Brief specifications:

Product Category: Reasoning Module
• Size: 82mm x 61mm x 7mm
• AI processor: Ascend 310 series
• 1 DaVinciV300 AI core, with a main frequency of 1.224GHz
• 4 TAISHANV200M processor cores, with a clock speed of 1.6GHz
• AI computing power: Half precision (FP16): 10TFLOPS
• Integer precision (INT8): 20TOPS
• Memory: Supports ECC/memory 12GB/bit width 96bit
• Encoding and decoding capability: -Supports H.264/H.265 Decoder hardware decoding, 40 channels 1080P (1920 x 1080) 30FPS，YUV420 -Supports H.264/H.265 Decoder hardware decoding, 4-channel 4K (3840 x 2160) 75FPS, YUV420 -Supports H.264/H.265 Encoder hardware encoding, 20 channels 1080P (1920 x 1080) 30FPS，YUV420 -Supports H.264/H.265 Encoder hardware encoding, 3-channel 4K (3840 x 2160) 50FPS, YUV420 -JPEG decoding capability 1080P (1920 x 1080) 512FPS, encoding capability 1080P (1920 x 1080) 256FPS， Maximum resolution: 16384x16384， Minimum resolution: 32x32
• Electrical Characteristics: Operating Voltage 12V/Typical Power Consumption 21W
• Supporting OS systems: openEuler 22.03, Ubuntu 22.04

Introduction to functions and application scenarios:

This inference module is a high-performance, low-power AI chip designed by Huawei specifically for fields such as image recognition, video processing, inference computing, and machine learning. The Ascend 310 series AI processor has one built-in AI Core and can support 96 bit wide LPDDR4X. The module can achieve a maximum computing power of 20 TOPS INT8. This inference module has rich external interfaces, supporting USB 3.0/USB 2.0, PCIe 3.0, RGMII, I2C, SPI, I2S, MIPI-CSI, MIPI-DSI, HDMI, SDIO, CAN and other interfaces. The module can achieve up to 40 channels of 1080P 30fps video access; There are 4 TAISHANV200M processor cores inside the module, with a maximum clock speed of 1.6GHz, suitable for different video processing needs of users, and can be used as embedded system CPUs. The inference module adopts a compact structural design, and the interface adopts MXM gold finger design. Different heights of MXM connectors are selected to achieve different configurations, which can be flexibly laid out in user products. It can achieve various data analysis and inference calculations such as images and videos, and can be widely used in scenarios such as robots, drones, intelligent monitoring, video servers, and assisted driving.