Home > Article > Technology peripherals > How AI is changing data center design
With global spending on AI systems set to double from 2023 to 2026, it’s clear that data center capacity will increase rapidly to meet demand.
Surprisingly, however, many data center operators have put the brakes on new projects over the past year, slowing investment, with London's vacant capacity falling in 2022-23 6.3%.
What’s behind this counterintuitive trend? To explain this, we need to understand some of the issues surrounding AI computing and the infrastructure that supports it.
Data centers have historically been built around racks using CPUs to handle traditional computing workloads, however, AI computing instead requires the use of GPU drivers Rack, compared with the same CPU capacity, it consumes more power, releases more heat, and takes up more space.
In practice, this means that AI computing power often requires more power connections or alternative cooling systems.
This is because the embedded infrastructure is built into the fabric of the data center complex, so the cost of replacing it is often very high unless it is completely economically unfeasible
In practice, Operators must ensure that their new data centers have a certain amount of space dedicated to the "split" between AI and traditional computing Center operators impose a permanently underutilized and unprofitable burden
A problem exacerbated by the fact that the AI market is still in its infancy. According to Gartner, we are currently in the midst of a hype cycle where expectations are overblown. As a result, many operators are choosing to stay on the sidelines during the design phase rather than commit too early to investing heavily in AI computing in new data center projects
Take a comprehensive approach during the design phase
To meet the need to be first movers while offsetting risks, operators need to design their data centers for maximum efficiency and resiliency in the era of AI computing, which requires a new holistic design approach.
1. Involve more stakeholders
So, to guarantee uptime and reduce the risk of costly problems over the life of the site, teams need to be more thorough in the data center planning phase.
At the beginning of a project, especially during the design phase, input from the wider team and expertise should be sought. In addition to seeking power and cooling expertise, designers should engage with operations, cabling, and security teams early to understand potential sources of bottlenecks or failures
2. Integrating AI into data center operations
3 # During peak periods, such as during training runs or when running enterprise-level models in production, AI places a greater load on the data center. During these periods, AI computing tends to significantly exceed traditional expectations in terms of power consumption, cooling requirements, and data throughput.
At the most basic level, this means that the underlying materials in the data center are under greater pressure. If these underlying materials or components are not of high quality, it means they are more likely to fail. Since AI computing means a dramatic increase in the number of components and connections at a site, this means that cheaper, lower quality materials that work well in traditional sites could bring the data centers running AI computing to a standstill
To avoid false economic risks, operators should avoid purchasing lower quality materials, such as substandard cables, to save money. These materials are prone to failure, requiring more frequent replacement, and most seriously, failure of substandard materials and components often results in a site's downtime or downtime, impacting its profitability. Therefore, operators should carefully select materials to ensure that they are of reliable quality See, that's not the case. Rewritten content: While the infrastructure requirements of AI computing may be the main reason why operators delay investment, in the long term, this is not entirely the case
To ensure that companies have every possible advantage in running their website, they need to ensure they learn and mature as the situation evolves
This means designing holistically from the start, Leverage AI itself to discover new efficiencies for their sites and invest in high-quality components that can handle greater AI computing needs.
The above is the detailed content of How AI is changing data center design. For more information, please follow other related articles on the PHP Chinese website!