EdgeCortix introduces SAKURA-II edge AI accelerator for genAI applications
EdgeCortix, a Japanese fabless semiconductor company, has unveiled the SAKURA-II edge AI accelerator, specifically designed to deliver high performance and energy efficiency for generative AI applications at the edge.
The platform is built on the company’s proprietary second-generation Dynamic Neural Accelerator (DNA) architecture, with a focus on addressing the complexities of Large Language Models (LLMs), Large Vision Models (LVMs), and multi-modal transformer-based applications.
The DNA architecture is a runtime reconfigurable neural processing engine with interconnects between compute units that can be reconfigured to achieve high parallelism and efficiency. It uses a patented approach to reconfigure data paths between the compute engines.
EdgeCortix claims that the AI accelerator can deliver up to 60 trillion operations per second (TOPS) for 8-bit integer operations and 30 trillion 16-bit floating-point operations per second (TFLOPS).
In terms of memory bandwidth, EdgeCortix says that the platform offers up to four times more DRAM bandwidth compared to its competing AI accelerators. It also provides software-enabled mixed-precision support, allowing it to achieve near FP32 (32-bit floating point) accuracy.
“Whether running traditional AI models or the latest Llama 2/3, Stable-diffusion, Whisper or Vision-transformer models, SAKURA-II provides deployment flexibility at superior performance per watt and cost-efficiency,” says Sakyasingha Dasgupta, chief executive officer and founder of EdgeCortix.
The performance of the SAKURA-II edge AI accelerator can be elevated by utilizing the MERA software suite, a versatile compiler platform that supports a variety of hardware configurations. This software incorporates advanced quantization and model calibration, resulting in reduced model size and improved inference speed without compromising accuracy.
The SAKURA-II can implement an array of models, such as Llama 2/3, Stable Diffusion, Whisper, and Vision Transformer models. Additionally, it includes built-in memory compression, which reduces memory usage and enhances overall efficiency, making it particularly beneficial for generative AI applications.
“We are committed to ensuring we meet our customer’s varied needs and also to securing a technological foundation that remains robust and adaptable within the swiftly evolving AI sector,” Dasgupta adds.
The announcement follows its 2024 forecast, which predicts that efficient edge AI chips will significantly transform processing power and provide tailored functionalities for generative AI and language models.
Read more:
EdgeCortix’s Sakura-I chip selected by BittWare for AI inference solutions
EdgeCortix raises $20 million for technology development and global expansion
Comments