Gartner: Generative AI will drive the transformation of Chinese enterprise data center design

According to news on April 15, 2024, a 2024 CIO and technology executive survey recently released by Gartner showed that more than 60% of Chinese companies plan to Deploy generative artificial intelligence (GenAI) within 24 months. Since Chinese companies tend to deploy GenAI locally rather than through the public cloud, the current infrastructure environment cannot support GenAI projects. This will promote the design transformation of Chinese enterprise data centers.

Zhang Lukeng, research director at Gartner, said: “Due to security and data privacy concerns and regulatory requirements, some enterprises prefer to deploy GenAl solutions or fine-tune large language models (LLM) on-premises. Deploying on-premises GenAl is not just a simple hosting requirement for data centers, but may change the strategy of enterprise data centers because model training requires large-scale GPU clusters."

Gartner defines five types of GenAI deployments. method (see Figure 1). Depending on the GenAI deployment method chosen by the enterprise, China's chief information officers (CIOs) and infrastructure and operations (I&O) leaders need to understand the impact of GenAI deployment and how to take action.

Figure 1: Five methods for generative artificial intelligence deployment

Gartner: Generative AI will drive Chinese enterprise data centers Design Transformation

China’s CIOs and I&O leaders must prepare for the impact of this technology on the data center.

Deploying GenAI on-premises will force I&O leaders to change how they design hosting environments

The impact of deploying GenAI on the data center determines the types of workloads run, as some GenAI workloads require High-end graphics processing unit (GPU). Due to the limited supply of high-end GPUs in the Chinese market, I&O leaders who want to deploy GenAI on-premises will need to change the way their hosting environments are designed.

I&O leaders cannot solve supply shortages alone and must work with the business, AI engineers, and functional teams to address this challenge.

Zhang Yingling said: “Chinese CIOs and I&O leaders who prepare infrastructure resources for GenAl deployment should proactively work with business and related teams to predict the impact of different workloads on data center costs and schedules. , to develop a data center macro strategy for GenAl deployment. If training models require high-end GPU clusters, you need to fully understand various hosting options by balancing costs, risks, and opportunities (such as purchasing alternative hardware or leasing GPU resources).”

Deploying large-scale GPU clusters requires transforming and upgrading data center infrastructure and equipment

Building a basic model from scratch or fine-tuning a model requires deploying large-scale GPU clusters, which will bring consequences to existing data centers subversion. Because the training of GenAI models requires high throughput, low latency, and lossless infrastructure. To support such high-performance computing clusters, network, storage, power supply, and cooling systems must be upgraded. In some cases, existing facilities will need to be retrofitted to host the upgraded infrastructure (see Figure 2).

Figure 2: The impact of large-scale GPU clusters on data centers

Gartner: Generative AI will drive Chinese enterprise data centers Design Transformation

Zhang Yingling said: "China's CIO and I&O leaders need to work with data scientists and engineers to clarify the GPU cluster size and GenAI performance requirements to determine infrastructure such as network and storage. Requirements. At the same time, it is also necessary to analyze power requirements, cooling efficiency, racks, space, etc., to determine the gaps in the existing data center environment in deploying large GPU clusters, and to select the most suitable data center transformation solution.

The above is the detailed content of Gartner: Generative AI will drive the transformation of Chinese enterprise data center design. For more information, please follow other related articles on the PHP Chinese website!