Since the introduction of the M1-SoC with Metal Performance Shaders interface, Apple has shown that the ARM architecture is excellent not only for gaming but also for developing AI applications, paving the way for features like Apple Intelligence. Another Californian chip manufacturer has been using the ARM architecture for its own SoC design for years. Qualcomm aims to extend the success of its mobile SoCs, Snapdragon, which have been used in various versions in mobile phones by Huawei, Xiaomi, and Samsung since 2007, to notebooks.
With the Snapdragon X Elite and Snapdragon X Plus, there are two versions of this CPU type available that can handle typical notebook tasks. The SoCs differ in the number of cores. The Qualcomm Snapdragon X Elite is a 12-core CPU (all high-performance cores in Oryon design) that drives LPDDR5X memory and has an integrated Adreno GPU. The clock frequency is up to 3.8 GHz, with a boost mode on up to two cores at a maximum of 4.3 GHz. The Qualcomm Snapdragon X Plus is the smaller sibling of the Snapdragon X Elite – with ten instead of twelve Oryon cores and a scaled-down Adreno GPU. Windows 10 and 11 run on these SoCs and work best with native ARM applications and drivers.
Machine learning and AI models need to be adapted to the ARM64 architecture of the Snapdragon SoC to execute them efficiently. Nvidia’s CUDA interface is not usable. Qualcomm and Microsoft provide development tools to make this transition as easy as possible. This is meant to encourage developers to use the new acceleration hardware. Model porting is curated through Qualcomm’s AI Hub and Hugging Face. Initial applications like GIMP, Blender, and Cubase already benefit from the new hardware of the Snapdragon X SoCs.
With DirectML, ONNX, and WebNN, developers can create their applications to be more hardware-independent. However, the adaptation to ARM64 is not yet fully mature. Open-source tools such as llama.cpp, the GGUF format, and various quantization types enable fast execution of local models even on the ARM CPU.
To determine how modern machine learning and generative AI models perform on these systems, a test was conducted using an HP OmniBook X with Snapdragon X Elite X1E-78-100-SoC and 16 GB of memory and an Acer Swift 14 AI with Snapdragon X Plus. The results showed promising performance for local AI applications on Snapdragon notebooks.
Overall, the transition to using ARM architecture for AI applications on Snapdragon notebooks is promising. Developers are encouraged to adapt their models to this architecture to take full advantage of the hardware’s capabilities. With the right tools and support, the Snapdragon SoCs could become a strong contender in the notebook market, particularly for AI-driven applications.
In conclusion, Qualcomm’s effort to bring ARM architecture to notebooks is a significant step towards enhancing AI capabilities on portable devices. As more developers adapt their applications to this architecture, users can expect better performance and more efficient AI solutions on their Snapdragon-powered notebooks.