the university conducted a “breast cancer tissue image analysis” test. Results showed that a single A100 GPU completed model training in just 4 minutes, a 54-fold speed increase compared to the over 4-hour CPU-based training process. This impressive result further solidified the decision to purchase DGX A100, showcasing the advantages of utilizing the university’s AI Lab resources for faster, higher-quality research.
Among the vendors, DynamiX Technology proposed both the DGX A100 and the AI-Stack platform by infinitix. After thorough evaluation, I-Shou selected DynamiX’s proposal due to AI-Stack’s comprehensive portal, resource allocation, and pricing functions, which set it apart from other platforms that lacked robust pricing functionality. The project was completed by the end of last year.
Quick Model Training through Web Interface
Chen recalls that setting up the platform’s container image was a challenge due to the complexity of certain department-specific parameters required for AI-Stack. However, the technical support and execution capability of infinitix enabled the university to overcome these challenges.
For instance, MATLAB, frequently used in medical imaging, has an authentication mechanism that could hinder user experience if each login required account verification. infinitix wrote the authorization code directly into the AI-Stack platform, allowing users to access MATLAB without repeated logins, enhancing convenience.
AI-Stack benefits I-Shou in several areas. The platform is hosted on Ubuntu OS, and since some professors are more familiar with Windows than Linux, AI-Stack’s isolation features allow users to avoid complex Linux commands by executing tasks through a web interface with a few clicks, enabling them to quickly initiate training.
Given limited AI computing resources, the university prioritizes fair distribution to meet academic and industry research needs, anticipating potential competition for resources. Effective usage controls based on “money” and “time” ensure efficient resource allocation, and AI-Stack meets these expectations with precise design.
In terms of monetary control, AI-Stack supports hourly billing, as well as monthly or yearly pricing options. For time management, AI-Stack requires users to specify a project period during the application process, ensuring approval only when timelines are set. This mechanism prevents prolonged resource monopolization, encouraging applicants to produce timely results and enabling administrators to monitor usage for resource coordination if needed.
Additionally, each user in the AI Lab receives 100GB of storage for frequently used data or models. AI-Stack’s alert function notifies users when storage approaches the limit, prompting them to download less critical data or models to local storage to conserve resources.
Overall, I-Shou aims to maximize the efficiency of every AI computing resource through AI-Stack, enabling faculty and medical researchers to undertake more academic and industry projects, produce more research papers, and help cultivate AI talent for universities, hospitals, and enterprises alike.
