Gabia announced on January 9 that it has launched NPUaaS, a service that allows users to subscribe to Rebellions' neural processing unit (NPU) "ATOM Max" through the cloud.
NPUaaS is a service that provides NPU infrastructure on a cloud subscription basis. NPUs are semiconductors specialized for inference tasks that apply trained artificial intelligence (AI) models to actual services. Compared to graphics processing units (GPUs), NPUs consume less power and offer higher cost efficiency, attracting attention as an alternative for companies seeking to reduce GPU dependency.
Gabia's NPUaaS, launched on an instance (VM) basis, enables detailed environment configuration and customization at the OS kernel level. It supports precise infrastructure configuration tailored to specific AI workload requirements and can be flexibly scaled according to service objectives.
The ATOM Max utilized in Gabia's NPUaaS provides computing performance of 128 teraflops per chip, along with 64GB of NPU memory, 16 vCore CPU, 256GB of system memory, and 4TB of free traffic. It supports various workloads including LLM, vision AI, multimodal AI, and physical AI, with multi-configuration options available depending on deployment scale.
Gabia's NPUaaS primarily targets enterprises that need to precisely control their infrastructure environment for specific AI workloads. It is particularly suitable for workloads requiring fine-tuned high-performance inference environments such as large language model (LLM) serving, real-time video and audio analysis requiring low latency, and financial, healthcare, and public sectors that require kernel-level isolation due to security policies.
"We will combine NPUaaS with professional consulting to create an environment where companies can focus on AI business without infrastructure burden," said Dae-won Jung, Executive Vice President at Gabia.