Lenovo increases investment in AI computing power

Author | Liu Baodan

Editor | Zhou Zhiyu

Facing the certainty opportunities brought by AI, Lenovo has put forward the vision of AI for All, with computing power being the most important part of it.

Recently, Lenovo officially released new computing infrastructure products including servers, storage, data networks, and edge computing. This includes the new generation server Lenovo ThinkSystem WR5220 G5, the new NetApp AFF A all-flash series, the disaster recovery integrated solution Rescue 1110, and the Lenovo ThinkSystem 100G core switch.

This release further enriches Lenovo's "One Horizontal and Five Verticals" AI-oriented infrastructure layout. Looking at the development trajectory of Lenovo's computing power, it is evident that Lenovo's computing power is entering a rapid development stage.

In August last year, Lenovo officially announced its AI computing power strategy, aiming to comprehensively build AI-oriented computing infrastructure. In February this year, Lenovo first unveiled the "One Horizontal and Five Verticals" strategy at the Mobile World Congress (MWC) in Barcelona, laying out Lenovo's blueprint for the development of computing infrastructure. "One Horizontal" refers to Lenovo's comprehensive heterogeneous intelligent computing platform, while "Five Verticals" include servers, storage, data networks, software-defined solutions, edge infrastructure products, and solutions.

In April, Lenovo officially launched the comprehensive heterogeneous intelligent computing platform. It integrates Lenovo's five major technological innovations and serves as the infrastructure foundation for large-scale model training and inference in the AI 2.0 era. Chen Zhenkuan, Vice President of Lenovo Group and General Manager of China Infrastructure Business Group, stated that this platform is the core of Lenovo's infrastructure strategic framework in the AI 2.0 era.

With the release of these new products, Lenovo's computing infrastructure will undoubtedly accelerate its implementation.

Taking the new generation server Lenovo ThinkSystem WR5220 G5 as an example, this is a 2U2S server designed to provide computing power for scenarios such as cloud computing, big data, artificial intelligence in large data centers, virtualization, online transactions, high-performance computing, critical business flows, and business collaboration.

For example, the newly released NetApp AFF A all-flash storage series and other products can boost customer IT workloads such as generative AI, virtualization, and enterprise databases, with performance improvements of up to 2 times and verified data availability of 6 nines.

Behind the accelerated layout, Lenovo has seized the market opportunities brought by the iteration of AI technology, especially the booming demand for AI computing power. IDC predicts that by 2026, China's smart computing scale will reach 1271.4 EFLOPS, with a compound annual growth rate of 52.3% over the next five years, while the compound annual growth rate of general computing power scale during the same period is 18.5%.

Like two sides of a coin, behind market opportunities lie challenges. Major tech companies worldwide are heavily investing in AI, and for Lenovo to capture more market share in the AI computing power market, it must continuously enhance its capabilities to cope with intense market competition.

Lenovo is undoubtedly one of the most comprehensive enterprises in terms of AI layout, with its presence in AI intelligent terminals, AI-oriented infrastructure, and AI-native solution services According to the financial report, the company's R&D expenses reached 14.8 billion yuan in 2023, with R&D investment scale approximately 6 times the pre-tax net profit, demonstrating its determination.

As a technology company with a 40-year development history, Lenovo has successfully navigated through multiple technology cycles. Whether the company can recreate its past glory in the AI era, time will tell.

The following is a transcript of the discussion (edited):

Q: The rapid development of AI technology represented by large models has placed what demands on China's computing power infrastructure?

Chen Zhenkuan: Computing power is a core driving force in today's era of artificial intelligence. At the end of 2022, the international emergence of large models like ChatGPT ignited the demand for large language models, leading to rapid development in AI computing power. In the past, computing power was based on general-purpose computing, but large models have ushered in the era of AI 2.0, with an annual growth rate of 10 times in computing power demand.

The entire computing power infrastructure still cannot fully meet the demand for computing power, especially as infrastructure based on CPUs cannot meet the needs of AI computing power development. New acceleration computing infrastructures such as GPUs, TPUs, and others have received significant attention. Over the past year, the share of acceleration computing has increased from around 20% to over 1/3 last year, marking the development of computing power.

Q: How will China's computing power industry develop in the future? How is Lenovo preparing to embrace this trend? What are the strategic layouts?

Chen Zhenkuan: Computing power not only requires servers but also imposes more stringent requirements on high-speed storage and data scheduling. Computing between computing nodes and clusters requires high-speed network connections. To support large-scale training, inference in various scenarios, computing power elasticity is also highly demanded. Additionally, energy efficiency and cooling requirements are also very high.

This year, we first proposed the "One Horizontal, Five Verticals" strategic framework. "One Horizontal" is the Lenovo Wanquan heterogeneous intelligent computing platform, which can help our users more efficiently schedule computing power when using computing infrastructure. The heterogeneous intelligent computing platform can also perform GPU virtualization, support collective communication algorithm libraries, achieve minutes-level power-off training resumption, and unify scheduling between traditional HPC scientific computing computing infrastructure and AI infrastructure across clusters.

The "Five Verticals" include servers, storage, data networks, software and hyper-convergence, edge infrastructure products, and solutions. Lenovo has already had a comprehensive layout in the "Five Verticals." However, we are now repositioning it with AI as the focus, giving it a comprehensive upgrade to make it more capable of AI, fully AI-enabled, and able to accommodate the development of computing power in the AI era.

Q: In April, Lenovo and the China Intelligent Computing Industry Alliance launched the Heterogeneous Intelligent Computing Ecological Industry Alliance. What plans are in store, and what drives the alliance?

Chen Zhenkuan: As a leading enterprise in the computing power industry, Lenovo has a responsibility to drive the development of the entire industry. I believe that alliance members contribute different dimensions of strength at three levels First, at the AI chip level. There are many excellent AI chip companies internationally, as well as many in China. Especially in the past year or two, there has been increasing maturity in inference and advancement in training.

Second, at the system and equipment level. Companies like Lenovo play a role in efficiently applying different chip capabilities for different application scenarios, such as training and inference, contributing to the industry.

Third, at the software and application level. For example, enterprises dealing with large models, including those building enterprise and industry-specific large models.

Chip level, system equipment level, and application software level, these three levels are indispensable. The entire industry ecosystem alliance should leverage the capabilities of all members in different dimensions to set standards together and achieve better integration.

Question: What are the positioning differences between Lenovo Neptune and Lenovo ThinkSystem servers this time? What are their respective characteristics? Compared to the two Neptune servers released at last year's computing power conference, what are the technological upgrades?

Chen Zhenkuan: Today, we mainly released four servers, two Lenovo Neptune servers, and two Lenovo ThinkSystem servers.

The two Lenovo Neptune servers are the next-generation products released last year. The first one is WR5220 G5, which is currently the most widely used product in the market, with higher performance and low power consumption. Compared to last year, we have incorporated more AI management engines, which can reduce unplanned downtime caused by components and also reduce the frequency of component replacements.

The second one is WA5480 G5, a server specially designed for AI, supporting up to 10 double-width PCA GPUs. It currently supports GPUs from multiple brands and manufacturers, not only international but also various domestic GPUs. It is the most suitable server for AI training and inference in AI scenarios.

The other two are Lenovo ThinkSystem servers, one is SD520 V4, a high-density server suitable for high-density deployment and high-performance computing. The other is Lenovo ThinkSystem SR630 V4, a 1U2CPU model mainly used in cloud computing and computing power leasing scenarios.

Question: What is the future market plan for Lenovo's comprehensive heterogeneous intelligent computing platform?

Chen Zhenkuan: "One horizontal" is a comprehensive upgrade of the "five vertical" capabilities, reconfiguring and upgrading the infrastructure layout today to achieve true AI orientation. This is the role played by Lenovo's comprehensive heterogeneous computing platform, which will be ubiquitous.

Question: Lenovo's server business has grown rapidly in recent years, what are the reasons behind this growth? Also, we see many companies in the server business, what are Lenovo's advantages?

Chen Zhenkuan: IDC's "China x86 Server Market Report for the First Quarter of 2024" shows that Lenovo's x86 server sales revenue grew by 200.2% year-on-year, jumping to the third position in the domestic market share This year marks the beginning of the 5th decade for Lenovo. In each decade, Lenovo has gone through various cycles, including the entrepreneurial period, the era of PCs, the era of internationalization, the era of diversification, and more. Our products and services, whether in PCs or infrastructure, are present in over 180 countries and regions worldwide, with the ability to build a comprehensive product capability integrating "end-edge-cloud-network-intelligence". Lenovo's ranking among the top three is the steady progress of its AI-driven infrastructure strategy.

Question: What is the overall development of AI servers in China, and what challenges do they face? This time, a new product of full-stack computing power infrastructure was released. What are the highlights that you can share with us?

Chen Zhenkuan: In 2023, the overall size of the accelerated server market in China reached $9.4 billion, a 104% increase from 2022. Domestic chip performance has significantly improved, and the related industry chain is gradually forming. At the same time, the application scenarios of AI chips are becoming more diversified, and more targeted AI chips are being invested in more precise segmented markets.

Lenovo's new generation of servers are specially crafted for customers with a focus on outstanding computing power, ultimate energy efficiency, and AI intelligent operation and maintenance. Firstly, in terms of performance, the number of processor cores has increased by 2.25 times, AI workload performance has improved by 2 times, and the output performance per rack in cloud server applications has increased by 42%; memory bandwidth has increased by 14%, fully supporting CXL 2.0, and E3.S capacity has doubled. In terms of energy efficiency, the energy consumption per core of the processor has been reduced by 70%, and all models support liquid cooling mode, achieving a data center PUE below 1.1 through 98% coverage of power consumption components. In terms of AI intelligent operation and maintenance, intelligent analysis of key components such as memory and hard drives can be conducted to effectively avoid or reduce the number of component failures, thereby reducing unplanned downtime for customers