Qualcomm AI Stack Arrives To Unleash A New Era Of Intelligence From The Cloud To The Edge
In case you haven't noticed, artificial intelligence in one form or another has emerged as a key cog in all kinds of devices and machines, from smartphones and mixed reality headsets
, to a plethora of IoT devices and automobiles (and everything in between). AI
has even found its way to toasters, a seemingly simple kitchen appliance.
The reason we mention all this is because it's the driving force behind Qualcomm's new AI Stack software tools portfolio, which is essentially an end-to-end AI foundation for developers in a wide range of industries from the data center to intelligent edge devices of all kinds. Or put another way, Qualcomm
invites developers to 'come one, come all in your native AI frameworks and our tools will accept your stuff', to paraphrase the announcement.
"This comprehensive Qualcomm AI Stack portfolio is a truly revolutionary step in improving AI development for OEMs and developers and achieving high performance across our product portfolio," said Ziad Asghar, vice president of product management, Qualcomm Technologies, Inc. "We’re delighted to extend our one technology roadmap with this new portfolio and expand our leadership in the Connected Intelligent Edge."
So what exactly is at play here? Let's have a look...
What this entails is a comprehensive portfolio offering developers optimized access to Qualcomm's AI Engine and dedicated AI cores across its portfolio of silicon solutions. And as part of this new offering, Qualcomm-optimized AI models can now scale across every AI accelerator and a wide range of the company's technologies.
AI Engine direct is an AI library that delegates and deploys existing models directly to the AI
accelerators on Qualcomm Technologies platforms and will now provide OEMs and
developers the ability to develop one feature, then move the same model across different
products and tiers," Qualcomm explains.
Circling back to what we stated earlier, AI is everywhere
these days, but it's not same collection of technologies across the board. Your smartphone, for example, processes speech recognition to turn commands into reminders or to look up directions, while an XR headset has sensors to track your movements. And in the automotive sector, there's a whole different set of sensors, such as LIDAR, radar and other machine vision technologies.
From top to bottom, Qualcomm AI Stack is said to support popular AI frameworks such as TensorFlow, PyTorch, ONNX, and runtimes including TensorFlow Lite, TensorFlow Lite Micro, ONNX runtime. It also includes inferencing SDKs such as Qualcomm's Neural Processing SDK for Android and, more recently, Windows as well.
Part of the proposition here is that a developer can build a app for one type of device or platform, and then have the freedom to more easily move across virtually of Qualcomm's products and tiers using a singular AI-optimized software package. It's a simple concept, but complex in execution—it took Qualcomm several years to develop its AI Stack, with critical cross-collaboration between its software and hardware teams.
"Qualcomm AI Stack can be used directly by OEM and developers to develop on mobile and compute products. It is also a building block for our domain specific SDKs that are available today, for example, the Snapdragon Ride SDK, the Intelligent Multimedia SDK and the Snapdragon Spaces SDK," Qualcomm says.