Wuwen Chip Quasar, an AI startup established just over a year ago, recently secured nearly 500 million in Series A financing, bringing its total financing to over 1 billion, making it the king of financing in the AI infrastructure sector. Its shareholders include state-owned assets, local funds, and notable venture capital institutions such as Shunwei Capital and Dacheng Financial Intelligence. One of the founders, Wang Yu, is the head of the Department of Electronics at Tsinghua University, and many team members have Tsinghua backgrounds with rich research experience. In 2016, DeepInsight Technology, incubated by Tsinghua NICS-EFC Laboratory, was acquired by Xilinx for 300 million.

Intel's Gaudi3 AI accelerator chip has found its first cloud computing customer in IBM Cloud, which plans to offer services to customers early next year to support artificial intelligence computing in hybrid and on-premises environments. IBM Cloud will integrate the Gaudi3 chip into its Watsonx AI and data platform to drive the development of new artificial intelligence capabilities. The integration of Gaudi3 with Xeon CPUs will provide cost-effective, secure, and innovative AI computing solutions. Despite facing competition from Nvidia, Intel...

The upcoming iPhone 17 series from Apple is expected to feature 12GB RAM to meet the memory demands of future AI large models. The current iPhone 15 Pro and iPhone 16 series with 8GB RAM are insufficient for supporting large models with around 7 billion parameters. A model with 13 billion parameters requires at least 5GB of memory, and after the operating system and applications take up space, 8GB RAM fails to provide enough room. The 12GB RAM configuration could offer ample space for larger AI models.

NAVIAI 2.0, independently developed by the Zhejiang Humanoid Robot Innovation Center, is a humanoid robot that stands 1.65 meters tall, weighs 60 kilograms, boasts 41 degrees of freedom, and has an AI computing power of 275 Tops. It showcases exceptional performance and a highly anthropomorphic design, with the ability to execute smooth and natural movements. It can perform complex tasks such as giving speeches, making tea, and playing chess, demonstrating a very high level of anthropomorphic motion mapping execution ability. At the exhibition, it not only served as the host but also displayed its ability to quickly learn new movements and perform high-precision tasks in industrial settings.

NVIDIA has open-sourced two new large models, Nemotron-4-Minitron-4B and Nemotron-4-Minitron-8B, which utilize structured pruning and knowledge distillation for efficient training. This significantly reduces training requirements, minimizing data and computing power consumption. Compared to traditional methods, the new models reduce training token data by 40 times and save 1.8 times in computing costs. By optimizing Llama-3.18B, structured pruning simplifies the model structure, while knowledge distillation enhances performance.

As the developer of the Llama foundation language model, Meta points out that the computing power required for future model training will significantly increase, with the expected training amount for Llama 4 being 10 times that of Llama 3. In light of this trend, Meta emphasizes the importance of building training capabilities in advance to avoid falling behind in the competition. Meta has released Llama 3, which has 80 billion parameters, and its upgraded version Llama 3.1, which reaches 405 billion parameters, making it its largest open-source model. To meet future model demands, Meta is

The Hangzhou government has launched measures to support high-quality development across the entire AI industry chain, aiming to establish the city as a national AI industry hub. The initiative focuses on infrastructure development, ecosystem construction, empowering the real economy, industry chain growth, and talent support, proposing 14 specific actions. Key components include supporting computational technology breakthroughs, increasing finan....

At the 2024 China Unicom Partners Conference, Shen Dou, Executive Vice President of Baidu Group and President of Baidu Smart Cloud Business Group, delivered a keynote speech titled "Deeply Embrace Artificial Intelligence+, Accelerate the Development of New Qualitative Productivity." He pointed out that with the rapid development of large model applications, the underlying computing power is increasingly strained, and a multi-core cloud has become an inevitable choice for Chinese enterprises. Bai

Amidst the increasingly fierce global technological competition, China has made remarkable achievements in the field of computing power. The Minister of Industry and Information Technology announced at the 2024 World Artificial Intelligence Conference that our country's computing power scale has soared to the second largest in the world.Image Source Note: The image is generated by AI, and the image is authorized by Midjourney The latest statistics show that as of the end of May, there are alread

Gaoxin Development, a leading computing power company with a market value exceeding 30 billion yuan, has experienced a significant short-term price surge with a static P/E ratio of 156.31 times, clearly above the industry average. The company has disclosed plans for the acquisition of equity in Huakun Zhenyu Intelligent Technology Company. Recently, the stock price has surged, but the audit and assessment of the target assets has not been completed. Data from the dragon and tiger list reveals large capital inflows.

NVIDIA's fourth quarter revenue and profit surged beyond analyst expectations, marking three consecutive quarters of record-breaking performance. Companies, industries, and nations worldwide are experiencing a surge in AI demand. The scale of data center infrastructure is expected to double within five years.

In 2024, large models are fully rising, with various industries scrambling to set up their frameworks. Large models not only enhance productivity but also play a crucial role in the digitalization process. One of the challenges is the insufficient depth of understanding large models have about industries and the credibility of content generated. Insufficient computing power has become a significant constraint on the development of large models. Future trends in computing power development include dedicated computing power, distributed training, energy consumption sustainability, and the combination of software and hardware.

Baichuan Intelligent and Pengcheng Laboratory announced a collaboration to develop the longest window large model based on domestic computing power. This breakthrough addresses the technical limitations of large models utilizing domestic computing, demonstrating the potential to promote local large model open-source initiatives and supporting intelligent transformation.

Kunlun Wanwei's revenue grew by 8% in the third quarter, with R&D expenses increasing to 620 million yuan. The existing computing power is sufficient to meet future demands for 1-2 years. Significant progress has been made in overseas expansion, and the Opera browser has launched the cutting-edge AI feature, Aria, as its AI assistant.

Li Peng, Senior Vice President and President of Carrier BG at Huawei, stated at the Global Broadband Forum that the rise of AI applications, represented by large models, has led to an explosive demand for intelligent computing power. Li believes that in the era of the digital economy, 5G is leading breakthroughs in general internet technologies, reshaping the new paradigm of the digital economy, and accelerating various innovations. The five major trends for the future development of the digital economy are becoming increasingly clear, and by 2025, the demand for intelligent computing power will reach 100 times the current level. To fully unleash this computing potential, more advanced network capabilities and greater bandwidth are required.

By 2025, the scale of computing power will exceed 300 EFLOPS, with intelligent computing power accounting for 35%, achieving balanced development of computing power between the eastern and western regions. The focus will be on intensive construction of intelligent computing centers in western computing power hubs and regions with relatively good foundations for artificial intelligence development. Promote the coordinated development of intelligent computing power with different computing architectures and general computing to meet various business needs. Accelerate the construction of edge computing power and promote the ubiquitous distribution of cloud-edge-end computing power. Strengthen industry-specific computing power construction to meet the digital transformation needs of various industries.

Computing resources may become a crucial dividing line in the competition between Google and OpenAI. Google's computing resources are still rapidly growing, with Gemini's training power reaching 5 times that of GPT-4. Ample computing resources and efficient infrastructure will become Google's advantages in AI research and commercial applications.

The AIGC industry is on the verge of taking off, with the emergence of large models and the widespread application providing a certain acceleration for the reshaping of future industry paradigms. The market size of China's AIGC industry is expected to reach 1,144.1 billion yuan, and a complete model-as-a-service industry ecosystem will be established in the future. Both open-source and closed-source models of large models will collectively create a diverse and thriving ecosystem.

1. A 100 billion parameter model requires 266 8-card A100 servers with a single card computing efficiency of 44%. 2. To improve the performance of large models, it is necessary to optimize aspects such as frameworks, IO, and communication. 3. Compared to GPT-4, domestic large models have discrepancies in computing power, algorithms, and data.