Cutting-Edge AI Accelerator Powers Complex Generative AI Tasks

EdgeCortix has unveiled its latest innovation, the SAKURA-II platform, which is equipped with the second generation Dynamic Neural Accelerator (DNA) architecture. This state-of-the-art platform is specifically designed to address the most demanding Generative AI tasks in the industry. With a focus on flexibility and power efficiency, SAKURA-II enables users to efficiently handle a wide range of complex tasks, including Large Language Models (LLMs), Large Vision Models (LVMs), and multi-modal transformer-based applications, all while operating within the strict environmental constraints at the edge. The platform boasts low latency, exceptional memory bandwidth, high accuracy, and a compact form factor, delivering unmatched performance and cost-efficiency for various edge AI applications.

SAKURA-II is well-suited for a multitude of industries, including manufacturing, industry 4.0, security, robotics, aerospace, and telecommunications. The platform features EdgeCortix's latest generation runtime reconfigurable neural processing engine, DNA-II, which offers power efficiency and real-time processing capabilities. By simultaneously running multiple deep neural network models with minimal latency, SAKURA-II can achieve up to 60 trillion operations per second (TOPS) of effective 8-bit integer performance and 30 trillion 16-bit brain floating-point operations per second (TFLOPS). Additionally, the platform supports mixed precision to meet the demanding requirements of next-generation AI tasks.

With its advanced MERA software suite, the SAKURA-II platform provides a heterogeneous compiler platform, advanced quantization, and model calibration capabilities. The suite includes native support for popular development frameworks like PyTorch, TensorFlow Lite, and ONNX. MERA's flexible host-to-accelerator unified runtime is capable of scaling across single, multi-chip, and multi-card systems at the edge, streamlining AI inferencing and reducing deployment times for data scientists. Integration with the MERA Model Library, along with seamless access to Hugging Face Optimum, gives users a wide range of the latest transformer models for seamless transition from training to edge inference.

"SAKURA-II's remarkable performance of 60 TOPS within 8 watts of typical power consumption, coupled with its mixed-precision and memory compression capabilities, positions it as a key technology for cutting-edge Generative AI solutions at the edge," stated Sakyasingha Dasgupta, CEO and Founder of EdgeCortix. "Whether deploying traditional AI models or the latest Llama 2/3, Stable-diffusion, Whisper, or Vision-transformer models, SAKURA-II offers deployment flexibility with superior performance per watt and cost-efficiency."

The SAKURA-II Edge AI accelerator will be available as a stand-alone device, two different M.2 modules with varying DRAM capacity, and single and dual-device low-profile PCIe cards. Customers can pre-order M.2 modules and PCIe cards for delivery in the second half of 2024.