Infinity Turbine, a U.S.-based company focused on using waste heat to generate power, has developed a pulsed supercritical CO2 (R744) heat pump that provides efficient cooling for data centers with graphics processing units (GPU), says the company’s CEO, Greg Giese.
According to the company’s website, one Nvidia A100 GPU ‒ a new AI processing unit that doubles the previous generation’s memory capacity with a bandwidth of 2 terabytes per second ‒ generates 0.3kW (0.8TR) heat at temperatures between 40 and 60°C (104 and 140°F). As such, a data center operating with 50,000 GPUs must dissipate 15MW (4,265TR) of heat.
“CO2 technology is emerging as a compelling option for cooling data centers, particularly given its efficiency at high temperatures and eco-friendliness as a natural refrigerant,” Giese told NaturalRefrigerants.com. “Our unique design without mechanical components provides a high COP for handling the substantial heat loads of GPUs.”
Design details: According to Giese, the principal operating mechanism of the heat pump relies on cascade heat transfer and ejector pumps with no moving parts.
- In a heat exchanger, liquid CO2 absorbs GPU waste heat, causing it to be heated to a supercritical state with a temperature above 31°C ( 87.8°F) and pressure above 73.8bar (1,070psi).
- The heated CO2 then passes through an ejector pump system, which expands the refrigerant to create a modulated or pulsed pressure drop.
- The refrigerant automatically cools with the pressure drops, returning it to liquid CO2. According to a press release written by Giese, dropping the CO2 pressure from 80 to 10bar (1,160 to 145psi) lowers the refrigerant’s temperature to between −10 and −20°C (14 and −4°F), depending on the expansion.
- The cycle uses one-way valves to prevent back pressure and provide a reliable flow into the next heat exchanger.
Cost savings: By leveraging the Joule-Thomas effect, a supercritical CO2 heat pump can reduce data center energy consumption by up to 50% compared to traditional air- or water-cooling methods, Giese noted.
- Cooling large GPU data centers using air technology accounts for 30‒50% of total facility energy usage. According to Giese, the pulsed supercritical CO2 heat pump could cut energy use by 50%, slashing energy costs by up to $3.3 million (€3 million) annually based on a rate of $0.10 (€0.09) per kWh.
- Although water-cooled technology offers higher energy efficiency rates in data centers than air-cooled options, a supercritical CO2 system, with less infrastructure complexity, would cut water pumping and chilling costs.
- “The absence of mechanical components enhances durability, reduces maintenance, and minimizes operational noise,” Giese told NaturalRefrigerants.com.
Industry interest: With the rising demand for AI computation and the increasing use of GPUs, data centers want to cut energy usage.
- “Data centers are searching for ways to meet regulatory and environmental standards while managing escalating heat loads,” Giese said, with his company seeing substantial growth potential for CO2 heat pumps in the U.S. and Canada. He told NaturalRefrigerants.com that several data center operators have expressed interest in Infinity Turbine’s technology.
- Canadian manufacturer M&M Carnot reports that since 2012, it has installed Aquilon CO2 cooling systems in “dozens” of data centers in Canada. In 2022, the company released a white paper to help data facility owners navigate the cooling options.
- A data center in Spain, owned by international information technologies provider KIO Network, will use CO2 (R744)-based cooling and air-conditioning systems manufactured by Danish OEM Advansor.
Quotable: “Infinity Turbine is at the forefront, helping advance this technology for direct cooling of high-performance GPUs like the Nvidia A100, known to generate significant heat loads,” Giese said. “Our systems drive energy savings by utilizing low-grade heat more effectively.”
“CO2 technology is emerging as a compelling option for cooling data centers.”
Greg Giese, CEO Infinity Turbine
