Qualcomm is adding to its Snapdragon range of devices with a new cloud-based AI processor. The custom-made chip is designed to bring AI processing to the data center and partners such as Microsoft and Facebook are already onboard with the new Cloud AI 100.
Unlike other AI accelerators, which are based on mobile devices and have their own limitations, the Cloud AI 100 doesn’t use Qualcomm’s existing hardware. Instead, it’s an entirely-new purpose-built chip, although Qualcomm has leveraged its knowledge from Snapdragon to build this custom chip.
The custom made chip promises to offer peak performance in excess of 50x compared to the Snapdragon 820. The custom chip is built using a 7nm process. Qualcomm says it should offer 10x the performance of other AI inference solutions on the market, which are based on CPUs, GPUs or FPGAs. The Cloud AI 100 supporsgt the Glow, XLA and ONNX runtimes, as well as frameworks like Caffe2, Keras and Google’s TensorFlow.
Why does all of this matter? With AI very much the future of technology, the requirements we’ll need will put serious pressure on crowd-based platforms. As a result, companies are looking at edge computing to help address that.
Edge computing allows you to offset some of the heavy lifting of the cloud by doing some of the initial processing first. Then, the cloud has less resource-intensive tasks to do, but also means the amount of data being transmitted is significantly lower. With 5G about to deliver low latency and usher in a new era of AI-based solutions – such as mixed reality, gaming and more – the ability to transmit less data with lower latency will be crucial. Other potential benefits of edge-computing based AI could be things like personal assistants and real-time translation, faster image recognition in autonomous cars and more.
Qualcomm’s new processors bring flagship performance and AI to affordable devices
As a consumer, you won’t be looking for Cloud AI 100 support in your next smartphone. However, the benefits of this may be notable depending on the services you use. For example, Facebook uses Qualcomm’s AI for its real-time translations and its messenger chat bots, so the Cloud AI 100 platform could potentially allow Facebook to create new use cases that we’re yet to even dream of. All of this will become more apparent once Qualcomm begins shipping these chipsets to its partners in the second half of 2019.
Article source: https://www.technobuffalo.com/qualcomm-bringing-ai-processing-cloud