Supermicro CEO would like it if you could all build new, greener datacenters
Using guess-who's datacenter equipment, natch
Transitioning just half of all datacenters to sustainable operating models could cut global energy costs by $7 billion, Supermicro CEO Charles Liang claims.
Building greener datacenters was a central theme of the company’s nearly hour-long Computex keynote earlier this month, where Liang touted Supermicro’s efforts to reduce carbon emissions through modular server design and encouraging liquid cooling.
“It’s our responsibility to reduce the environmental impact of technology by using less power, reducing carbon footprint and e-waste,” he said.
Supermicro is no stranger to green datacenter initiatives, as Liang highlighted throughout the presentation. This included the company’s contributions to the MN-3 supercomputer cluster, which is among the most energy efficient supercomputers today and has held the number-one spot on the Green500 for the last few years.
Liang also talked up his corporation’s efforts to normalize liquid cooling in the datacenter. He highlighted the company’s recent collaboration with Swedish datacenter operator GleSys to repurpose water used to cool the facility to warm homes in the region, a technique billed by Dell'Oro Group analyst Lucas Beran as the holy grail of datacenter cooling.
Supermicro’s efforts in the arena haven’t only been practical. Many of the biggest barriers to adoption relate to misconceptions about the technology, Liang explained.
“Our new redundant liquid cooling technology simplifies the deployment and addresses concerns of serviceability,” he said. “This solution will be pretested and validated across a broad portfolio of systems.”
Liang also claimed the liquid cooling “doesn’t cost anything extra on initial hardware acquisition costs,” and instead highlighted the lower operating costs associated with more efficient cooling technologies.
“With optimized hardware and software system designs, our total IT solutions allow our customers to achieve the best performance per watt, including [power use efficiency] up to 1.06 or better,” he said.
Supermicro’s enthusiasm for people buying liquid-cooled Supermicro hardware comes as a wave of high-wattage compute platforms are slated to hit the market.
Many of Intel, AMD, and Nvidia’s latest GPUs and AI accelerators are now sucking down 600W or more, while many CPUs have thermal design powers topping 300 watts — 500W in the case of Nvidia’s Grace Superchip CPU.
- Plans for Dutch datacenter to warm thousands of homes
- Nvidia brings liquid cooling to A100 PCIe GPU cards for 'greener' datacenters
- Intel plans immersion lab to chill its power-hungry chips
- Intel commits to net-zero greenhouse gas emissions by 2040
Supply chain efficiencies
Liang also highlighted the importance of sustainable supply chains during a brief presentation alongside Intel Fellow Shesha Krishnapura. Krishnapura is a long time champion of disaggregated compute architectures, including blade servers.
“It’s not just about the cost of ownership, which everyone understands and gets,” Krishnapura said. “Green computing isn’t just about running the datacenter in the most energy efficient way and using less of the natural resources and less electricity. It’s also about reducing the e-waste.”
He pointed to disaggregated servers as a prime example. By enabling a single chassis to utilized across multiple processor generations without being replaced, e-waste can in large part be eliminated.
“When you upgrade the servers, for example, a CPU-complex-only upgrade will reduce the e-waste by 83 percent,” he claimed. “That means only 17 percent by weight of the components need to be upgraded. Thus, the shipping costs get reduced significantly as well as the cost of making it.”
This philosophy isn’t limited to Supermicro’s blade servers either. On stage the company demoed its recently announced Universal GPU Servers, by swapping out a GPU daughter board equipped with AMD’s MI250 for an Nvidia A100 SXM module.
Green computing catches on
Supermicro wasn’t the only vendor touting modularity and liquid cooling at Computex. Earlier this week, Nvidia unveiled its first Grace and Grace-Hopper Superchip reference designs, which also utilize a disaggregated chassis. Two of Nvidia’s HGX Grace-Hopper or four HGX Grace nodes can be slotted into a single chassis for system power.
Perhaps unsurprisingly, Supermicro was one of the launch partners slated to build systems based on the HGX platform. Other vendors included Asus, Foxconn, Gigabyte, QCT, and Wiwynn.
Nvidia also announced liquid-cooled versions of its PCIe-based A100 GPUs in a bid to reduce the impact its power hungry chips have on datacenter power consumption.
While liquid cooled A100s have been around for some time, it’s only been available in the larger, higher TDP SXM form factor. The new PCIe-based cards bring liquid cooling to a more mainstream audience.
The announcements also follow Intel's $700 million sustainability initiative, announced last week, which will see the chipmaker construct a 200,000 square foot ‘mega-lab’ to research and develop innovative liquid and immersion cooling tech. ®