The Amaretti AI Module series (available in both E1.S and E3.S form factors) delivers 60 TOPS in 10 watts (6 TOPS per watt) through the EdgeCortix SAKURA-II processor with up to 32GB memory, enabling up to 20B parameter LLMs and VLMs without compromising battery or thermal performance. With pre-validated AI capabilities and full ecosystem compatibility, it accelerates time-to-market while supporting TensorFlow, PyTorch, ONNX, and Hugging Face frameworks for parallel model execution with minimal latency.
Amaretti
Features
- E1.S / E3.S AI Module
- AI Accelerator: SAKURA-II
- Up to 1920 TOPS of inference performance with air-cooled Dual CPU Servers
- Use 20% of the Watts with TPUs compared to training GPUs
- GenAI LLMs for up to 20B parameters
- 16-week lead times, significantly less than the typical for GPU servers
- Up to 32GB per module
Benefits
Ultra-Efficient Intelligence
The Amaretti E1.S / E3.S AI Module delivers 60 trillion operations per second (TOPS) of AI computing power while consuming only 10 watts. Amaretti E1.S / E3.S is powered by the EdgeCortix SAKURA-II accelerator, designed to deliver fast, real-time AI inference with high performance in a compact, low power form factor. SAKURA-II is designed to handle the most challenging GenAI applications at the edge, enabling developers to generate new content based on diverse inputs such as images, text, and audio. Amaretti features up to 32GB memory, and the module delivers an industry-leading 6 TOPS per watt. This level of efficiency allows enterprises to run up to 20 billion parameter LLMs and VLMs locally, without compromising thermal management or system power constraints.
Accelerating Your AI Market Opportunity
Amaretti E1.S / E3.S enables the delivery of specialized AI hardware that maintains full compatibility with broader hardware ecosystems, offering a significant time-to-market advantage through pre-validated processing capabilities. Beyond raw text processing, Amaretti delivers low-latency performance for Text-to-Speech, Visual Language Models, and YOLO object detection. By leveraging its proprietary MERA SDK, Amaretti ensures that AI tasks never compete for resources, excelling in time-sensitive requirements for mission-critical robotics, secure AI agents, aerospace, defense and intelligent industrial monitoring applications.
Comprehensive Safe AI Framework Ecosystem
Amaretti provides flexibility for AI architects and system integrators, featuring seamless integration with industry-standard frameworks including TensorFlow, PyTorch, ONNX, and Hugging Face. This modular approach protects infrastructure investments while delivering up to a 10x reduction in operational costs compared to cloud-dependent architectures. Crucially, Amaretti eliminates security and data privacy concerns through its on-prem architecture, allowing enterprises to leverage their proprietary IP without ever exposing sensitive data to the public web.
Specifications
| Amaretti AI Module | ||
| AI Processor | EdgeCortix SAKURA-II | |
| Form Factor | E1.S/ E3.S (EDSFF) | |
| Power (Max) / Idle | 10 Watts / 6.5 Watts | |
| Performance | 60 TOPS | |
| Interface | EDSFF (PCIe Gen 3) | |
| Supported Frameworks | TensorFlow, ONNX, Pytorch, Hugging Face | |
| Supported OS | Linux, Windows | |
| Certification | SCE, FCC | |
| Dimension | E1.S: 33.75mm x 118.75mm x 9.5mm (5.9mm w/o Heat Sink); E3.S: 11.27cm x 7.6cm x 0.75cm (5.9mm w/o Heat Sink) | |
| Operating Temperature | 0°C ~ 70°C |
Ordering Information
| Part Number | Note |
| UCT106-016-HS | Amaretti - E1.S AI Module with EdgeCortix SAKURA-II and 16GB, 9.5MM thickness |
| UCT106-017-HS | Amaretti - E1.S AI Module with EdgeCortix SAKURA-II and 32GB, 9.5MM thickness |
| UCT106-016-E3 | Amaretti – E3.S AI Module with EdgeCortix SAKURA-II and 16GB, 7.5MM thickness |
| UCT106-017-E3 | Amaretti – E3.S AI Module with EdgeCortix SAKURA-II and 32GB, 7.5MM thickness |