SSD-Sized AI Module Turns Spare Drive Bays into 20B LLM Powerhouses - tech
(hx) 05:30 PM CEST - Apr,15 2026
- Post a comment / read (1)
Unigen has launched the Amaretti E1.S AI Module, a compact accelerator based on EdgeCortix's SAKURA-II chip that fits standard M.2 or E1.S slots like a regular SSD. It delivers 60 TOPS of INT8 performance and 30 TFLOPS of BF16 at roughly 10W, while supporting up to 32GB of LPDDR4x memory.
This efficiency enables secure on-prem execution of large language models and vision-language models with up to 20 billion parameters, ideal for PCs, laptops, servers, and even ARM devices like the Raspberry Pi 5.
The module offers full compatibility with PyTorch, TensorFlow, ONNX, and Hugging Face frameworks, allowing multiple units to be stacked for scaled generative AI workloads without heavy power or thermal demands.Products include pre-installed heatsinks and are available with 14-week lead times, as confirmed on Unigen's official site and recent announcements.
All key facts from the original report—performance metrics, form factor, memory, power draw, model support, and compatibility—align directly with Unigen's product page and EdgeCortix specifications.
|
Updated:06:33 PM CEST Jun,03
(new)
