Suiyuan Technology's Zhang Yalin: Adhering to Independent Research and Innovation, Mastering the 'Source' of AI Development
As a top-tier global event in the field of artificial intelligence, the "2022 World Artificial Intelligence Conference (WAIC)" landed in Shanghai World Expo Center from September 1st to 3rd. This year's conference, themed "Intelligent Connectivity, Metaverse, and Boundless Possibilities," attracted more than 200 exhibitors from home and abroad, showcasing the latest research progress and cutting-edge applications in the AI industry, and making positive contributions to promoting data-intelligent applications to empower digital transformation.
Independent and Controllable: Mastering the "Source" of AI Development
Since the training side requires a large amount of data to train artificial neural networks, AI training chips need to be deployed in the cloud, focusing on absolute computing power. On the other hand, AI inference chips focus on comprehensive indicators and need to consider factors such as computing power per unit energy consumption, latency, and cost.
Zhang Yalin stated that since its establishment in 2018, Suiyuan Technology has established the original intention of "making large chips and competing with hard technology," and decided to challenge the AI training chips with the highest technological threshold. At that time, the most cutting-edge algorithms in artificial intelligence were all generated abroad. The fundamental reason is that the cloud training chips, as the source, have not yet been independently and controllably produced domestically. Domestic manufacturers developing artificial intelligence are like drifting duckweed, with the risk of being "stuck in the neck."
"Suisi 1.0 was successfully taped out in just 18 months because we have built a practical and experienced engineering team that achieved the predetermined goals through focused and precise daily execution." Zhang Yalin pointed out that Suiyuan Technology now has a high-level, complete, and battle-hardened full-chain team, including: an innovation research institute, a chip hardware team, a software R&D team, and a product system team. These teams cover aspects from market, product architecture development, engineering mass production, customer support, and the supply chain, providing strong support for rapid product iteration and large-scale implementation.
The Suisi 2.5 AI inference chip is based on the second-generation GCU-CARA architecture. As the computing core of the Yunsui i20 high-performance inference card, it provides strong full-precision AI computing power, an advanced storage scheme, and leading computing efficiency, widely supporting model inference for various applications such as vision, speech, NLP, search, and recommendation.
Computing Power Empowerment: Promoting Digital Economic Development
It is worth noting that Suiyuan Technology's newly launched "CloudBlazer POD" will contribute to accelerating the construction of computing power infrastructure in China.
According to the introduction, CloudBlazer POD is an artificial intelligence acceleration cluster product developed by Suiyuan Technology for large-scale, intensive AI computing power application scenarios. It integrates AI acceleration hardware, an integrated development and management platform, and supporting AI application software and services in a one-stop manner. It addresses the current pain points of weak standards and specifications, lack of professionalism, and long delivery cycles in AI computing center construction, providing customers with standardized, out-of-the-box AI computing center products that can be widely used in the computing power base of smart cities, research institute development computing power clusters, and other enterprises and institutions' standardized AI computing power needs.
This product has five main features: First, it is green and low-carbon, with a PUE as low as 1.1 or less; Second, it has high-speed interconnection, providing nearly 1 TB/s of aggregate communication bandwidth within the node and 3 200 Gb/s (RDMA) transmissions between nodes; Third, it has computing power synergy, based on the "Sui Chi" intelligent computing platform, it can unify the management and scheduling of hardware resources such as CPUs, memory, storage, and AI acceleration cards; Fourth, it has efficient deployment, supporting pre-optimized solutions, pre-integrated devices, and pre-optimized systems; Fifth, it has ultra-high computing power, with a typical configuration achieving 8 PFLOPS of TF32 floating-point computing power per unit, supporting on-demand horizontal scaling, and supporting clusters of over a thousand cards.
In Zhang Yalin's view, AI will inevitably develop in the direction of integration, systematization, compliance, and marketization in the future. Whether it is AI chips, AI systems, or AI software, their integration will be enhanced, computing power density will increase, energy efficiency will improve, and the range of applications that can enable customers will also become increasingly wide.
Founded in March 2018 in Shanghai, 燧原科技 has R&D centers in Shanghai and Beijing. As the first domestic startup to enter the data center market with high-end AI training products, it is dedicated to providing affordable infrastructure solutions for the development of the AI industry, offering independent intellectual property rights in high-computing power, high-efficiency, programmable general-purpose AI training and inference products. The company's R&D team members primarily possess over 15 years of experience in the development and mass production of high-end chips and related software ecosystems, with extensive practical experience in engineering and product development. They have successfully developed and mass-produced numerous large-scale chips.
In June 2019, China Zhongji Investment participated in the Series A financing round of 燧原科技. China Zhongji Investment is highly interested in the development of related industries and has invested in companies such as Haiguang Information and Inspur Cloud. They look forward to collaborating with more excellent enterprises and industry peers to jointly promote the growth and strength of the industry.
