Category Archives: Air Cooling

Air cooling was once the dominant method for cooling processors on a PCB, employing heat sinks, fans, and blowers. These solutions are still popular, but faster-running, higher-powered devices, like most GPUs used for AI can’t be sufficiently cooled with air. Liquid cooling is needed. [1,7]

There isn’t a universal application point where engineers must switch from air cooling to liquid cooling. The decision mainly depends on allowable temperature and heat flux. Other considerations include higher costs and power needs, added hardware, space and weight, and required more monitoring. [1,3]

Fig 1 – Air Cooling Heat Sink vs. Liquid Cooling. [9,10]

In practice, once devices reach several hundred watts in a compact package, or when the required thermal resistance approaches 0.1°C/W or lower, liquid cooling often becomes the preferred solution.

Low Thermal Resistance

Engineers usually don’t switch to liquid cooling simply because a chip reaches a particular wattage. They switch when the required junction-to-ambient thermal resistance falls below what can be achieved economically and reliably with available airflow and heat sink models. [2,3,5]

Consider an equation and example:

T_j = T_α+ P x R_αJA                              [2]

Where:

T_j = junction temperature

T_α = ambient temperature

P = power dissipation

R_αJA = thermal resistance from junction to ambient

If keeping the chip within its temperature limits requires an unrealistically low thermal resistance, air cooling may not be practical.

Example:

Ambient = 40°C

Maximum junction temp = 90°C

Chip power = 500 W

The allowable thermal resistance is:

R_αJA = 90 – 40 /500 = 0.1°C/W

Achieving 0.1°C/W with air is extremely difficult. Liquid cooling is the practical solution. [2,3]

High Heat Flux

Heat flux (power per unit area) is often more important than total power. [3]

For example:

• A 100 W device spread onto a large heat sink may be easy to air-cool
• A 100 W device concentrated in a 1 cm² die can be much harder

Modern high-performance processors, GPUs, laser diodes, RF power amplifiers, and power electronics often reach heat fluxes that make air cooling difficult.

Fig 3 – Heat Flux Measures the Amount of Heat Energy Transferred from a Warmer Side to a Cooler Side. [12]

Typical industry guidance suggests:

• Air cooling: comfortable below ~10–20 W/cm² [1,6]
• Advanced air cooling: can sometimes handle 50–100 W/cm² with aggressive airflow [1,6]
• Liquid cooling: often becomes attractive above ~100 W/cm² [1,6,7]

In basic terms:

• Air cooling is usually limited by how much heat you can transfer from the heat sink to the air.
• Liquid cooling is usually limited by how much heat you can get from the chip into the coolant.

Where Air Cooling Works Well

Fig 4 – Heat Sinks from Advanced Thermal Solutions, Inc. [9]

• Total heat dissipation is relatively modest, up to a few hundred watts
• There is enough surface area for heat sinks
• There is sufficient airflow through the enclosure
• Higher fan noise is acceptable  [1,3]

Air cooling is used in many consumer electronics products and networking devices.

Coolants Remove More Heat

Since water-based coolants have much higher heat capacity and thermal conductivity than air, they can transport large amounts of heat away with relatively small temperature rises. [3,9]

For example:

• Air’s specific heat capacity is about 1 kJ/kg·K 
• Water’s is about 4.2 kJ/kg·K   

And water is about 830 times denser than air, so a small coolant flow can carry away an enormous amount of heat compared with an equivalent volume flow of air. [9]

Fig 5 – Tubing and Ports on a Liquid Cooling System [9]

That’s why modern AI accelerators, high-power motor drives, and high-density data-center servers are increasingly moving toward liquid cooling: not because air cooling is impossible, but because the required heat removal density becomes impractical with fans and heat sinks alone. [1,7,8]

The Benefits of Faster Flow Rates

Air cooling performance increases when the rate of airflow increases. Thus, many heat sinks are supported with fans and blowers to enhance heat transfer into the passing air. [3]

Liquid cooling, similarly, is enhanced with higher flow rates. A common expectation is that a coolant that stays longer at the heat source (chip) will absorb more heat. But, at slower flow rates, the coolant heats up as it moves through a cold plate and downstream in the cooling loop. As a result, the returning coolant has less of a temperature difference from the chip, and heat transfer becomes less effective. [3,4]

With a faster flow of liquid coolant, the fluid temperature stays more uniform and keeps a steady, higher temperature difference from the chip. This makes heat extraction more efficient and improves the lowering of chip temperatures. [3,4]

Flow rates can be optimized per application. There are diminishing returns on performance as flow rates increase, as well as added costs. Your liquid cooling system provider should provide performance data and recommendations for proper flow rates. [3,4]

In-Between Air and Liquid Cooling

There are options that fall between air and liquid cooling methods. Engineers should look through these choices before designing in a liquid system. They could provide simpler, lower-cost, but reliably effective cooling       

The expanded options include:

1. Better PCB thermal design (thermal vias, thicker copper planes)
2. Larger heat sinks
3. Forced-air cooling with optimized airflow paths
4. Heat pipes or vapor chambers to spread heat
5. Remote heat exchangers connected by heat pipes
6. Liquid cold plates       [3,4]

Fig 6 – Heat Pipes May Meet Cooling Needs in Place of Air or Liquid Cooling [9]

Not a Tradeoff

Traditional air cooling is inadequate for many of today’s thermal management needs. Other methods, including enhanced air cooling  systems, may offer solutions. But for many applications, including fast-growing AI units and data centers, liquid cooling is essential. In this case, there is no tradeoff because its use is the only option. [1,6,7]

Consider This

Here are some analogies for comparing air to liquid (water) cooling:

Air is a ghost. Because air is so light and empty, a single cubic meter of it can only grab a tiny handful of heat before it gets too hot and gives up.

Water is a sponge. Because water is about 830 times denser and packed tight with mass, that same one cubic meter acts like a massive thermal sponge. It can swallow up a staggering amount of heat before its temperature rises even a single degree.

To cool a hot system, you either have to blow a hurricane of lightweight air past it, or gently glide a tiny, heavy stream of water over it.

References

1. ASHRAE. Thermal Guidelines for Data Processing Environments, latest edition.
2. JEDEC Solid State Technology Association. JESD51 Series: Methodology for the Thermal Measurement of Component Packages.
3. Frank P. Incropera, David P. DeWitt, Theodore L. Bergman, and Adrienne S. Lavine. Fundamentals of Heat and Mass Transfer, Wiley.
4. David A. Reay, Ryan McGlen, and Peter Kew. Heat Pipes: Theory, Design and Applications.
5. Texas Instruments. Thermal Design by Insight, Not Hindsight (Application Report).
6. Open Compute Project Foundation. Advanced Cooling Solutions documentation.
7. NVIDIA. Data Center Liquid Cooling technical papers and deployment guides.
8. National Institute of Standards and Technology (NIST). Thermophysical Properties of Fluids Database.
9. Advanced Thermal Solutions, Inc., https://www.qats.com
10. Chatsworth Products, https://www.chatsworth.com/en-us/resources/blogs/2026/5-misunderstood-facts-about-direct-to-chip-liquid-cooling/
11. Advanced Thermal Solutions, Inc, https://www.qats.com/
12. EngineerExcel, https://engineerexcel.com/flow-of-heat/