Beijing-based Bitmain is widely known for their ASIC bitcoin mining rigs, but has been moving into the AI business with their offshoot Sophon, which specializes in TPU-equipped devices. The company has recently collaborated with 96Boards to launch the Sophon Edge Development Board to bring deep-learning capabilities for any number of different projects, including edge AI applications.
The Bitmain Sophon EDB features both RISC-V and Arm architectures through the onboard BM1880 SoC, which packs a dual-core and single-core RISC-V processor, as well as a TPU capable of pumping-out 1 TOP performance on 8-bit integer operations, along with up to 2 TOPS with Winograd convolution acceleration. According to 96Boards, the BM1880 TPU has tailored support for DNN/CNN/RNN/LSTM operations and models, plus deep learning frameworks, including Caffe, ONNX, Pytorch, and TensorFlow.
As for the rest of the hardware specs, the Bitmain Sophon EDB features 1Gb of LPDDR4, 8Gb off eMMC, a micro SD card slot, 3X USB 3.0 (support for camera, U-disk, etc.), Gigabit Ethernet, Wi-Fi/Bluetooth combo, and a 96Boards 40-pin low-speed GPIO header with 2X I2S for audio. It also sports support for H.264 and MJPEG encoder/decoder (1x 1080p @ 60fps or 2x 1080p @ 30fps H.264 decoder, 75fps for FHD images).
In terms of software, the Bitmain Sophon EDB runs Linux and can take advantage of Linaro 96Boards, Raspberry Pi, and Arduino add-on modules.
Developers can leverage off-the-shelf modules and develop cutting-edge DL/ML applications, like facial detection and recognition, facial expression analysis, object detection and recognition, vehicle license plate recognition, voiceprint recognition, and more!
The Bitmain Sophon EDB is not yet available, but can be reserved for $129 (CNY 899/China) and is expected to ship by November 30th.