- Nvidia’s Rubin generation AI servers use coolant that is often hotter than a hot tub because it enters a closed loop at up to 45°C and exits at around 55°C, without any performance penalty.
- The principle used here is that of a temperature gradient with server chips always hotter than the coolant, allowing for smooth heat transfer.
- This approach allows Nvidia to keep about 2.6 million gallons per megawatt per year near zero and could save a 50 MW site more than $4 million per year.
Nvidia’s Rubin generation is the first of its kind in many ways, but perhaps the one that stands out is the fact that it is 100% liquid cooled, implementing it as a core design feature at the platform level with every chip and networking component covered by a closed loop.
Nvidia’s reference design aims to enable AI factories that effectively consume zero water by implementing a closed-loop circuit that does not leverage evaporative water cooling 99% of the time.
The approach goes even further by potentially removing industrial cooling plants from the equation by using liquid cooling where the temperature can reach 45 degrees Celsius for the coolant entering the system and around 55 degrees Celsius when it leaves.
A numbers game for a data center that analyzes them on an industrial level
Nvidia’s approach is not the first of its kind in terms of thermal design or implementation; However, IBM beat it by 16 years by using 60-degree coolant to cool its supercomputer prototype at ETH Zurich.
The principle was the same: a much lower carbon footprint, significantly lower energy consumption and a loop filled once and left alone. What Nvidia adds is scale: its Vera Rubin-based DSX design extends the approach to an entire installation rather than a single machine.
“The NVIDIA DSX reference design for AI factories has zero water consumption – we have eliminated massive amounts of power consumption and virtually all water consumption,” said Ali Heydari, director of Nvidia data center cooling and infrastructure.
The coolant is a 3:1 water/propylene glycol mixture, and it easily carries heat away from the chips much hotter than the coolant itself.
The savings aren’t just based on water: Industrial cooling plants reduce their energy costs by about 4% for every degree they increase in operating temperature, and Nvidia estimates that a 50-megawatt hyperscale facility could save more than $4 million per year in cooling-related energy and water costs by adopting its design.
With cooling historically accounting for up to 40% of a data center’s total electricity bill, Nvidia’s approach is ambitious, but it should pay for itself quickly: by eliminating server fans altogether while allowing coolers to run only when necessary, and at considerably lower power than a loop requiring cooler coolant.
It’s worth noting that DSX is a reference design: a set of best practices for building an AI factory, not a census of what the industry has actually built, where many designs still consume very large amounts of water, even as AI data centers continue to pop up in drought-affected parts of the United States.
Nvidia may seem to be championing a “greener” AI data center for the future, at least in its own designs, but that’s only part of the story – and the part it leaves out is a little less flattering. Nvidia didn’t embrace hot water cooling because it’s sleek, eco-friendly, or smart, even though it’s all three. He adopted it because his accelerators became so energy dense that moving air through them was no longer physically viable, and once you commit to liquid cooling, heating it up is simply the efficient way to do it.
The sustainability win is real and is a consequence of a thermal problem that Nvidia created for itself by building the most energy-dense chips in the industry. Hot water is not the solution. This is the bill coming due, paid in the most efficient currency available.
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds.
