Abstract:
Qualcomm officially released two new generation flagship mobile phone processors at the annual Snapdragon Summit on September 22, local time, namely Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6. Both chips take AI as the core upgrade direction, focusing on improving the personalization capabilities, device-side reasoning capabilities and multi-modal processing capabilities of mobile phones when running AI agents.
Qualcomm said that the new generation of chips can help AI agents in mobile phones further understand users’ personal habits and usage scenarios. Both processors are equipped with a new perception hub that can locally run lightweight AI models with up to 200 million parameters. With this design, the mobile phone can run the personal recording assistant directly on the device and distinguish the voices of different speakers. At the same time, it can establish a longer-term "memory" based on the user's daily usage, thereby providing more accurate suggestions for automated tasks.
Qualcomm also stated that the new perception center can support complete voice input and voice output AI agents. In other words, the mobile phone does not need to hand over all voice processing tasks to the cloud, but can complete the complete process locally from listening to user instructions, understanding semantics to generating voice replies. This enables mobile AI to be more real-time, while also helping to reduce reliance on network connections.
In terms of AI computing architecture, the 6th generation of Snapdragon 8 Extreme Edition adds a new accelerator unit to run AI models more efficiently. Qualcomm does not simply stack larger computing resources to improve AI performance, but optimizes the model architecture common to current generative AI and agent applications.
The higher-end Extreme Edition can directly run a hybrid expert model (MoE) with 30 billion parameters locally on the mobile phone. Although the hybrid expert model contains 30 billion parameters as a whole, when processing specific tasks, all parameters will not be activated at the same time. Instead, a part of the experts will be selected to participate in the calculation based on the current task. Therefore, the amount of calculation required for actual reasoning can be reduced while maintaining a large model scale.
This capability has also enabled Qualcomm to begin to connect the end-side AI capabilities of flagship mobile phones with the current increasingly large-scale generative AI models. For comparison, Apple announced a hybrid expert model with 20 billion parameters at the Worldwide Developers Conference held in June this year, which Apple calls the most advanced model among its third-generation basic models. Qualcomm has further brought the 30 billion parameter-level MoE model to the mobile phone this time.
In addition to AI, the new generation of Snapdragon chips also enhances mobile phone imaging capabilities. The new CPU can provide more detailed pixel-level camera control for a more professional shooting experience, while improving video anti-shake and motion understanding capabilities.
Among them, the Snapdragon 8 Extreme Edition 6th generation supports up to 8K 60-frame video recording, and also supports 4K 240-frame ultra-high-definition video shooting, which can be used to produce high frame rate slow-motion videos. In addition, the chip also adds Qualcomm's new advanced professional video encoding format APV (Advanced Professional Video) to further meet the needs of professional-level video shooting and post-production.
Audio is also part of this upgrade. Both new chips can use AI to improve vocal effects and reduce environmental noise, while supporting Qualcomm's new "Voice Bubble" speech bubble technology. This technology can separate the user's voice from the surrounding environmental noise during the call, thereby improving the clarity of the call voice.
At the same time that Qualcomm released the new chip, Motorola also announced the launch of the Motorola Signature 27 mobile phone equipped with the 6th generation of Snapdragon 8 Extreme Edition. The machine is expected to be officially launched later this year, which also means that a new generation of Qualcomm flagship chips will soon enter the actual consumer market.
It is worth noting that Qualcomm has been actively promoting the expansion of AI from mobile phones to more hardware devices in recent years. The company has previously revealed that it is developing more than 40 AI devices with partners, covering smart glasses, wearable devices and other new terminals. However, Qualcomm still believes that mobile phones are likely to continue to become an important device for consumers to use AI, and will not necessarily be quickly replaced by specialized AI hardware.
The reason is that mobile phones already have mature cameras, microphones, displays, communication capabilities and a large amount of user data. At the same time, they are devices that consumers carry almost everywhere. Therefore, if an AI agent needs to continuously understand the user's environment, personal habits and context, mobile phones actually have a natural advantage.
Qualcomm's product line this time also reflects the transformation of mobile phone AI from simple generative functions to a more continuously running "intelligent" model. In the past, AI on mobile phones was mainly used for relatively independent tasks such as photo processing, text generation, translation, and voice assistants. However, the new generation platform hopes to enable AI to perceive user behavior for a long time, proactively provide suggestions and even perform tasks based on context.
This also explains why Qualcomm places special emphasis on the new perception center and local memory capabilities this time. For AI agents, it is not enough to have a powerful model. It also needs to continuously obtain information about the user's current environment and previous use of the device. Deploying these capabilities directly onto mobile phone chips can reduce the need for frequent data uploads to the cloud while reducing latency.
Another technology route that Qualcomm is currently promoting is to allow mobile phones to directly run larger-scale local models. The hybrid expert architecture allows the model to activate only some parameters relevant to the current task as the parameter size continues to increase, thereby reducing power consumption and computing pressure. This is especially important for smartphones with limited battery capacity.
Therefore, the upgrade of the Snapdragon 8 Extreme Edition 6th generation series is not just a CPU or GPU performance improvement in the traditional sense, but Qualcomm’s attempt to redesign the way mobile phones serve as AI computing terminals. From local small models and perception centers, to 30 billion parameter-level MoE models, to voice AI agents that can continue to work, AI is gradually becoming one of the core capabilities in the design of flagship mobile phone processors.
As new generations of chips gradually enter the mobile phone market, competition among future flagship mobile phones may increasingly revolve around on-device AI. Whether mobile phones can run larger models, understand more complete personal context, complete complex tasks without relying on the cloud, and whether these capabilities can continue to operate with controllable power consumption will all become important indicators for the next stage of mobile platform competition.
Comments