We will introduce our new Rebellions' AI accelerator for Datacenter Inference, ATOM, which achieve outstanding latency and energy efficiency. A performance snapshot was submitted to MLPerf 3.0 benchmark last April and proved that our chip can achieve #1 latency performance compared to other same-class AI accelerators/GPUs.This accomplishment utilizes the versatile SoC architecture designed for different types of neural networks and applications, including computer vision, language models and generative AI. We are actively collaborating with different partners and customers spanning from the finance markets to KT datacenter. Some examples of collaboration works will be introduced in this talk.

TBA