Taiwan's next-generation supercomputer for AI set to launch in May

Taiwan is set to launch its next-generation supercomputer for artificial intelligence (AI) in May, with plans to significantly expand its capabilities in the coming years, the Taipei Times reported.

According to the report, the new system, which will be equipped with Nvidia's H100 graphics processing units (GPU), will initially offer a computing capacity of 16 petaFLOPS. Further upgrades are expected to increase its power to as much as 200 petaFLOPS. This cutting-edge supercomputer is designed to facilitate the rapid development of generative AI and strong AI, which are key to advancing AI technologies such as machine learning, natural language processing, and autonomous systems.

The National Centre for High-performance Computing (NCHC) is overseeing the construction and operation of the supercomputer, which will be a vital resource for researchers in fields including physics, chemistry, mathematics, atmospheric science, engineering, life science, and artificial intelligence. The system will be available to researchers, academic institutions, and government agencies through an application process, the report added.

This initiative is part of Taiwan's broader efforts to lead the development of next-generation technologies, with the supercomputer serving as a critical tool in pushing the boundaries of AI research.

Taiwan's current AI supercomputer, Taiwania 2, built in 2018, has a computing power of nine petaFLOPS and has been extensively used for AI tasks such as deep learning and data processing. Taiwania 2 was built using Nvidia's V100 GPUs and has played a crucial role in advancing AI capabilities within Taiwan. However, as AI continues to evolve rapidly, there is an increasing need for more powerful systems, according to the Taipei Times report.

To address this, the NCHC has designed a successor to Taiwania 2 that utilizes Nvidia's H100 GPUs, providing a substantial performance boost, starting with the 16 petaFLOPS of initial computing power, the report stated.

Additionally, as per the report, the NCHC aims to increase the new supercomputer's computing power to 100 petaFLOPS and eventually to 200 petaFLOPS, depending on the available funding for the project. The expansion of this supercomputer's capabilities will not only advance generative AI models, which are used to generate text, images, and videos, but will also support the development of strong AI. Strong AI refers to machines capable of human-like cognition, designed to learn, reason, and adapt in ways that mirror human intelligence. This will enable more complex and sophisticated AI applications.

The supercomputer will also be integrated into an AI cloud service platform, which will include a traditional Chinese version of the Trustworthy AI Dialogue Engine. This platform aims to support a variety of industries by providing computing power and a robust language model base, reducing costs for companies developing large language models and encouraging the widespread adoption of AI technology across various sectors. It will allow industry operators to create customized knowledge bases and applications tailored to their specific needs and datasets, the report stated.

Beyond the new supercomputer, Taiwan's NCHC also operates other high-performance computing systems, such as Taiwan 3 and Forerunner 1, which are used for diverse computing tasks. In line with Taiwan's broader technological ambitions, the NCHC plans to further enhance computing power and energy efficiency through initiatives like the Taiwan Chip-based Industrial Innovation Program and the Southern Silicon Valley project, the Taipei Times reported.

These efforts aim to achieve a total of 480 petaFLOPS of computing power in the government sector by 2029, with the overall national computing power expected to exceed 1.2 exaFLOPS by that time when contributions from the private sector are included, the report added.