Introduce

Marsha is a next-generation, CPU-based ultra-heterogeneous AI engine designed to maximize AI inference performance by optimizing CPU architecture. Unlike traditional AI solutions that rely heavily on GPUs, Marsha leverages cutting-edge CPU features such as AVX-512, AMX-TILE, and VNNI to deliver high-performance AI inference capabilities on low-power and cost-effective devices.

This breakthrough enables edge devices, embedded systems, and traditional IT infrastructures to run complex AI models with reduced power consumption and without the need for specialized hardware like GPUs.

With Marsha, AI inference can be executed on a wide range of devices, from IoT sensors to industrial equipment, with near-GPU performance, making it ideal for lightweight and resource-constrained environments.