A new AI module runs generative models locally, reducing power use and cloud reliance while handling complex workloads with efficiency.
Unigen has introduced the Amaretti E1.S AI Module, its first hardware solution built specifically for generative AI. The module is designed for on-premise deployment and can handle large language and vision models with up to 20 billion parameters, offering an alternative to cloud-based AI processing.
At its core, Amaretti delivers 60 trillion operations per second while using only 10 watts of power. It runs on the EdgeCortix SAKURA-II accelerator, which is optimized for real-time AI inference in compact systems. With up to 32GB of memory and an efficiency of 6 TOPS per watt, it’s built to run demanding workloads locally without pushing thermal or power limits.
The module is intended to work alongside AMD and Intel servers, fitting into existing enterprise infrastructure without requiring major changes. It supports a range of AI tasks beyond text generation, including text-to-speech, visual language models, and object detection systems like YOLO, all with low latency.
Unigen is positioning Amaretti as a way to speed up deployment of AI systems by offering pre-validated hardware. Its MERA software development kit manages how tasks are distributed across the system, preventing resource conflicts and keeping performance stable in time-sensitive applications such as robotics, aerospace, defense, and industrial monitoring.
The system also integrates with widely used AI frameworks like TensorFlow, PyTorch, ONNX, and Hugging Face, making it easier for developers to adopt without overhauling their existing workflows. Running AI workloads locally instead of in the cloud can significantly reduce operational costs and avoids sending sensitive data outside the organization, which is a key concern for many enterprises.
Amaretti builds on Unigen’s earlier AI products focused on object detection, including the Cupcake Edge AI Server, Biscotti E1.S module, Poptart E3.S module, and Poundcake VMS Server, but shifts the company more directly into the generative AI hardware space.
“As part of our expanding roadmap for Edge AI modules and servers, the Amaretti E1.S is the first in a line of modules designed to support the latest in Generative AI,” said Paul W. Heng, Founder and CEO of Unigen.
Click here for the original announcement.
