Two weeks ago we reported on Nvidia's announcement of a series of AI products. It announced the integration of some Nvidia technologies into the ARM chip design, especially to provide a partnership agreement for the optimization of AI processing for IoT devices.
Not to be outdone, Qualcomm announced its Internet of Things and AI optimization system-on-chip this week, focusing on computer vision processing products. The company will launch new QCS603 and SCQ605 chips in this field, and these two products deserve our attention.
Seshu Madhavapeddy, Qualcomm's vice president of product management in the Internet of Things, gave a brief introduction to the two products.
Special application
Madhavapeddy said the new system-on-a-chip is designed for the cutting-edge IoT and can process as much data as possible on the local chip to minimize the amount of data transferred to the infrastructure. At the same time QCS603 and 605 have special equipment for visual intelligence. Special features such as security cameras, motion cameras, wearable cameras, virtual reality cameras, robots and smart displays are all available.
Mr. Madhavepeddy went on to explain the difference between IoT cameras and mobile phones. IoT cameras need to operate in low light environments such as 1 lux. Their stable environment is also different: smartphones only need to take less blurred snapshots, while smart device cameras shoot the correct identifiable video for the operator independently from the moving helmet or flying drone.
603 and 605 can operate in such a low brightness and unstable environment. They can also work with the AI ​​system to handle the obstacle avoidance behavior of the robot.
Device side AI
Qualcomm's AI processor includes his Xiaolong Neural Processing Engine (NPE) framework. The framework is compatible with modules created based on the mainstream deep learning libraries of Tensorflow, Caff, Caff2, Android Neural Networks API and Qualcomm's own Hexagon Neural Network.
These modules can be built into Qualcomm's AI engine via source platform-specific software development kits (SDKs). Qualcomm's AI engine performs optimization reasoning: the image data is evaluated based on the machine learning module on the device. Module training is likely to continue in the cloud with a powerful full GPU.
The 603 and 605 are low-power products designed to operate on battery-powered equipment vendors. They all integrate Wi-Fi, Qualcomm's Adreno GPU, multiple Qualcomm Cryo ARM CPU cores, a Hexagon 685 Vector processor and the AI ​​engine mentioned above.
Free video capabilities and powerful Wi-Fi
Image information no longer means a picture, but a video. The 604 and 605 have monster-level video capabilities. The 605 can simultaneously handle streaming of 4K (ultimate HD) and 1080P (full HD) video information at 60fps. At low frame rates the chip can process more video information at the same time. The 603 can handle 4K and 720P streaming at 30fps at the same time.
In general, QCS603 and 605 are similar products, 603 low power consumption, and various indicators are also lower. For example, 605 has 8 core CPU, 2x2 802. 11ac Wi-Fi, and 603 is 4 core CPU, 1x1 802.11ac Wi-Fi.
High power-level performance on low-power system-on-chips.
Qualcomm's visual intelligence platform provides 2.1 TOPS (operating per second, similar to floating-point FLOS per second, but no floating-point operations) for deep neural network inference. The above-mentioned computing power is provided by a dual 14-bit Spectra 270 image signal processor that can support dual 16 megapixel sensors.
Catch up with the AI ​​boom
Just as the 1980s personal computers proved that high-value computing can be done not only on mainframes. Smart IoT devices and the chips behind them prove to us that AI computing can be done on the device side, not just in the cloud. The purpose of the two changes is the same: as long as the developer cares, the user's equipment is not too big and stupid.
Ready, a series of hardware advancements, including the chip mentioned in this article, will lead the growth of AI's cutting-edge technology. The sooner we can liberate AI from the data center (as its only possible device), the more convenient it is for end users and developers.
The above is about the Internet of Things - Qualcomm released two system-level chips, dedicated to the Internet of Things and visual intelligence related introduction, if you want to know more information, please pay attention to eeworld, eeworld electronic engineering will provide you with more complete, More detailed and updated information.
KENNEDE ELECTRONICS MFG CO.,LTD. , https://www.axavape.com