KT Cloud releases AI inference GPU infrastructure service

South Korea’s KT Cloud on Monday launched the service AI SERV, which allows the use of infrastructure for high-performance graphics processing units (GPU) designed for artificial intelligence (AI) inference at reasonable cost. 

AI inference requires the use of low-capacity GPUs at all times, while large-scale AI learning intensively employs large-capacity GPUs for a short time.

“Use of the infrastructure for inference learning incurs more cost than necessary,” a KT Cloud source said. “We expect high demand for GPU infrastructure services specialized for inference.”

AI SERV uses slicing technology in which the GPU service, which was previously provided in a single unit with a value of 1, is divided into five units each with a smaller value of 0.2. This prevents infrastructure waste by reducing the minimum usable unit. 

KT Cloud targets as clients of AI SERV corporate providers of AI services after the company completes AI development and learning.
 
Write to Ji-Eun Jeong at jeong@hankyung.com