As technology progresses at a dizzying pace, the demands placed on data centre infrastructure have reached unprecedented levels. With the rise of Artificial Intelligence (AI), High-Performance Compute (HPC) platforms, and GPU systems, the nature of data storage and management has dramatically shifted.
Today, we dive deep into the evolving challenges and solutions surrounding these cutting-edge technologies, focusing mainly on the critical issues of extreme-density racks and cooling solutions.
Understanding Extreme Density and Its Implications
AI solutions and HPC platforms in data centres now require extremely high-density racks. A “high density” IT rack can consume up to 35kW of energy, translating to equal heat production. Taking this a notch higher, “Extreme Density” racks have an energy consumption ranging from 50kW to a staggering 150kW per rack.
Addressing the challenges posed by these extreme-density racks is a challenge. Beyond the sheer amount of energy and heat they generate, there are other concerns. These encompass issues related to airflow, direct-to-chip cooling methods, immersion capacity, heat exchange, floor planning, power distribution, real-time monitoring, resilience, maintenance, fire management, data storage, and cabling.
Cooling: A Major Concern in Modern Data Centres
Historically, air has been the preferred medium to cool IT racks in data centres. Oper8 Global, for instance, applied air cooling up to a density of about 35kW per rack. But with racks now demanding up to 150kW or even more, air cooling falls short. As the demand grows, alternative solutions like direct-to-chip (DLC) cooling and immersion cooling are emerging. However, each of these solutions comes with its own challenges, necessitating innovative approaches. This is where entities like Oper8 Global are stepping in, leading the way in modular HPC infrastructure solutions.
The Surge of Hyperscale Data Centres
“Giant data centres more than 20 times larger than what they were just a few years ago are coming to Australia due to the surge in the use of artificial intelligence and the huge demands it puts on data storage.” “AI is real. The AI of now is very different from the AI of 10 years ago,” said Robin Khuda
Oper8 Global has been seeing the same increase in AI, GPU and HPC requirements through our EMEA team supporting our European customers over the past three years, with early-stage Australian projects being planned over the last six months. AI, GPU and HPC outside the cloud provide a number of benefits, as mentioned later in this article.
The Energy Question
The focus on energy, sustainability and overall efficiency for such massive hyperscale data centres should also expand into the real impacts on the energy grid and truly look at where the energy is sourced from.
You’d be hard-pressed to find a couple of unloaded power stations close to Sydney and Melbourne needed to support expected population growth, the almost mandated use and recharging of EVs, and spare capacity for more hyperscale data centres.
Most people typically think a 100MW hyperscale data centre needs 100MW of power. In fact many industry peers regard the 100MW statement as the IT capacity the facility can support across the IT racks within. When considering a PUE of 1.3, a 100MW facility requires 130MW from the grid.
When considering the grid loss in Australia (from power station to market customer) is approximately 10%according to AEMO, the 100MW hyperscale data centre needs approximately 144MW from the nation’s power stations 24 hours a day.
The Edge: A Solution for Australia
Australia needs to follow Europe’s example in getting to the edge. From an Australian perspective, how do we see and deliver more Edge Data Centre projects in the UK?
Oper8 Global is deploying Edge Data Centre solutions for customers on-premise for reasons including latency (another AI and HPC requirement), direct platform control, Intellectual Property and trade secret protection, GDPR and, in many cases, to differentiate their company’s performance, output and market leadership.
What is interesting is that in a country the size of the UK, latency is one of the primary drivers for HPC and AI at the edge.
It’s also worth remembering not all AI platforms are like ChatGPT, with the end users worldwide and connected via the internet. Most AI and HPC solutions are designed to meet the specific compute and performance needs of the customer.
While there are many benefits in building and deploying at scale (e.g., hyperscale in the 100MW range), it poses significant headaches for countries reaching the installed capacity of major transmission paths and power generation. New power stations take time, approvals, and money.
While wind and solar are key parts of many national energy strategies, consideration must be given to the location of significant manufacturing and industrial hubs and where hyperscale (data factories) can fit into the grid.
In the meantime, and as alternative speed-to-market options for business and government in Australia, HPC and AI at the edge spread the energy load across the national grid, meets the latency requirements of high-performance computing, provides direct customer control of the platform, and faster to build, commission and operate.
Looking Ahead: The Potential of the Next Five Years
While millions of customers might be in the cloud, millions of locations have no cloud infrastructure or hyperscale data centres. These smaller locations will still be a key part of the global IT marketplace with localised modular and micro data centre infrastructure.
As with European customers, AI, GPU and HPC platforms are helping differentiate innovative and high-performing businesses while underpinning research, analytics, and machine learning outside the cloud, mainly where latency is a key requirement.
There are many customers spread across Australia, New Zealand, and the Pacific who can and should be considering AI, HPC and GPU solutions and how it delivers value for the business. Solving the extreme density data centre infrastructure challenge while distributing the load across the energy grid is where Oper8 Global can help.
As noted in the Financial Review, giant data centres 20 times larger might be coming to Australia; the question is when and how does it help you?
Oper8 Global delivers, manages and supports HPC, GPU and AI solutions at the Edge (customer-defined locations) with extreme density racks supported through our modular, prefabricated and micro data centre solutions. Our infrastructure solutions are HPC, AI and GPU platform agnostic to help you achieve the expected performance from your IT investment.