Taking a Hybrid Approach to Data Center Cooling Saves Energy
It is a fact that chips are working harder than ever before. In our data-driven world, we rely on powerful computing systems that generate a lot of heat. The embrace of artificial intelligence (AI), machine learning (ML), and other high-performance computing functions is accelerating power demands. Since air cooling reaches its limits around 20 kilowatts per rack, traditional air-cooled systems can’t manage the heat waste as quickly or efficiently as liquid cooling technologies can.
Cooling already plays a significant role in overall data center power consumption. According to the International Energy Agency (IEA), cooling alone can represent around 40 percent of total power required to operate a data center. Due to this, it’s crucial to devise efficient cooling strategies, especially when IT data center owners and operators are under intense corporate and regulatory pressures to reduce energy consumption and increase infrastructure sustainability.
One Plus One, Equals Three.
This rapid acceleration of energy consumption is threatening sustainability objectives. A quarter of data center rack densities are already exceeding 20 kilowatts and five percent are exceeding 50 kilowatts according to the Uptime Institute. This kind of power is required for cooling central processing units (CPU) and high-density graphics processing units (GPU) essential for digital image processing, video rendering, AI computing, etc.
As the rack densities increase and temperatures rise, many data center owners and operators are evaluating or have already switched to liquid cooling technologies. That’s because liquid cooling increases thermal efficiency through power usage effectiveness (PUE), creates reductions in capital and operational expenses, and increases data center performance and reliability by maximizing computing capacity in the given space.
Two of the primary ways of applying liquid cooling are immersion liquid cooling and direct-to-chip liquid cold plates, but some don’t realize these technologies can be combined to take a hybrid approach to data center cooling.
What’s the difference?
Immersion and direct-to-chip are two distinct liquid cooling technologies, but both can be critical to thermal management of data centers. Key differences include how the technology is applied in the server and how the heat is transferred.
Immersion liquid cooling is a niche application, but it’s changing the way computer servers and data storage systems are managed and maintained. It involves submerging electronic components, server drawers, and racks into a bath of dielectric fluid. This fluid is a special type of coolant that is chemically engineered to serve two functions: cool the hardware through heat absorption and dissipation, and act as an insulator, preventing electrical interference that could disrupt normal server operation.
Direct-to-chip technology, which use devices known as cold plates or water blocks, is a conductive cooling device that protects hardware from overheating. This device is mounted directly to a microchip and the heat waste is transferred to a coolant that circulating through microchannels. Cold plate designs align seamlessly with existing data center infrastructure. The cooling efficiency of this technology is pivotal in cooling increasing compute power.
How do I integrate Direct to Chip and Immersion Liquid Cooling?
Immersion liquid cooling is a promising technology for managing high-power servers, and companies will gain even greater operational savings when it is paired with sophisticated, effective cold plate technology. As a proven innovator in thermal management, Mikros Technologies’ developed proprietary technology that enables the company to build higher quality cold plates for immersion liquid cooling applications. These designs can provide localized cooling where needed, thus increasing coolant bath efficiency. Here’s how it works:
- A cold plate is secured to a heat-generating component and the heat waste is transferred from the component to the cold plate (direct fluid contact).
- Next, the heat in the cold plate is then transferred to the dielectric fluid that surrounds it.
- Lastly, the heat in the dielectric fluid dissipates as it cycles through a larger volume of dielectric fluid in the enclosure.
Benefits of Technology Integration
Dielectric fluids alone will absorb and dispense heat, but when combined with the precision and high heat flux cooling capabilities of cold plates, the results are superior. Mikros Technologies will design a custom solution for your engineering department or production team, but we also offer standard cold plates that are ready to use now. These off the shelf products have ultra-low thermal resistances of 0.015-0.040 °C-cm²/W, high heat fluxes over 1000 W/cm², and will protect your CPU, GPU, and application-specific integrated circuits (ASIC) from overheating.
Cold plates also extend the life of your data center hardware by reducing temperature gradients on CPU, GPU, and ASIC die. Plus, the modular nature of cold plates allows for easy integration and scaling within immersion data center infrastructures. This flexibility ensures that as data demands grow, and chips get hotter, the cooling system can adapt and expand accordingly.
In addition, Mikros Technologies’ proprietary microchannel cold plates are built for reliability and durability. Their thin, lightweight design and brazed copper construction, ensures a reliable performance under high thermal loads and when submersed in dielectric fluid.
The Future of Data Center Cooling is Fluid
Cooling accounts for a sizeable portion of energy use. As the next generation of chips reach one kilowatt of power and data centers consume more and more electricity, it is important to embrace liquid cooling for thermal management strategies. The pairing of direct-to-chip and immersion liquid cooling systems is highly effective for increasing data center performance and sustainability for businesses.
If you have questions or would like a consult, contact us.