With the growth of artificial intelligence in data centers of all types comes not only requirements for applications and services, but the need to cool the power-hungry systems that run those workloads. With that in mind, Lenovo has announced some new enterprise hybrid AI solutions, and cooling technologies to deal with the heat they generate.
AI workloads in particular are driving up data center densities, with power consumption now topping 20 kilowatts per rack. Air-cooling systems struggle to keep up at this point, so data centers are increasingly turning to liquid cooling — almost one-quarter now have at least some workloads cooled this way.
Lenovo has offered such technology for a while, now, and has just introduced the sixth generation of its liquid cooling platform, Neptune. This now extends its reach through its ThinkSystem V3 and V4 server portfolios with options for direct open-loop water cooling for CPUs. The new generation also adds a new cold plate design for CPUs and accelerators to maximize heat extraction, new on-memory cooling for high performance computing systems, and warm water cooling that allows hot water reuse in a facility.
“For large language models that are really taking hold of the market today, sustainability is going to be a huge hurdle for them. And that’s where Neptune liquid cool portfolio comes in, and helps [companies] achieve both balancing their sustainability goals with their mission to go develop and bring new AI technology to market,” said Robert Daigle, Lenovo’s director of global AI business, at a media briefing on the new products.
For companies that can’t install liquid cooling their own data centers, Lenovo has partnered with colocation provider Digital Reality to provide liquid cooling in more than half of its global data centers.
Jeremy Roberts, senior analyst at Info-Tech Research Group, said, “AI is subject to heavy scrutiny for its carbon footprint, and a solution — even if it does end up being more complex or expensive — that can reduce that carbon footprint in a meaningful way could be meaningfully beneficial for the purchaser.”
Liquid cooling is a great idea in theory, and has been around for a while, but vendors have struggled with the practicalities, he said: “This all sounds good, but it’s historically been difficult to do in practice. There’s a reason it’s niche and why Lenovo is pitching this as a new mainstream.”
There are a number of issues around the technology that companies need to consider, said Roberts. To begin with, most workloads don’t require a complex liquid cooling infrastructure. AI workloads might, and that’s probably why Lenovo is pushing this, he said. But liquid in data centers is not generally a good thing if it can be avoided, and managing plumbing could be a challenge.
“Do I see large-scale adoption in the short term? Probably not,” he concluded. “But do I see the need for this type of cooling solution increase as computing needs grow? That feels more likely.”
New AI services
At the same time as it unveiled the new cooling systems, Lenovo also announced new AI services running on Nvidia chips and software.
New services include Lenovo AI Fast Start for Nvidia NIM inference microservices, to give developers an easy to use and manageable containerized inference engine for Nvidia’s AI foundation models; AI Fast Start for Innovators to help companies build AI use cases with the combination of any of the 165+ turnkey Lenovo AI innovators program solutions; and upcoming AI advisory services, which will include training and tools to help customers engage and enable their employees.
There’s also Lenovo AI Discover to help customers assess their AI readiness, offering recommendations across the realms of people, process, technology, and security.
It also introduced new Fast Start services, some with broad appeal and some for extremely niche markets, including an AI-powered kiosk with a virtual employee to serve customers; a video-based analytics system for improving manufacturing efficiency and compliance; a video building block for self-checkout and retail loss prevention; and an AI-based system for reducing bird strikes at airports.
Source:: Network World