Deutsch
Español
Français
Italiano
Nederlands
Polski
Português
Русский
Türkçe
Svenska
Chinese
MagyarMWC 2024 | AI Everywhere: Qualcomm shows impressive on-device demos on Snapdragon X Elite PC, phones and IoT
Since OpenAI first presented ChatGPT to the astonished world public in 2022, the topic of generative AI has not disappeared from the headlines, whether in stock market letters or the tech sector. The abbreviation for artificial intelligence was also omnipresent at the Mobile World Congress 2022. However, it is important to distinguish where artificial intelligence has its home. You can see this in Samsung's new Galaxy S24 "AI phone" with many new Galaxy Ai features: while some of them run directly on the device and also work offline, many require access to the cloud.
"On-device" is therefore an important factor, for example for data protection reasons, but of course also due to offline capability and performance, regardless of the availability of the respective cloud provider. At the MWC 2024 in Barcelona, which starts today, Qualcomm demonstrated today's technical possibilities of AI algorithms running directly on the device or what will soon be possible, at the latest in the summer, when the first Snapdragon X Elite-based AI PCs come onto the market.
Snapdragon X Elite NPU against Intel Core Ultra 7 155H
In the demo video above, an image is generated using Stable Diffusion 1.5 in the open source editor Gimp with the help of generative AI: On the left is 45 TOPs Hexagon NPU in the Snapdragon X Elite demo device, and on the right is Intel laptop with Core Ultra 7 155H. The red Qualcomm test device, already known from the Qualcomm Summit in October 2023, completed the image generation in 7.25 seconds, three times faster than the Intel PC next to it, which took 22.26 seconds.
The second test is almost even more impressive: While the Snapdragon X Elite laptop spits out an image every second, Intel's image generation takes an average of 12 seconds, even though Meteor Lake also has an integrated NPU. Of course, it should be noted that we do not know the specific parameters of this test, but at least in Qualcomm's world, the Snapdragon X Elite notebooks expected in 2024 have tangible advantages, especially with on-device AI.
Individual image adjustment using LoRA on an Android phone
The demo above is also exciting, showing for the first time how generative AI can customize the result according to user requirements using "Low Rank Adaptation" or LoRA for short. Whether cats with crystals or noodles instead of classic fur - there are virtually no limits to the imagination. All of this runs directly on an Android phone with a Snapdragon chip.
First multimodal 7+ billion parameter model locally on a smartphone ...
Another first is the LMM (Large Multimodal Model) with more than 7 billion parameters shown by the Qualcomm AI research team, which runs locally on an Android phone. The video above shows how artificial intelligence accepts not only text but also speech and images as input and uses them to hold a conversation with the user. The whole thing is called LLAVA or Large Language and Vision Assistant by Qualcomm.
... and of course also on a Snapdragon X Elite PC
In the last demo above, a 7+ billion parameter LMM is also running on a Windows-on-ARM PC, naturally based on the Snapdragon X Elite, which also accepts audio files as input and can analyze them - in this case music, for example. Qualcomm calls this Audio Reasoning. In addition to these demos, Qualcomm also announced the AI Hub at MWC 2024, which app developers can use to make their applications AI-ready in no time. Qualcomm also has new hardware in the field of communication. Find out more about the Snapdragon X80 5G modem and the FastConnect 7900, the first chip to combine Wi-Fi 7, Bluetooth 5.4, and Ultrawideband, here.
AI in IoT devices and current smartphones
Some consumer IoT devices based on Snapdragon are also already using AI features, a prominent example being the controversial AI Pin from Humane. The flagship smartphones launched at the weekend - Xiaomi 14, Xiaomi 14 Ultra, and Honor Magic6 Pro - each provide different examples of how on-device AI can be used to set different priorities. In the case of Xiaomi, the main area of concern is the camera, while Honor focuses more on user operation rather than tasks.
Source(s)
Qualcomm
