Microsoft Build: Are you ready for the new era of AI?
It was great to see everyone back in person again at Microsoft Build. The event was packed and a lot of exciting news came out of it from Qualcomm Technologies and our partners. The following are key highlights, starting with Qualcomm AI Stack.
An open door to AI opportunity
As an application (app) developer or a device manufacturer, it's crucial to identify your main objectives. Are you aiming for broad compatibility across various platforms, enhancing performance, or adding features?
If you're developing an artificial intelligence (AI) application for devices with Snapdragon and performance is a priority, you have two primary options. You can choose between the Qualcomm Neural Processing SDK or the Qualcomm AI Engine Direct SDK. The choice depends on your specific needs and the goals of your application.
Qualcomm AI Engine Direct SDK is now accessible to any developer interested in utilizing our Qualcomm AI Engine and its accelerator cores (central processing unit (CPU), graphics processing unit and Qualcomm Hexagon processor). What was once exclusive to a select group of customers is now open to any developer who wishes to take advantage of it.
The all-in-one choice: Qualcomm Neural Processing SDK
The Qualcomm Neural Processing SDK supports heterogenous computing and system-level configurations. Additionally, it is designed to direct AI workloads to all accelerator cores on our platforms. This SDK offers developers a high degree of flexibility, including support for inter-core collaboration and access to a variety of advanced features.
The Qualcomm Neural Processing SDK provides dedicated application programming interfaces (APIs) for power management for assisting the developer in fine-tuning their core frequencies and modifying parameters, ensuring ideal performance of their AI models. Additionally, our Qualcomm Neural Processing SDK offers CPU that allows for the task to be complete even when Hexagon is not available.
This SDK is particularly beneficial if you’re a developer looking to enhance the speed and efficiency of your AI models using the Qualcomm AI Engine. The best part is you don't need to focus on specific accelerator cores. Instead, the SDK handles the complexities, allowing you to concentrate on building and refining your AI models. This makes the Qualcomm Neural Processing SDK a nearly perfect choice if you’re aiming for high performance and efficiency in your AI applications but have no preference in accelerator core choice.
For control and customization: Qualcomm AI Engine Direct SDK
The Qualcomm AI Engine Direct SDK offers a unique degree of control and customization compared to the Qualcomm Neural Processing SDK, providing lower-level, highly customizable unified APIs that facilitate faster AI models on all AI accelerator cores. Using the Qualcomm AI Engine Direct SDK, you can directly manage and fine-tune various aspects of your AI models, such as setting up model frequencies, adjusting parameters, and specifying the specific accelerator IP core you wish to target.
It can also delegate workloads from popular runtimes including Qualcomm Neural Processing SDK, TensorFlow Lite and ONNX runtime. This heightened level of control empowers you to actively optimize your AI models to align with specific needs and objectives, allowing for a more tailored performance.
If scalability is your goal and you want your model to function on all platforms, including those beyond Snapdragon technology, you might lean toward a popular runtime like Tensorflow Lite with our pre-built TFlite delegate.
The key difference between our SDKs
An important distinction between the Qualcomm AI Engine Direct SDK and the Qualcomm Neural Processing SDK is the level of control and customization offered by the former. The Qualcomm AI Engine Direct SDK does not include an automatic workload allocation feature to different IP cores, unlike the Qualcomm Neural Processing SDK. This characteristic of the Qualcomm AI Engine Direct SDK emphasizes its capacity for fine-tuned control and customization, but it also requires your careful planning and management of AI models.
Qualcomm AI Stack now downloadable
Lastly and perhaps more importantly, we have made our Qualcomm AI Stack fully downloadable within our Qualcomm Package Manager. Developers across the board can now access our latest AI software tools.
More news to come
We are always working on new features and enhancements to make our software tools more powerful and easy to use. Stay tuned for more announcements in the next blog post. Thank you for using Qualcomm AI Stack and happy coding.
