Intel, AMD Discuss How AI Will Test and Revolutionize Data Centers

During an opening keynote at Data Center World 2024, execs from Intel and AMD detailed the massive shift AI is bringing to the data center – and their visions for where the industry is heading.

Drew Robb

April 17, 2024

5 Min Read
Intel, AMD Discuss How AI Will Test and Revolutionize Data Centers
Left to right: Data Center World's Bill Kleyman, Intel’s Jen Huffstetler and AMD’s Laura Smith discuss how AI is revolutionizing the data center at Data Center World 2024, April 16.Channel Futures

The opening keynote at Data Center World 2024 featured speakers from AMD and Intel discussing the transformative impact of artificial intelligence – and advanced graphics processing units (GPUs) – on the data center industry. They touched upon a wide range of topics including the latest advancements in AI algorithms and GPU architectures, how these are boosting energy efficiency, the role of AI-driven automation and how data center infrastructure is shifting dramatically to support burgeoning AI workloads.

Jennifer Majernik Huffstetler, chief product sustainability officer at Intel, started by explaining that it is not just processors that are facilitating the AI revolution. It takes a combination of:

  • Bigger and denser processors with far more cores.

  • Hardware and software optimized for energy efficiency to take advantage of the capabilities of AI.

  • Generative AI (GenAI) models optimized for specific use cases and workloads.

The Need for Domain-Specific Models

While ChatGPT may have become pervasive, it is not the best model for the enterprise, according to Huffstetler. Eighty percent of data remains on premises and the bulk of it is unused.

“To address business needs, smaller models are evolving that are domain specific,” said Huffstetler.

Related:Data Center World 2024: The Sessions I Don’t Want to Miss

She offered an example:One company owned a repository of 50,000 in-house documents. AI specialists trained a model on that content to provide insight for its consultants. The AI model operated within the firewall, thus eliminating many of the security, privacy and accuracy concerns related to internet-based models. As it focused solely upon in-house documents written by experts over many years, the model avoided the hallucination headaches that plague broader-based GenAI models.

Whether models are large or small, they demand a complete rethink when it comes to processing power, density, power and cooling, she added.

“New software and hardware solutions that use liquid cooling can cut energy usage by 40% but more innovation is needed,” said Huffstetler. “Through collaboration, we can lower energy consumption while enhancing performance.”

She urged patience in the headlong rush to an AI future. Virtualization technology took close to a decade to become pervasive. During that time, steady and sometimes spectacular gains took place.

“GenAI’s maturity will arrive much faster, but there are challenges to overcome,” said Huffstetler.  “What is clear is that liquid cooling will be required by high-performance GenAI apps and racks as we can’t achieve density any other way.”   

Related:Omdia Analysts Discuss Powering – and Cooling – the AI Revolution

AI Reciprocity

There may be plenty of work needed to prepare data centers to host demanding AI applications, but it is a reciprocal arrangement. AI can be harnessed to glean insights that can lower energy consumption and boost efficiency. Software tools already exist that can lower energy consumption in high-end processing by 20% to 30%, Huffstetler said. As AI develops, expect it to further enhance energy usage – even in legacy data center facilities. AI-based robotics, for instance, could become a game changer within a couple of years.

“AI will have many waves; in ten years we will look back and hardly recognize the data center,” said Huffstetler. “Software is the biggest lever for greater efficiency. It is estimated that 30% of modern processing capabilities are currently being wasted due to lack of software integration and orchestration.”

New Builds Versus AI Retrofits

During the Data Center World keynote, Laura Smith, CVP of engineering solutions at AMD, explained that current data centers are not fit for AI purposes. Modernization efforts, however, run into inherent limitations.

“Current data centers don’t have the right infrastructure to adopt new technology which is why there is such a big push to modernize them,” said Smith. “Space, power, cooling, floor design and budget constraints make it difficult.”

Despite the obvious problems that lie ahead, Smith believes the journey is worth it. Upgrades can drive density upwards to enable the facility to get more out of an existing space.

Before anyone embarks on AI-based upgrades, she recommended liaison with top management to explain the return on investment from what could be a large investment. Some organizations will act now to capitalize on the AI boom. Others will hedge their bets. But both camps should design their facilities and plan their upgrades with the future in mind.

“Even if you are not an early adopter, design with a view to what you are likely to need up the line,” said Smith.

She said it’s important to look beyond the latest technology. Her advice was to look a few years ahead to ensure you don’t get caught in perpetual upgrades. Such is the pace of progress that something new is likely to arise from the AI, GPU, cooling or power sector that might necessitate another round of upgrades. Thorough homework and due diligence can help data centers avoid the scenario where they complete one modernization initiative and have to launch straight into the next one.  

“The market is rapidly changing,” said Smith. “No one knows what the ideal architecture will be, but we still need to prepare.”

AI Better Than Humans? Not So Fast

Vlad Galabov, head of the data center practice at Omdia, rounded out the opening keynote by stressing today’s environment of constant change and steady technological advances. What seems like a huge AI initiative today may be modest in a couple of years.

“The science of AI is progressing so rapidly that model building is being disrupted,” he said.

Galabov doesn’t buy into the idea that AI will supplant humans – at least any time soon. He believes immense development remains to be done on the next phase of AI.

First there was predictive AI. Now we are in the GenAI phase. That is being followed by an artificial generative intelligence (AGI) phase, he said. AGI can be defined as human-like intelligence with the ability to self-teach and do far more than what is trained on.

“AGI has the potential to exceed human intelligence and capabilities in 20 years,” said Galabov. “That will only happen if we improve AI algorithms by four times each year for two decades in tandem with major advances in software and hardware optimization.

Data Center World is run by Informa Tech, Data Center Knowledge's parent company.

About the Author(s)

Drew Robb

Drew Robb has been a full-time professional writer and editor for more than twenty years. He currently works freelance for a number of IT publications.

Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.

You May Also Like