1. Products introduction
At the 21st China International Industrial Fair(CIIF) in 2019, Nodka launched the industrial computer with AI reasoning and accelerated ability, ebox-3000aipc series. These products are developed based on Intel high-performance CPU platform, compact and appearance design, integrating AI reasoning calculation, edge calculation, light source controller and I / O data collection into the same one IPC platform, and are applicable to machine vision industry applications based on deep learning target detection technology.
Intel Movidius Myriad X VPU visual processing unit is the world's first SOC equipped with a dedicated neural network computing engine, which can be used to accelerate deep learning and reasoning, such as UAVs, robots, smart cameras, virtual reality and other products. Myriad x is an integrated hardware module on the chip. It is designed for high-speed, low-power, accurate operation of neural network based on neural network, so that the device can see, understand and respond to the surrounding environment in real time. It can provide 1 TFlops (trillions of times per second) of computing performance, and the overall performance can exceed 4TFlops.
Unique visual and AI workload design Intel Movidius™Myriad™X VPU has 16 VLIW programmable SHAVE cores, a special vision accelerator and 2 CPUs, all connected by an intelligent memory structure. It is a full-featured vision SOC designed for ultra-low power consumption and high performance. Highly parallel sparse data structure design and optimization make myriad Vpu very suitable for deep neural network applications and other modern visual workload.
16 programmable SHAVE cores Intel®Movidius™Myriad™X VPU structure includes a complete set of interfaces, a set of enhanced imaging/visual accelerators, a set of 16 special vector VLIW processors called SHAVE, and an intelligent memory structure that integrates heterogeneous resources for efficient processing. The optimized software library enables device manufacturer to run custom and proprietary operations on 16 high-performance SHAV.
The full name of Intel Open VINO is Open Visual Inference&Neural Network Optimizatio. It is a free software kit for neural network model optimization and reasoning calculation, which is mainly used in computer vision applications. OpenVINO can enhance the function and performance of visual system through the processor (CPU), integrated GPU and deep learning accelerator (FPGA, movidius Vpu) chips based on Intel Architecture.
2. Product details
3. Product value
1.Low Power and high environmental adaptability
Based on low power high performance VPU solutions, support TensorFlow, Caffe ,OpenVino structure, 64bit Ubuntu 16.04 & Windows 10; whole machine fanless, it can be used in waste sorting, vehicle AI equipment and other harsh field environment, improving the applicability of environment requiring waterproof and dust-proof.
There are many limitations in industrial application of using NVIDIA separete graphics card for AI reasoning calculation:
-
The power consumption of hundreds of Watts has ultra-high requirements of industrial computer and on-site power supply;
-
High power consumption, requiring the chassis to provide fan cooling, reducing dust and water resistance;
-
The volume of the separete graphics card impedes the miniaturization of the industrial computer and cannot meet the requirements of various industrial equipment;
-
The life cycle of commercial series graphics card is short. Once the original graphics card is shut down, the equipment needs to be retested and debugged, including driver and system compatibility tests, which will increase product quality and cost;
-
The cost of Geforce series high-performance graphics card is high. The high price of Tesla series and Jetson series limits its wide application in industrial equipment.
All of Intel AI computing hardware uses the unified Intel Open VINO for application development. When switching hardware, basically do not change the program, just change different hardware configuration parameters, which greatly facilitates backward compatibility of industrial devices.
2. High integration and compactness
It integrates AI reasoning calculation, edge calculation, light source controller and I / O data collection, so as to reduce product selection worries for customers, reduce product space occupation, and do not need to worry about product compatibility. It is suitable for machine visionapplications.
4. Product application
In the field of computer vision, two methods are widely used. First one is deep learning, another one is tranditional computer vision. Deep learning has advantages in object detection and object recognition, and it can replace the traditional computer vision. Open VINO includes a deep learning tool with a model optimizer and reasoning engine, as well as an optimized computer vision library for OpenCV * and OpenVx *. Model optimization can optimize the network model developed by the developers based on some open deep learning framework, aiming at the target platform chosen by the developers, and convert the results of these optimization into an intermediate representation file to build an IR file. In the next step, the inference engine will read the IR files, and then use the corresponding hardware plug-ins to download these IR files to the corresponding target platform for execution. So openvino helps developers to reason, not to train.
With Intel Movidius™VPU and Open VINO™ software, Nodka eBOX-3000AiPC products can help industrial developers quickly and efficiently deploy the trained network model to the target platform for reasoning operation.
