Revelion, an artificial intelligence (AI) semiconductor company, announced on the 21st that its neur..

Revelion, an artificial intelligence (AI) semiconductor company, announced on the 21st that its neural network processing unit (NPU) has been expanded to the core functions of SK Telecom (SKT)'s four commercial AI services.

Previously, the two companies announced in June last year that they would launch a test to apply domestic NPU to SKT's major AI services. Since then, in December of the same year, Revelion's AI semiconductor "ATOM-Max" based server was commercially applied to the "relationship estimation function" of the A-dot call summary for the first time, and it has been running services for the people for more than half a year so far.

In particular, based on the stable operation performance of the A-Dot currency summary, Revelion quickly expanded its application to SKT major services. Specifically, since June this year, NPU has been introduced sequentially to A-dot call summary "One-Line Summary Function," A-dot "Voice Synthesis Function (Streaming TTS)," AI Contact Center (AICC) "After-Counseling and Search System," and Scamvanguard "Bossy Text Detection Function."

Currently, there are about 14 million real-time AI traffic per day covered by the Revelion infrastructure in these four services. Converting this into tokens, the minimum unit for AI models to process text, exceeds 4 billion per day.

The expansion shows that domestic NPUs have become a proven infrastructure that can handle large-scale real-time traffic beyond the level of laboratory demonstrations.

By applying Revelion's NPU, SKT is streamlining company-wide computing resources by rearranging GPUs and NPUs according to the size and performance characteristics of AI models. In addition, infrastructure operational efficiency was increased by forming an "NPU Farm" that combines dozens of NPUs and hundreds of cards, and several in-house services shared the same NPU resources.

Park Sung-hyun, CEO of Revelion, said, "The past year has been a time for domestic NPUs to move from possibility to proven infrastructure. The fact that the four services that the people use every day are running stably on the Revelion infrastructure is a solid proof above all else, he said. "We will continue to expand the base of AI services at home and abroad based on large-scale commercial operations."

Park Byung-kwan, head of SKT Core Platform, said, "As demand for AI computing increases, we are actively introducing and utilizing Revelion's NPU as well as GPUs. We selectively utilized and redeployed GPUs and NPUs according to the characteristics of each AI service task, increasing the efficiency of using AI resources across the company, he said. "We plan to expand NPU Farm in the future to maximize synergy using domestic NPU semiconductors and respond quickly to demand for AI computing at home and abroad."