DLA based on OpenCL*: low power support unmanned strong calculation

High-power led 1W red light imitation lumens led Sanan 30mil gold wire copper bracket 40

Probe domestic high frequency needle JKT-50L high frequency needle with 3 probes

In the past, people thought that the concept of “unmanned driving” and “machine driver liberating human drivers” was far from satisfactory. However, Intel is letting such an idea fall. In order to drive the data center on the wheel, Intel's current power in the field of unmanned driving is described as "not a thousand miles". The six technologies are "steps" and help others. The key to driving a car is a thousand miles.

do you know? In the future, a driverless car will generate 4TB of data in 90 minutes. This data can help the car understand its environment and allow the car to travel safely on the road. If no in-vehicle artificial intelligence uses machine vision and deep learning to analyze data in real time, it is difficult for a driverless car to get on the road easily. But such artificial intelligence loads typically consume a lot of power and are difficult to operate with limited power in the car. Therefore, if you can use lower power to support more powerful computing, you can use artificial intelligence more effectively to help the car make decisions. Intel's development of OpenCL*-based DLA (Deep Learning Accelerator) can effectively solve this problem, reducing power consumption and overwriting calculations.

Energy efficiency multiplication: optimized partitioning of CPU and FPGA

Artificial intelligence is a hybrid algorithm. It is not just computer vision. From natural language processing to decision-making, there are different requirements for computing. Therefore, only the target can maximize power consumption. At present, Intel has advantages in using CPU for prediction and sequential execution, but at the same time, there is a large amount of parallel execution in the in-vehicle artificial intelligence algorithm. With Intel's acquisition of Altera, the outstanding performance of FPGAs in parallel execution allowed this problem to be solved. Intel has adopted a hybrid architecture in the driverless field, including CPU, FPGA, and future Mobileye-specific vision chips and visual computing, to fully utilize the characteristics of each chip to ensure the calculation effect, thus ensuring labor on an open and mixed platform. The realization of intelligence on unmanned driving, while ensuring the uniformity of performance and power consumption has been engineered.

By optimizing partitioned CPUs and FPGAs, Intel is able to increase energy efficiency by a factor of six, and with very low power consumption—only 23 fps/watt—can perform complex artificial intelligence algorithms in real time.

40 watt vs. 250 watts: Energy efficiency advantages of FPGAs over GPUs

Cars use deep learning in multiple cameras to perform real-time object detection and classification

Driverless cars need to be “eyes” at all times. Multi-camera neural networks use complex deep learning algorithms to detect and classify objects: which are pedestrians? Which is the box? -- These power consumption will be controlled below 40 watts, and by contrast, the GPU will require 250 watts of power to complete this series of actions. So, using the Intel® Arria® 10 FPGA and Intel® Xeon® CPU running on Wind River Linux* demonstrates the obvious power advantages of Intel technology.

If you want to experience the efficiency and convenience of a high degree of driverless driving as soon as possible, you can not ignore the energy consumption in the mass production process. Therefore, Intel allows each part of the hybrid architecture to play its own strengths. This is both technology accumulation and energy reduction. With this key technology, Intel can help unmanned vehicles to go light.