KT Cloud has launched a cloud-based neural processing unit (NPU) infrastructure service.
This development enables more streamlined access to AI research and development infrastructure and marks a development in South Korea's AI sector.
Why it matters: This launch is the first in the Korean market using NPUs, AI semiconductors recognized for their optimized design for AI tasks.
The Key points
- The service launched on May 30 uses the Atom" NPU from Rebellions, a local fabless semiconductor design specialist.
- The infrastructure allows customers to create resources, manage, monitor, and deploy the NPU as needed, potentially reducing research time and costs for AI companies.
- According to MLPerf, an AI semiconductor technology verification competition, Rebellion's Atom used in the service was 1.4x to 3.4x faster for image processing and 1.4x to 2x faster for language modeling than NVIDIA's A2 and T4 and Qualcomm's Cloud AI 100.
The Big Picture: This advancement is part of a more significant move to develop NPUs for the burgeoning mega-scale AI industry.
NPUs offer faster computing speeds and lower power consumption than comparable GPUs.
KT Cloud's offering is in line with the government's "K-Cloud Project," which aims to strengthen domestic cloud competitiveness by developing and applying domestic AI semiconductors in data centers.
The service is designed to support computing resources for small, medium, and venture-backed enterprises, with expansion to corporate customers planned for H2 2023.