The edge has emerged as a major area of infrastructure investment, with 27% of IT resources expected to be allocated to the edge by 2026. As this low-latency computing environment becomes more densely populated, direct-to-chip (DTC) liquid cooling technologies will prove crucial to edge computing architectures.

Traditional air cooling is only effective for compute densities of up to 10 kilowatts (kW) per rack. When faced with the typical high thermal design power (TDP) at the edge, air cooling isn’t efficient enough, JetCool CEO Bernie Malouin told SDxCentral.

Single-phase immersion cooling similarly faces challenges at the edge, including high flow rates, increased wear and tear to devices, system complexity and maintenance. Despite attempts to cool high TDP servers with single-phase immersion cooling, “recent developments haven’t achieved more than 1,000 watts of cooling capacity,” and “the operational conditions have not yet met the compliance standards set by chip manufacturers,” Malouin said.

Coping with rising chip temperatures These single-phase setups are often experimental, pushing chip temperatures to more than 80°C and requiring 7 liters of flowing liquid per minute. This scenario does increase server performance, but “at the cost of efficiency,” he said.

Liquid cooling vendor JetCool is hoping to capitalize on these constraints with its liquid-assisted air-cooling SmartPlate System, which has been “accelerating at the edge” in particular, Malouin said.

With the same form factor as conventional air-cooled servers, data center operators using JetCool technology can deploy high performance computing (HPC) (HPC) like artificial intelligence (AI) applications at the edge without the physical constraints of other liquid cooling solutions, according to the vendor.

Two-phase direct-to-chip liquid cooling, for example, effectively cools high TDPs associated with AI processors, but the waterless option relies on per- and polyfluorinated substances (PFAS) – “a forever chemical that is being regulated out for safety concerns,” Malouin said.

JetCool’s approach with single-phase DTC uses safer fluids and is able to cool more than 1,000W TDPs per socket, which means it’s capable of supporting the latest Nvidia AI GPUs.

In addition to efficient cooling, edge computing environments require location-agnostic cooling systems that need minimal maintenance. This remotely located and monitored environment also necessitates flexibility in the liquid cooling system to adapt to harsh conditions.

“With no maintenance for the lifetime of the server, these liquid-assisted systems can truly be a plug-and-play solution that easily expands in tandem with the growth of edge computing strategies,” Malouin said.

JetCool saves customers energy Today, JetCool’s liquid cooling system is deployed in customer environments with temperatures more than 35°C and servers at full capacity with 1,000W TDP. According to test data from JetCool’s OEM partners, the SmartPlate system reduces power consumption by 15% compared to air-cooled servers running identical workloads.

The vendor’s SmartPlate technology is a closed-loop system that’s integrated into the server, meaning it doesn’t require external plumbing or pipes. According to the company, the technology saves up to 150W of power per server, which makes it possible to add more servers to the rack and maximize compute in power constrained environments like the edge.

JetCool customer Sabey Data Centers, for example, is using the system in production. Since integrating the technology, the data center operator has reported reducing total power consumption, not just cooling-related power consumption, by 13.5%.