Turning up the power


HomeHome / Blog / Turning up the power

Aug 08, 2023

Turning up the power

Server CPU TDPs are set to soar as AI and other applications demand more power. GRC’s John Bean and Alex McManis tell DCD how immersion cooling will keep them cool Technological progress, like so many

Server CPU TDPs are set to soar as AI and other applications demand more power. GRC’s John Bean and Alex McManis tell DCD how immersion cooling will keep them cool

Technological progress, like so many things, typically starts slowly, but then the momentum rapidly becomes unstoppable: the internet, PCs, smartphones, flat-screen TVs and so on. When a technology reaches a certain point in terms of both utility and price, adoption positively explodes.

Immersion cooling is now on the cusp of doing likewise. It has been nibbling away at the data center market for some 15 years or so. But with Moore’s Law increasingly challenged by physical reality, and data centers’ need for more raw compute power to deliver resource-intensive AI applications, among other pressing demands, immersion cooling is a uniquely capable technology that will enable data centers to deliver.

“Immersion cooling has a lot of benefits that go beyond density. Sure, density means that we can achieve much higher kilowatts per rack. But the larger play is TCO – total cost of ownership: it’s simpler to build it without the airflow engineering required of air cooling infrastructures,” says Alex McManis, vice president of Digital Asset Solutions at Green Revolution Cooling (GRC).

“Then there’s the energy savings that cut operating costs. You have lower power usage from the servers because you’re not running fans. And, finally, there’s the efficiency of the cooling solution in removing excess heat,” he adds.

Moreover, immersion cooling enables the heat generated by servers to more easily and directly be captured and re-used, at a higher and therefore more viable temperature, says John Bean, chief technology officer (CTO) and vice president of engineering at GRC.

Moreover, the single-phase immersion cooling solutions offered by GRC and a number of other vendors are also easier to use than the more complex two-phase immersion coolers also available on the market, he adds.

“Early in my career I focused on mainframe chillers. I’ve been involved in the Open Compute Project (OCP) and ASHRAE, and have seen the promise of immersion continue to grow. It became apparent to me that single-phase immersion cooling was probably a more practical path than two-phase immersion cooling. While there’s no such thing as a perfect solution, in terms of the breadth of possible deployments, single-phase simply looks better,” says Bean.

These twin CTO and engineering roles mean that Bean has to turn his mind to both immediate, practical engineering issues on behalf of GRC’s growing roster of customers and technology partners, as well as the longer-term technologies and innovations that will shape the future of immersion cooling.

“As a technology CTO, I’m more focused on future technologies, innovation, and understanding the market. As an engineering leader, I’m more about the implementation and execution of those aspirations from a technical standpoint. So, not only setting the technology path but also implementing the technology paths,” he says.

Indeed, Bean is one of a number of key people helping to decide the direction that GRC’s research and development efforts will take over the next few years, which means that he has a considerably better idea of what the future holds for immersion cooling than most.

While a typical PC CPU runs at around 65-105 watts, the server CPUs supported today by immersion coolers are designed to run at up to 400 watts, wringing lots more performance out of the silicon. The research and development being pursued by Bean’s team at GRC is intended to take support for server CPUs running all the way up to 1,000 watts – well beyond anything even the most advanced air-cooled solutions will ever be practically capable of supporting.

There are a number of areas that GRC’s research and development is targeting.

“We see innovations coming in terms of how we manage the fluid flow around the server itself, making sure that we’re getting good fluid motion at these higher heat-flux CPUs or GPUs. There’s a number of emerging technologies that we have some IP [intellectual property] around to do that. We’re looking forward to expanding our portfolio to manage that,” says Bean.

“There will be continual changes to the fluids, too. But those are not necessarily going to be the biggest levers. I think the biggest will simply be better flow through the heat-transfer surfaces (or across them).

Other means include developments in thermal interface materials, improved heatsink design, etcetera; research to improve the thermal interface materials that behave well thermally also needs to be balanced with good materials compatibility,” says Bean.

GRC is, therefore, also re-examining heat-sink technology, where Bean believes that new designs questioning old assumptions, could drive radical improvements. He cites recently published research that indicates that vapor chamber heat sinks outperform conventional heat sinks with heat pipes (to transport heat evenly) embedded within them.

The fluids, of course, are one of the key elements of immersion cooling systems. Single-phase immersion cooling fluids are, naturally, more stable than the fluids used in two-phase systems, because in two-phase it is the boiling effect that is used to affect the heat transfer. This also makes these cooling systems – and the servers they are cooling – more challenging to run and maintain.

But even among single-phase immersion cooling systems, there is a wide variety of liquids that can be used, and research is ongoing to create fluids even more suitable for heat transfer.

“Right now, we typically use a synthetic hydrocarbon of polyalphaolefin (PAO), although we do have some partners who use GTL – gas-to-liquid,” says Bean.

“These fluids are formulated and refined according to different methodologies, whether that’s focusing on their thermal conductivity or their coefficients of expansion. But there’s a number of characteristics that can be tweaked just a little bit, one way or the other, that can make that fluid better suited to, for example, heat transfer,” says Bean.

Developing fluids for immersion cooling technology, in some respects, is as much an art as a science – some adjustments may improve heat transfer at the expense of viscosity, for instance, which may then affect liquid flow. It is the knowledge of the team at GRC that can bring all that together to drive overall performance improvements.

“We’ve been using a fluid partner for some years now and have gone from using their PA-06 to PA-05 to PA-04; that’s basically lower and lower viscosities. From our own testing, as we transitioned from PA-06 to PA-04, we saw a six-to-seven percent improvement in the heat transfer coefficient. That’s not trivial,” says Bean.

It’s these kinds of incremental improvements across every element of immersion cooling – thermal interface materials and heat sink design; fluid dynamics, chemistry, and viscosity – that GRC is planning to push over the next few years that will, together, drive revolutionary change in immersion cooling.

The role of the server is also going to be a critical factor in immersion effectiveness- possibly the most important. Today’s servers are designed for air and modified for immersing in the GRC systems. We know that purpose designed servers for immersion will open up huge incremental improvements to the impact of immersion. And we know that the server itself will cost less contributing to the TCO benefits of moving from air and other technologies to immersion!

Moreover, adds Bean, while the air cooling companies undeniably have a 20+ year head start, the scientific fundamentals of both air cooling and immersion cooling are similar: Bean, who only joined GRC in 2021, is looking to apply those decades of lessons learned to immersion cooling technology – but over the next few years, not decades.

“There’s a long runway still in front of us,” says Bean. “And we’re only really at the beginning of the journey.”

In other words, it would be wise to keep an eye on some of the research and development that will be coming out of GRC and other companies involved in immersion cooling, as the transition from air-cooling to immersion is now starting to heat up.

To find out more about the future of immersion cooling, check out GRC's latest white paper, The Future of Immersion Cooling: The Path to Cooling 1000W Chips, and Beyond

Another immersion liquid provider certified with GRC

Data centers won't use liquid cooling till they have to. Could that time be near?

When humans really need to cool down, fans are not a patch on immersion in a pool of cool water. So why not adopt immersion cooling in data centers?