2024 and, above all, next year, could decree the true debut of the ARM platform on Windows PCs. Previous attempts were unsuccessful, either because the SoCs presented were not yet up to expectations, or because Windows on ARM could not yet ensure broad application support. Not only the native ones, but also and above all those developed for the x86 architecture.
Microsoft has granted to Qualcomm the rights to exclusively design and build SoCs compatible with the Windows on ARM platform until 2024. With the expiration of the contract, AMD and NVidia have announced that they are working on ARM processors for modern PCs: they will be placed on the market during 2025.
For the moment, however, Qualcomm can enjoy some competitive advantage. The new Snapdragon performers that already today the company led by Cristiano Amon is showing itself to be particularly aggressive, certain that it can truly revolutionize the PC market and make inroads into a sector historically dominated by Intel and AMD. And all this although Pat GelsingerIntel’s number one, states that he does not see a real threat from ARM SoCs in the PC sector on the horizon.
Aggressive Qualcomm: Snapdragon X Elite offer better performance than Intel Core Ultra in AI-related processing
The workloads related toartificial intelligence (AI) are and will put modern computer hardware to the test. For this reason, what Microsoft calls PC AI, a new generation of systems ready to move AI processing from the cloud to the localeare equipped with a NPU (Neural Processing Unit), a specialized unit designed to speed up the computations needed to train and run generative models.
In the short video published by Qualcomm on YouTube, two portable systems can be seen compared: the first based on SoC Snapdragon X Elite; the second on an Intel Core Ultra 7 processor. According to the San Diego (USA)-based company, the Snapdragon workload related to AI. Thanks, first of all, to the Qualcomm NPU 45 TOPS (Tera Operations Per Second).
The tests carried out by Qualcomm involve the generation of images with Stable Diffusion and GIMP using only and exclusively local resources.
The results of the comparison between Snapdragon X Elite and Intel Core Ultra 7
Starting from the same prompt that is, from the same request conveyed to the generative model using the natural language, the Snapdragon X Elite-based laptop completed the image generation task in 7.25 seconds; the Intel Core Ultra 7 system, however, took 22.26 seconds.
Indeed, asking for the repeated generation of multiple images with the same subject (for example a “majestic lion basking in the golden afternoon sun“), Qualcomm found that its Snapdragon a new image every 2 seconds while Intel’s competitor needs 12 seconds to create just one.
According to Qualcomm, the company did not use it tricks and behaved honestly during the test. For example, the engineers who designed the Snapdragon chips first made sure that version 1.5 of the AI-based image generator “worked with the most efficient configuration” for the Intel processor, splitting the processing between CPU Meteor Lake, GPUs and NPUs. At the same time, however, it must be said that the Qualcomm machine appears to have benefited from a specially optimized version of Stable Diffusion.
Qualcomm followed Intel’s guidance to configure the AI plugin for GIMP based on OpenVINO
Qualcomm experts used Stable Diffusion v1.5 on both systems. Furthermore, as you can see in the video, the Intel system uses the graphics editor GIMP with the OpenVINO AI plugin developed and shared on GitHub by Gelsinger’s company. In the case of OpenVINO, obviously, oenology has nothing to do with it: the acronym “VINO” in fact stands for Visual Inference & Neural Network Optimization.
OpenVINO is an open source software toolkit developed by Intel to optimize and implement models deep learning. The tool helps programmers create efficient and scalable AI solutions with just a few lines of code. It supports models from several existing frameworks and covers various categories, including language models, computer vision and generative AI needs.
Qualcomm also specified that it had scrupulously followed Intel instructions to configure and optimize OpenVINO on GIMP.
The challenge has just begun: Arrow Lake and Lunar Lake are coming
The conclusions reached by Qualcomm are, in the end, not as surprising as they might appear. The 45 TOPS of the Snapdragon
Meteor Lake processors represent the first family with integrated NPU from Intel. The company of which Gelsinger is “helmsman” has anticipated that AI performance will grow significantly with the arrival of new generations Arrow Lake (desktop) e Lunar Lake (laptop).
Both arriving by the end of 2024, they could – according to what was declared by Intel spokespersons – triple the performance of applications based on artificial intelligence, precisely through a series of improvements made to NPU and GPU.
The images published in the article are taken from the Qualcomm video.