MA2485 AI Processor Chip with 4 TOPS DNN Inference 4K Video Processing and Ultra-Low Power Consumption

MA2485 is a Movidius Myriad X series Vision Processing Unit (VPU) designed for edge AI and embedded vision, offering ultra-low power consumption, 4K video processing and DNN inference acceleration.

• Type: Movidius™ Myriad™ X Vision Processing Unit
• Processor Base Frequency: 700MHz
• Maximum Memory Speed: 1600MHz
• Memory Types: 4Gb 1.1V LP-DDR3/LP-DDR4 SDRAM with 32-bit interface, clocked at up to 1600MHz
• Max Operating Temperature: 105°C
• Minimum Operating Temperature: -40°C
• Package Size: 8.8 mm x 8.0 mm x 1 mm
• Package: VF2BGA380

• Core Architecture: Neural Network Engine (NCE): Peak 4 TOPS (FP16), dedicated to DNN inference
• Vector Processor: 16×128-bit VLIW SHAVE, designed for vision and image processing
• Control Core: 2×LEON4 (SPARC V8), responsible for control and scheduling
• Hardware Accelerators (20+): 4K Codec: H.264/H.265 4K@30fps encoding, M/JPEG 4K@60fps, Optical flow, stereo depth, image scaling/rotation, histograms, etc.
• Memory System: Integrated 4Gb (512MB) LPDDR4, POP stack, 32-bit bus width, 1600MHz, bandwidth 6.4GB/s
• Comparison with MA2085: MA2085 has no on-chip memory but supports up to 16GB of external memory; MA2485 is better suited for compact / low-power designs
• Performance and Power Consumption: Typical power consumption: 1.5W (full load), standby <0.5W
• Inference Capabilities: Classification / Detection: Real-time execution of MobileNet, YOLO, SSD, etc.
• Multi-stream Video: Simultaneous processing of 16×1080p@30fps
• Programming Frameworks: Supports OpenVINO, TensorFlow, Caffe, MXNet

• Edge AI Devices: USB accelerator dongle (NCS2), mini-PCIe module, AIoT gateway
• Smart Vision: Security cameras, drones, AR/VR, industrial quality inspection
• Embedded Systems: NAS accelerator cards (e.g. QNAP Mustang-V100, 8×MA2485), embedded industrial PCs