On July 30th, with the theme of "Starting from the Core," the Cixin Technology AI PC Strategy and the launch of its first chip took place in Shanghai. Facing the era of edge-side generative AI that has already arrived, as well as the third PC industry revolution, Cixin Technology has established a development strategy of "one core for multiple uses," taking the lead in focusing on the AI PC field, creating a new generation of AI PC computing power base, and releasing the first heterogeneous high-energy-efficient SoC chip, the Cixin P1.
"Cixin Technology is committed to promoting the development of the edge-side AI ecosystem with high-energy-efficient computing power solutions, and making life, learning, and work more intelligent, efficient, and humanized through the application of AI technology," said Sun Wenjian, the founder and CEO of Cixin Technology at the launch event, expressing his hope to create a better future with AI.
Advertisement
Sun Wenjian, Founder and CEO of Cixin Technology
One Core for Multiple Uses: Building a New Generation of AI PC Computing Power Base
With the release of ChatGPT, artificial intelligence has entered the era of deep learning 2.0, with generative AI becoming the main theme of development. As generative AI large models are deeply deployed and applied in various industries, users' needs in terms of energy efficiency, data privacy, and personalized customization have become more intense, and the mixed deployment of cloud-based public large models and edge-side private large models has become an industry consensus. As the best carrier for edge-side generative AI, the PC industry has put forward new demands for hardware, especially for the SoC as the core computing unit in terms of AI computing power.
Based on this, Cixin Technology has determined the development strategy of "one core for multiple uses," aiming at both the global and domestic markets, building an edge-side AI ecosystem, taking the lead in focusing on the AI PC field, and creating a new generation of AI PC computing power base.
Sun Wenjian said, "The rich functions of Cixin Technology's chips greatly meet the needs of customers in various scenarios; on the other hand, through landing in multiple scenarios, the sales volume of products increases, which spreads the R&D costs of the products, bringing customers a cost-effective product experience."
Regarding the new generation of AI PC computing power base, Sun Wenjian introduced, "The AI PC computing power base of Cixin Technology has the characteristics of being heterogeneous, high-energy-efficient, and secure." It heterogeneously integrates CPU, GPU, and NPU, and is optimized based on data link bandwidth to achieve efficient operation. In terms of energy efficiency, it adopts a multi-core heterogeneous design and a dedicated NPU, as well as low-power memory technology, achieving high energy efficiency through software and hardware collaborative optimization. In terms of security, Cixin P1 has the latest security features in the Arm®v9 architecture, such as PACBTI, MTE, secure EL2, supports common commercial and national encryption algorithms, and meets the needs of TPM (Trusted Platform Module) and TCM (Trusted Cryptographic Module), providing system-level security and privacy protection.
"The new generation of AI PC computing power base of Cixin Technology, facing generative AI, supports the deployment of mixed artificial intelligence," Sun Wenjian further pointed out, by building a rich open ecosystem of software and hardware, empowering developers, and continuously exploring the application scenarios and advantages of edge-side AI PCs.
Launch of the First Heterogeneous AI PC ChipAt the press conference, based on the AI PC strategy, Cixin Technology released its first commercial product, a heterogeneous high-energy-efficient SoC specifically designed for AI PCs—Cixin P1. Sun Wenjian introduced: "After rigorous testing, Cixin P1 fully meets the mass production requirements and will officially enter the productization stage."
Cixin Technology Launches First AI PC Chip
Cixin P1 uses an advanced 6nm manufacturing process, offering features such as rich AI heterogeneous computing resources, a comprehensive security engine, a variety of peripheral interfaces, and support for multiple operating systems. The powerful multimedia engine supports 4K120 frame display, 8K60 frame video decoding, and 8K30 frame video encoding; the high-performance memory subsystem is equipped with a 128-bit LPDDR5 low-power memory, with a capacity of up to 64GB, a data transfer rate of up to 6400Mbps, and a bandwidth of up to 100GB/s. At the same time, it has efficient power consumption management, providing precise dynamic frequency and voltage adjustment, multiple power domains, dynamic power gating, and standard PC power supply working modes.
In the core CPU part, designed with the Arm big.LITTLE® technology, it has 8 performance cores and 4 energy-efficient cores, with a maximum frequency of up to 3.2GHz and a multi-level cache design optimized for PC scenarios; at the same time, it integrates 2 SVE2 vector acceleration units to enhance machine learning instructions.
The integrated GPU provides a 10-core GPU processor, meeting the extreme desktop rendering and general AI computing needs. The new generation of hardware ray tracing, comparable to the game experience of the main machine; the new geometric graphics processing process (Delayed Vertex Shading DVS), achieving more than 40% power saving, and flexible variable rate shading (VRS), achieving more than 50% performance improvement. At the same time, the desktop GPU software stack facing multiple scenarios meets the needs of industry applications.
The powerful heterogeneous AI engine provides 45TOPS edge-side AI heterogeneous computing power, supports the deployment of edge-side large models with up to 10 billion parameters, runs LLM at more than 30 tokens/s, and provides edge-side AI support for multiple scenarios such as computer vision, natural language processing, and generative AI.
James McNiven, Vice President of Product Management at Arm's End Device Division, said: "AI is changing the way consumers use PCs. Cixin P1, developed based on Arm technology, brings the advantages of the Arm platform PC ecosystem to a wider audience, bringing a platform that is both high-performance, high-energy-efficient, and AI-optimized, redefining the user experience of various workloads and use cases."
Xu Yatao, Executive Vice President of Sales and Business at Arm Technology, said: "Cixin Technology is an important partner of Arm Technology. In recent years, the two sides have jointly promoted the iterative upgrade of Arm CPU technology and led the expansion of the advanced Armv9 architecture in the PC application ecosystem. In the future, we will continue to maintain close cooperation in key areas such as product research and development and technical ecology, to provide more reliable and efficient computing power solutions for the industry, and promote the widespread application of terminal-side AI in various fields."
Executive Vice President of Sales and Business at Arm Technology, Xu Yatao
Complete Solution Support Accelerates AI PC Research and Development Process"As a newcomer in the PC chip field, Cixin Technology has the opportunity to create an AI PC platform solution from a new perspective," said Chu Ranzhou, co-founder and vice president of system engineering at Cixin Technology. The "three integrations" strategy of Cixin Technology's platform solution has also emerged from this.
Chu Ranzhou, Co-founder and Vice President of System Engineering at Cixin Technology
The "three integrations" strategy refers to the integration of the advantages of the X86 and Arm architectures, the integration into the PC industry circle, and the integration into the world of AI. Based on the "three integrations" strategy, Cixin Technology has launched the AI PC platform solution of the P1 chip, which has features such as scalable heterogeneous computing, support for multimodal human-computer interaction, high-bandwidth storage, and platform-level security shield.
In addition, Cixin Technology's AI PC platform solution supports a rich interface including Video-In/out, GMAC/Ethernet, HDA/I2S, etc., providing a foundation for the universal benefit of AI; at the same time, based on the "one core for multiple uses" strategy, Cixin P1 will launch multiple specifications to support the landing of various product forms of AI terminals. As one of the cost elements of PC products, the type of PCB is related to the complexity of the SMT process and the yield. The Cixin P1 platform solution can support all types of 8-12 layer, through-hole, high-density board PCB, eliminating the trouble of customers having to carefully select PCB suppliers and SMT OEM factories when making products.
Chu Ranzhou said: "We actively integrate into the PC industry chain, can support PC manufacturers to seamlessly switch from X86 CPU to Cixin P1 chip, and have started to cooperate with mainstream ODM manufacturers and well-known BIOS manufacturers in the industry."
In terms of customer project support, Chu Ranzhou introduced that Cixin Technology has established a customer support system that is "based on the comprehensive linkage of market, support, and R&D teams, responds quickly to customer requirements, implements application scenarios, solves problems quickly, and shortens the product development cycle."
Promote AI PC industry innovation with software
"Hardware is like our body, and software gives a chip a soul." In the AI PC software solution link, Liu Gang, co-founder and vice president of software engineering at Cixin Technology, said.
Liu Gang, Co-founder and Vice President of Software Engineering at Cixin Technology
Faced with the transformation and opportunities brought by the end-side generative AI to the PC industry, Cixin Technology focuses on four major directions of boot firmware, kernel, graphics acceleration, and AI solutions for full-stack software innovation.At the firmware layer, Core Technology has achieved a significant breakthrough by supporting multiple operating systems with a single set of firmware and simultaneously supporting the ACPI and Device Tree specifications with a single Linux kernel. Core P1 has become one of the few products in the world to use a unified firmware to support multiple desktop operating systems.
In order to achieve an ultimate user experience for the Arm GPU on the PC side, Core Technology has independently designed the Core GO graphics engine. By introducing an application compatibility layer and implementing original optimizations at the core driver layer, it adapts to various mainstream desktop environments, is compatible with traditional applications, supports the OpenGL standard, and collaborates with different multimedia frameworks, providing a one-stop solution for industry pain points.
In addition, facing the many issues and challenges in the deployment of generative AI on the edge, Core Technology will launch the NeuralOne AI software stack in the future, providing support for heterogeneous AI accelerators to meet the diversity and complexity of edge inference needs. At the same time, it will provide a unified NeuralOne API to hide specific hardware details and reduce the difficulty of application programming.
Regarding the fragmentation of models and inference frameworks, Core NeuralOne provides a unified SDK to meet the needs of multiple engines and support a wide range of model formats. "Neural means neural network, and One represents unity. Together, we hope that through the Core NeuralOne AI software stack, we can fully leverage the powerful potential of heterogeneous computing power and better help ecosystem partners develop stronger AI applications," said Liu Gang.
Hardware and software collaboration to prosper the AI PC ecosystem
Core Technology is well aware that building an ecosystem is not an overnight task. By laying out a heterogeneous AI edge ecosystem, creating an Arm native development platform, actively participating in upstream open source construction, and promoting industry alliances and standardization, Core Technology hopes to empower developers to build a new PC ecosystem in a way that suits PCs, and prosper the AI PC ecosystem together with developers.
In the roundtable forum, guests such as Zhou Jie, General Manager of Core Technology's Ecosystem Strategy, Xie Wei, General Manager of ARM Technology's Global Service Market Department, Yan Yiqiang, Chief Researcher of Lenovo Group, Zhu Chen, Deputy General Manager of Kylin Software, Liang Yejin, Head of Alibaba Cloud Tongyi Edge Business, and Yao Hang, Co-founder and COO of Qingang Intelligence, discussed the development of AI PC technology, application scenarios, and the opportunities and challenges of industrial development around hardware and software collaboration.
Xie Wei said, "For many years, Arm has been deeply cultivating the ecosystem, not only on the mobile side, but also on the server side and the PC side. The performance of Arm architecture CPUs is already comparable to X86, and it can fully undertake the computing tasks on AI PCs. I hope to see more AI PC chip manufacturers like Core, designing more products to enter the market."
Regarding the impact of hybrid artificial intelligence on the industry, Yan Yiqiang said, "The industrial value brought by hybrid artificial intelligence is very large, and there is a huge market space for related infrastructure, equipment, and the underlying computing power support."Regarding the impact of AI PCs on the domestic PC industry chain, Zhu Chen said, "Generative AI has injected new vitality into the PC industry, bringing new imagination space to the upper-level applications. For the first time, AI chips are integrated into the motherboard, coupled with AI subsystems at the operating system level, enabling all upper-level applications to have AI capabilities, leading to higher quality development of the PC industry."
Liang Ye Jin discussed from the perspective of the collaboration between the inference engine and hardware, stating, "In today's era of a hundred flowers blooming for large models, the computing power on the end should not be wasted but fully utilized. This chip has launched the P1 chip, which has excellent computing power and rich interfaces. Whether the computing power can be used well still requires the collaboration with the design of the upper-level inference engine, which requires the joint efforts of the industry."
As for the matching problem between large models and hardware, Yao Hang said, "At present, there are many large models in China, and the hardware computing units of AI PCs show a trend of diversification. How to solve the matching between large models and hardware, as well as the application landing, cannot be separated from the efforts of the middle-layer partners. It requires all software partners to work together in soft and hard coordination from the model side, framework side, inference engine side, and hardware side to match the models and hardware."
This chip technology AI PC industry chain strategic cooperation launch ceremony
In order to better build the AI PC industry chain, Chen Jiefeng, the vice president of sales of This Chip Technology, invited partners from Lenovo Group, Arm Technology, Tongfang Dingxin, Wanmos, Union Software, Kylin Software, Jiangbo Long, Bai Ao Software, Wuwen Xin Qiong, and other upstream and downstream industry chain partners to jointly come on stage, complete the launch ceremony of the strategic cooperation of This Chip Technology AI PC industry chain, and open the prelude to the cooperation between This Chip Technology products and partners. Accelerate the commercialization of domestic AI PC innovative products, promote the industry to achieve sustainable development, promote the development of new quality production forces, and create a brilliant future for the AI PC industry together.
Comments