• 0
Interview with Intel Dai Jinquan: Help China's AI model efficient and sustainable development
统计 阅读时间大约211分钟

2024-02-21 04:19:49Interview with Intel Dai Jinquan: Help China's AI model efficient and sustainable development

来源:Fast information website
 Although AI is not a new technology. For many years, enterprises have been investing in l 特快资讯

  Although AI is not a new technology. For many years, enterprises have been investing in large amounts of artificial intelligence (AI), but as GPT-3.5 and GPT-4 have appeared one after another, the new era of AI is coming.After nearly a year of "Hundred Model War", the industry is going in the second half.

  The latest forecast of IDC shows that by 2027, global AI solutions will increase to more than $ 500 billion.At the same time, most enterprises will also experience the significant transfer of technical institutional rights to artificial intelligence implementation and artificial intelligence enhanced products/service applications.This means that AI technology is about to usher in a new round of growth, and it will also reshape thousands of industries.

  As the world's leading chip and infrastructure solution company, Intel is undoubtedly the vane of the industry and the AI pioneer of the industry.The company is constantly exploring the ability to add AI in the hardware and promote the popularization of AI applications through open and multi -architectural software solutions.

  At the 2023 Intel ON Technology Innovation Conference, Pat Gelsinger, CEO of Intel, pointed out that "AI represents the arrival of the new era. AI is giving birth to a new era of global growth. In the new era, composition powerPlay a more important role and let everyone usher in a better future. For developers, this will bring huge social and commercial opportunities to create more possibilities and create solutions for major challenges in the world to create solutions in the world.And benefit everyone on the earth. "At that time, the concept and products of the AI PC were proposed, and AI also opened another new track.

  Subsequently, at the Q3 financial report analysis meeting in 2023, Pat also said, "We are committed to allowing artificial intelligence to be widely used in various fields. We believe that the work load of artificial intelligence by 2030 will become a semiconductor that promotes $ 1 trillion in semiconductors.The key factor in the total market. We are trying to enable the market to seamlessly integrate artificial intelligence into all applications and operate efficiently. "

  The time is back to now, the end of the year is approaching, what is the situation of Intel's layout in AI, what unique understanding of AI, and how will it realize their vision in the future?We invited Academician Intel and Data Technology Global CTO to share it.

  Question: Intel has various AI computing devices such as CPU, GPU, GAUDI, FPGA, etc. In addition to the combination of components for heterogeneous calculations in the future, it will gradually integrate it with plate -level heterogeneous.The body, and connect these heterogeneous body together to give full play to the real ability of AI?

  Answer: Intel built the super -heterogeneous architecture of XPU, and one of the key technologies is the open -platform programming framework open by OneAPI, which can help developers on the heterogeneous hardware platform to play its optimal computing power.

  Q: In the new era of AI PC, what will happen to our lives?For ordinary people, the probability of price is the primary consideration. How long will we have a distance from AI, how can Intel cooperate with OEM manufacturers from the underlying basic software and hardware to make AI all touch it?

  Answer: Our vision is to promote everywhere. AI PC is an important part of it; based on the Intel Core Ultra processor, a platform that provides a combination of CPU, GPU, and NPU, bringing a new AI experience to the application ecosystem of the X86.Including the local AIGC application experience, thereby promoting the improvement of creativity and productivity.

  Question: What are the differences between the computing power requirements of AI training and reasoning. How does Intel solve different computing power needs through products?

  Answer: AI training includes pre -training (usually the scale of data sets is relatively large, requires a large amount of computing resources, the computing complexity is high), and fine -tuning (the size of the data set and the relatively small computing requirements);High requirements.

  Intel's AI acceleration from the GPU and AI accelerator from the data center, to the general CPU AI acceleration (such as the accelerator Intel AMX in the matrix operation of the fourth -generation Intel most expansion server), and the rich AI hardware product combination and optimized softwareCombined with different AI computing scenarios.

  Question: Now, the industry's understanding of AI hardware is still more inclined to equal with the GPU. Scientists have pointed out that in the future, AI computing will gradually gradually build a new ecosystem from universal to dedicated.The composition of hardware?

  Answer: Intel from the GPU and AI accelerator of the data center to the general CPU server (such as the accelerator specifically for matrix operations -Intel AMX), to the client on the clientPC, provides ubiquitous calculations to drive AI omnipipipipipip then every time.

  Question: Now the development of the AI industry is divided into large refinery models, that is, walking the universal model route, and the other is a specific model for the industry or scene. How Intel looks at the future AI technology development route, what does "smart emerge" mean for us?How will the future market develop?

  Answer: Now in the development of the AI industry, the large model of general models provides a high -quality basic base, while modeling and deployment for models for vertical fields provide more relevant capabilities on the basis of general models.Better to the ground.

  Question: For AI, landing is the most critical issue. The demand for AI in various industries has different needs. Each industry also has its own data. For different industries, how Intel can help AI go through the underlying basic software and hardware to walk AI to walk away.Enter the scene?

  Answer: The most important point of computing industry is how to provide computing power to support the application and algorithm on the upper levels.In this regard, Intel proposed "software definition and chip enhancement", that is, users use software to define what kind of computing power they need, and then how to provide better support from the perspective of hardware; Intel uses OneAPI's open standards, open source, open sourceThe optimization of the AI software framework, etc., helps AI landing.

  Question: Computing power, profile, and capacity are the three major capabilities that are currently in urgent AI. How can Intel unite one to embrace the future intelligent world?

  Answer: Intel is always committed to promoting faster transmission, more storage, and wider calculation.Therefore, we not only add AI capabilities in the products of the personal computing side and the data center, but also solve the problems encountered in practice with software and hardware collaborative optimization.Comprehensive development.

  Question: In the first half of this year, China staged a "Hundred Model Wars". Now, big domestic models are relatively mixed. Intel believes that the development of China's AI industry needs to break through?

1、Fast information website原创文章未经授权转载必究,如需转载请联系官方微信号进行授权。
2、转载时须在文章头部明确注明出处、保留官方微信、作者和原文超链接。如转自Fast information website
3、Fast information website报道中所涉及的融资金额均由创业公司提供,仅供参考,Fast information website不对真实性背书。