EdgeCortix was founded with a transformative vision: “Bring cloud-level artificial intelligence (AI) performance to the edge—enabling real-time, low-power AI computing in resource-constrained environments.” That vision has been brought to life in SAKURA-II, the flagship AI accelerator redefining how businesses deploy intelligent systems everywhere from the factory floor to city streets.
SAKURA-II is specially designed for edge AI tasks, delivering up to 60 TOPS of performance with top-tier power efficiency—outperforming traditional edge GPUs by up to 5 times in performance per watt. Thanks to EdgeCortix’s flexible dataflow-based Dynamic Neural Accelerator® (DNA) architecture, SAKURA-II attains high efficiency for demanding applications, from vision transformers (ViTs) and large language models (LLMs) to multi-modal generative AI. Supporting mixed-precision computing, SAKURA-II manages real-time, Batch=1 inference of multi-billion parameter models like Llama 2 and Stable Diffusion—all within an 8W, compact BGA package.
Integration is seamless: SAKURA-II supports both x86 and ARM platforms and is available as a compact M.2 module or a scalable PCIe card. This versatility makes it an ideal drop-in upgrade whether modernizing existing hardware or powering new deployments in embedded, industrial, or enterprise edge environments. The MERA software stack ensures compatibility with all major AI frameworks, offering a seamless transition from model development to real-time inference at the edge. SAKURA-II is already transforming multiple markets.
Read more: talkingiot.io
