SKT and Sapion Provide NPU Farm to University of Toronto, Canada for AI Joint Research

[Asia Economy Reporter Oh Su-yeon] SK Telecom and global AI semiconductor company Sapeon announced on the 25th that they will provide an AI semiconductor-based data center called the ‘NPU Farm’ to the University of Toronto in Canada and engage in joint AI research.

SKT, Sapeon, and the University of Toronto agreed on the 22nd (local time) through a memorandum of understanding (MOU) to provide the University of Toronto with an NPU Farm composed of Sapeon X220 and X330 AI semiconductors free of charge and to promote joint research for AI technology development using it. SKT explained that this MOU was established as part of R&D cooperation for AI advancement between South Korea and Canada.

An NPU (Neural Processing Unit) is a processor optimized for artificial intelligence (AI) deep learning algorithms. The NPU Farm is an NPU-based data center capable of processing big data quickly and efficiently like a human neural network.

The NPU Farm provided by SKT and Sapeon to the University of Toronto has a scale equivalent to a small AI data center with about 32 Peta OPS (Operations Per Second). The two companies will also provide a dedicated cloud platform (PaaS) that allows University of Toronto researchers to effectively utilize the NPU Farm.

Based on the established NPU Farm, SKT and Sapeon plan to promote various joint AI research projects with the University of Toronto, a world-leading research institution in the field of artificial intelligence.

The joint research includes the development of automatic generation and optimization technologies for artificial neural networks based on Sapeon AI semiconductors. Based on these research outcomes, AI companies are expected to be able to automatically generate artificial neural network models or optimize existing models to operate on Sapeon semiconductors.

The Sapeon AI semiconductors provided to the University of Toronto demonstrate superior performance compared to existing semiconductors, applying not only to data centers (IDCs) but also to various industries such as broadcast media image quality improvement and voice recognition performance enhancement, based on low power consumption and high performance.

In particular, the Sapeon X220, which was commercialized for the first time in Korea in 2020, recently received performance evaluations 2.3 to 4.6 times higher (data center performance measurement) compared to other products at the semiconductor performance test competition ‘MLPerf’. Additionally, Sapeon showed more than twice the power efficiency compared to other products in terms of performance per power consumption (based on maximum power consumption).

Lee Sang-ho, Chief Technology Officer (CTO) of SK Telecom, said, “It is especially meaningful that we are building AI semiconductor infrastructure in South Korea together with the University of Toronto, which is at the world’s highest level in the field of artificial intelligence,” and added, “We hope the cooperation among SKT, Sapeon, and the University of Toronto will contribute to the advancement of global AI technology.”

Ryu Soo-jung, CEO of Sapeon, forecasted, “If the excellent performance of Sapeon and the AI research expertise of the University of Toronto combine to achieve various AI research results, it will also lead to the expansion of the Sapeon AI semiconductor software ecosystem.”