(Report produced by Author: Guohai**, Chen Mengzhu, Yin Rui, Lu Ruiqi).
As an end-side, AI PC phones can solve problems such as data security and latency compared with the cloud
As an on-device large-scale model solution, AI PC phones can help solve problems such as data security and privacy leakage. For individual users, a digital twin can be formed by accessing local data, and the large model becomes a personalized assistant for the userFor enterprise users, the enterprise database can be accessed through the company's intranet to achieve intelligent collaborative office. This means that in order to obtain industry models, enterprises must upload key assets - data to the cloud, and once the trained model is shared by the whole industry, it means that to a certain extent, the competitive barriers of enterprises are eliminated, which gives rise to the demand for enterprises to train special models in the data wall, and AI PC mobile phones came into being. In most application scenarios, users have high requirements for the timeliness of inference, and inference in the cloud and then transmit the results back to the end and edge will inevitably have delays due to physical distance.
As a decentralized service model, AI PC mobile phones can help the cloud share the cost of computing power
Deploying large models on the device side and at the edge may lead to market demand for public clouds, private clouds, and on-premises data centers, as well as a balance of computing power across the cloud, devices, and edge. In the past, the market's attention to computing power was mostly focused on cloud service providers and large model training stages, but the cloud computing power was relatively limited, and it was difficult to meet the inference needs of all users during peak periods. Sharing computing power costs for the cloud through smart terminals (such as AI PCs, mobile phones) and edge devices may balance the demand for computing power for the cloud, devices, and edges.
Measurement of AI PC shipments and incremental revenue space based on penetration rate
Based on our analysis of AI PC shipments based on penetration, Canalys expects PC shipments to reach 26.7 billion units, we refer to this data to assume that the PC shipment intake scenario, which is 24/2.6/2.8/3.0 billion units;Similarly, the AI PC penetration scenario is hypothetically 5%, 10%, 20%, 30%, 40%, 50%, and 60% under seven different scenarios. According to the calculation, the range of AI PC shipments is 012~1.8 billion units. Assuming the price increase level of AI PC, the incremental market space under the two different scenarios of AI PC price increase of 1,000 yuan and 1,500 yuan is 12,018 billion yuan and 18,027 billion yuan, respectively.
GPU History: Born out of the need for specialization, it performs parallel computing tasks as a secondary processor
As the process evolves, GPUs have become more popular on personal PCs and mobile devices, and even with the exponential growth in performance, the momentum for independent graphics has gradually increased.
NPU: Born from the needs of neural network computing, compared with GPUs, it takes into account both specificity and energy consumption
The NPU (Neural Network Processing Unit) can simulate human neurons and synapses at the circuit layer, and use the deep learning instruction set to directly process large-scale neurons and synapses, and complete the processing of a group of neurons with one instruction. Compared with the von Neumann structure of CPUs and GPUs, NPU integrates storage and computing through synaptic weighting, thereby improving operational efficiency. NPU is a type of ASIC chip, which is currently mainly used in artificial intelligence tasks such as deep learning and machine learning.
The Snapdragon 8 Gen3 speeds up the running of large models on mobile phones, and the Snapdragon X Elite helps Qualcomm enter the AI PC track
Qualcomm unveiled the Snapdragon 8 Gen3 for AI phones and the Snapdragon X Elite for AI PCs at the 2023 Snapdragon Technology Summit, taking the lead in becoming the chip manufacturer that simultaneously seized the two end-side tracks for generative AI applications. As far as the mobile side is concerned, the phones equipped with the Snapdragon 8 Gen3 run the stable Diffusion model with only 0An image is generated locally in 6 seconds, which greatly optimizes the inference speed of generative AI on mobile devices compared to the 15 seconds of the Snapdragon 8 Gen2On the PC side, Qualcomm pioneered the Snapdragon X Elite platform, which not only expands the breadth of generative AI applications, but also serves as an important step for Qualcomm to enter the PC market.
PC equipped with NPU has become an important symbol of the opening of the AI PC era, and the iteration of the Windows system may bring about a wave of replacement
As a special engine for AI-related tasks, NPU has become one of the core components of the PC, opening up the possibility of AIGC landing on the device side, and is an important innovation of PC products in the AI era. NPUs have a better energy consumption ratio than GPUs, which means that as NPUs become more widespread on PCs, the division of labor with GPUs may become clearer. The GPU, which was first used as a dedicated chip for image processing, is also commonly used in deep learning-related fields, and the parallel computing capability of matrix multiplication makes GPU an important hardware unit that AI model training and inference rely on before the NPU. As Transformer has become the mainstream architecture of LLM and unified the two research branches of CV and NLP, neural network has become the basic unit for building AI models, and NPU as a tensor processor can complete large-scale neural network computing tasks in less time, so it is a more suitable acceleration engine for AI-related tasks than GPU. As one of the important infrastructures on the device side, NPU will provide the prerequisites for PC to carry large models, and its popularization marks the advent of the AI PC era.
In order to support the operation of high-spec large models on the device side, the heat dissipation module of the AI PC may be refactored
Heat dissipation capability is a major bottleneck for high-computing power chips, and with the continuous improvement of model specifications mounted on AI PCs and AI phones, NPU performance release may become more aggressive. From the perspective of data centers, the key pain points are power consumption and heat dissipation, and how to accurately calculate and smoothly discharge the calorific value and thermal effects generated by high integration and high energy density within the limited volume of the chip has become the most challenging problem in the 3D packaging of multi-source heterogeneous chips. At present, in terms of consumer electronics terminals and mobile phones, stainless steel VC vapor chambers have gradually replaced copper VC vapor chambers to become the main force of heat dissipationThe heat dissipation scheme of a PC is generally presented in the form of a combination, and the thermal module mainly includes a heat dissipation base, heat pipes, heat fins, cooling fans and other objects from a structural point of view.
The large model is implemented on the device side or makes AI PC mobile phone manufacturers become the traffic entrance in this service mode
The main beneficiaries of the implementation of large models on the device side include hardware (mainly NPU) design manufacturers, large model providers, and terminal equipment (AI PC mobile phone) manufacturers, among which terminal equipment manufacturers, as the only direct role for C-end users, may become the traffic entrance in this service model.
Lenovo Group: Pioneered the launch of the first AI PC concept machine, which is expected to be launched in the second half of 2024
In October 2023, Lenovo Group showcased its artificial intelligence personal computer (AI PC) product at the 9th Lenovo Innovation and Technology Conference, which is expected to be launched in the second half of 2024. According to Lenovo's official website, its AI PC is able to create a personalized local knowledge base and run individual large models through model compression technology to achieve natural interaction between users and artificial intelligence.
Lenovo Group: Launched AI Phone and AI Twin, a functional product that empowers both individual and business users
As far as individual users are concerned, AI Twin is defined as an extension of the user in the digital world, which is essentially a proprietary large model generated by an individual based on a local knowledge base, which can understand the user's way of thinking and give solutions to the greatest extent. Since AI Twin will only be stored on localized devices or home servers, and personal data will not be shared or uploaded to the public cloud platform, users' personal privacy and data security can be effectively protectedEnterprise-grade AI Twin covers a range of enterprise-level AI applications, which can connect various intelligent devices, edges, and private clouds within the enterprise, link various enterprise-level software, synthesize and analyze all kinds of information, and give optimal recommendations. For example, an enterprise-level AI twin can take into account the company's travel policies, approval processes, and employees' personal information and preferences to specify a reasonable plan for users.
Xiaomi: It plans to connect lightweight local large models to terminals and reach in-depth cooperation with WPS AI
In April 2023, Xiaomi set up an AI large model team and tested the model with 1.3 billion parameters on the mobile phone, and at the same time, Xiaomi's voice assistant "Xiao Ai" began internal testing of the large model version, becoming the first application of Xiaomi's large model. In August 2023, Xiaomi's latest 1.3 billion parameter model has successfully run through the mobile phone, and some scenarios can be comparable to the results of the 6 billion parameter model running in the cloudIn addition, Xiao Ai's monthly activity has exceeded 1100 million, and upgrade the capabilities of AI large models, and open invitation tests. In October 2023, WPS AI was unveiled at the new product launch event of the Xiaomi 14 series mobile phones, and users of the Xiaomi 14 series mobile phones can use the exclusive version of WPS with the WPS AI function through the Xiaomi community**. On the Xiaomi 14 series mobile phones, WPSAI supports input topics to generate PPT presentations with one click, and can provide the function of further refinement and adjustment. In October 2023, the Xiaomi 14 series debuted the Snapdragon 8 Gen3.
Humane: Introducing a new wearable AI PIN designed to interact with large language models
On November 9, 2023, Humane officially launched AI PIN, a wearable device powered by OpenAI designed to interact with large language models. The device, which allows users to make calls**, send text messages, and search for information by speaking, also boasts a laser display that turns directly into a mini screen in the palm of the hand, starting at $699.
This article is for informational purposes only and does not represent any investment advice from us. To use the information, please refer to the original report. ）
Featured Report**: Think Tank for the Future].