Bitcoin ASIC maker pivots to AI tensor ASIC chips

Bitmain announced its first products for accelerating artificial intelligence (AI) applications, applying its industry-leading expertise in Bitcoin mining hardware and system design.

The products are:

* The BM1680, a customized tensor computing ASIC (Application Specific Integrated Circuit), which is optimized for multiple types of inference and training functions for deep learning networks
* The SC1, an advanced fan-cooled module that combines the BM1680 into a compact, easy-to-integrate package

Bitmain’s new line of AI hardware products delivers superior performance and optimized costs when compared to traditional implementations using graphics processing units (GPUs). Bitmain’s Sophon hardware can be applied in a variety of industries and use cases including image and speech recognition, autonomous vehicle technology, enhanced security camera surveillance, robotics, the Internet of Things (IoT), and many other AI applications.

“Deep learning is very intensive computationally and our experience in creating high-performing hardware for Bitcoin has absolutely prepared us for this exciting area of computing,” notes Micree Zhan, Bitmain’s CEO. “AI hardware is an area that Bitmain is proactively developing to power the next generation of AI applications.”

The BM1680 and SC1 feature full compatibility with popular AI platforms, including mainstream Caffe, Darknet, Googlenet, VGG, Resnet, Yolo, Yoto2 and other models, making them simple for developers to use. Bitmain has demonstrated its hardware’s capabilities in motor vehicles, human/object detection, and face recognition and has also completed the training for Alexnet, Googlenet, VGG, Resnet and other mainstream AI networks.

Bitmain was founded in 2013 and released its first chip, a SHA-256 accelerator for bitcoin mining. They made the BM1384 chip as the bitcoin market rose in 2015. Tthey now have more than 600 employees, sell hundreds of thousands of mining systems a year and run their own mining business.