The Mini PCIe format enables the building of small embedded PC systems with KI functionality for deep learning inference applications. The KI functionality is accomplished by two Intel Movidius Myriad X MA2485 Vision Processing Units (VPUs). Using only 5 watts, the Mustang MPCIE-MX2 is suitable for low-power AI applications on the edge. The device is capable of multi-channel operation, allowing each VPU to be assigned a different DL topology performing calculations simultaneously, for example for object recognition and face recognition. Compatible with Intel’s Open Visual Inference Neural Network Optimization (OpenVINO) Toolkit, various AI training models can be integrated at the edge. The OpenVINO Toolkit optimises the performance of the training model and scales it to the target system.
The Mustang MPCIE-MX2 is compatible with Linux OS such as Ubuntu, CentOS and Windows 10. AlexNet, GoogleNet, SqueezeNet and Yolo and many other networks and topologies are supported.