What's next for Taiwan's Shan-2 AI supercomputer? "Taiwan Shan No. 2" was developed by the Ministry of Science and Technology in 2017 for 4 years and at a cost of 5 billion yuan, and it has entered the top 20 supercomputers in the world. With the end of the planned budget in 2020, the commercial operation rights will be transferred to TZI Cloud. How will this supercomputer, which started as a national project, be used in the future? What kind of help can it bring to the development of Taiwan's AI industry? Huang Shimin, deputy general manager of TAZ Cloud's Enterprise Solutions and Innovation Cooperation Development Division, said that Taiwan Shan No. 2 is a cluster designed for AI and Data, especially the network transmission efficiency and files of each node of computing power The reading efficiency is particularly outstanding. Huang Shimin mentioned that connecting nodes through low-latency network (InfiniBand100G) technology is 14 times faster than traditional Ethernet in terms of transmission efficiency; and because the reading efficiency of training data sets will greatly affect the training time of AI models As for the effect, it also spent a considerable budget to build a high speed file system (high speed file system), hoping to reduce the research and development time of AI projects, and at the same time improve the performance of data processing and model optimization. Which industries are being helped? In the past year, based on Taiwan Shan No. 2 and TWCC, taizhiyun has assisted a variety of industries including smart medical care, self-driving cars, and semiconductor manufacturing to carry out projects in various fields. It has served more than 150 companies so far, and it is expected to increase by the end of this year. to more than 250. Huang Shimin said that although the commercial operation right of Taiwan Shan 2 was transferred to TAIC Cloud, the property rights still belong to the country, which naturally complies with local laws and regulations. Therefore, it is especially suitable for some industries with sensitive data and cannot leave the country, such as health care Or finance, life insurance related industries. In addition, it can also bring great help to the fields of self-driving cars and biomedicine. Huang Shimin shared that there are some companies in China that are working on ADAS systems or smart car cabins. They organize a large amount of road condition data collected on the road into training data sets, and continue to strengthen the self-driving system. What these customers care about is not only computing power, but also whether your platform can become its data pipeline. Because there is so much data, an automated process is needed, from shooting the data to the cloud, cleaning and processing it into a data set and then being trained.... Can this process be automated as much as possible? For this reason, taizhiyun also hopes to build a platform that meets the needs of customers through cooperation with different industries. In addition, start-up companies engaged in new drug research and development can also use the platform to help them fulfill customer needs. 5G and blockchain will be the trend Huang Shimin said that currently the three well-known cloud platforms mainly provide comprehensive cloud services Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure. They do not provide AI supercomputer services in Asia, although they There are also services similar to GPU as Servce, but the unit price is not cheap. When TAIC Cloud's AI supercomputer service was built, it was expected to assist Taiwan's industrial innovation and academic research circles to enhance their competitiveness in developing world-class AI research and development. Therefore, it is also more favorable in terms of price. He believes that from the perspective of the services provided, the three major platforms are not competitors of taizhiyun, but partners.
Forum Role: Participant
Topics Started: 0
Replies Created: 0