2.2a Fundamentals of heat transfer: conduction, convection, and radiation in servers¶
🌡️ Context Introduction¶
Every server in a data center generates heat — lots of it. When you run complex AI workloads, GPUs and CPUs can reach temperatures that would melt standard electronics if not properly managed. Understanding how heat moves is the first step in keeping your infrastructure cool and reliable. This section covers the three fundamental ways heat transfers in servers: conduction, convection, and radiation. Think of these as the "physics behind the fans."
⚙️ Conduction — Heat Travels Through Solids¶
Conduction is heat transfer through direct contact between solid materials. In a server, this happens when a hot GPU chip touches its heat sink.
Key points: - Heat moves from the hot component (e.g., CPU die) to a cooler surface (e.g., heat sink base). - Materials matter: copper and aluminum are excellent conductors; air is a poor conductor. - Thermal paste (also called thermal interface material) fills microscopic gaps between the chip and heat sink to improve conduction.
Real-world example in a server: - A GPU generates heat at its core. - That heat conducts into the copper heat pipes attached to the GPU. - The heat pipes then carry the heat away to the fins of a radiator.
🌬️ Convection — Heat Moves Through Fluids (Air or Liquid)¶
Convection is heat transfer through the movement of a fluid — typically air or liquid coolant. This is how most data center cooling works.
Two types of convection:
| Type | How it works | Server example |
|---|---|---|
| Natural convection | Hot air rises naturally, cooler air replaces it | Passive cooling in low-power devices |
| Forced convection | Fans or pumps actively move the fluid | Server fans blowing air over heat sink fins |
Key points: - Server fans create forced convection to pull cool air through the chassis and push hot air out the back. - Liquid cooling uses pumps to move coolant directly over hot components. - The greater the airflow or liquid flow, the more heat can be removed.
Real-world example in a server: - A server's front fans pull in cool air from the cold aisle. - That air flows over the heat sink fins (heated by conduction from the CPU). - The now-warm air exits the server into the hot aisle.
☀️ Radiation — Heat Travels as Electromagnetic Waves¶
Radiation is heat transfer without any physical medium — it travels through empty space as infrared waves. In servers, this is the least dominant method, but it still matters.
Key points: - Every object above absolute zero emits thermal radiation. - In a server, hot components radiate heat to nearby cooler surfaces (like the chassis walls). - Radiation becomes more significant at very high temperatures (e.g., near GPU hotspots). - Shiny surfaces reflect radiation; dark, matte surfaces absorb and emit it better.
Real-world example in a server: - A hot GPU backplate radiates heat toward the server chassis side panel. - The chassis panel then warms up and may radiate that heat into the surrounding air.
🛠️ How These Three Work Together in a Server¶
A typical server cooling path uses all three methods in sequence:
- Conduction: Heat moves from the CPU/GPU die into the heat sink base.
- Conduction (continued): Heat travels through copper heat pipes to the fin stack.
- Convection: Fans blow air across the fins, carrying heat away into the data center aisle.
- Radiation: Some heat also radiates from hot surfaces to the chassis and surrounding equipment.
📊 Visual Representation: Server Heat Transfer Sequence¶
This diagram shows the sequential pathway through which heat generated by an NVIDIA GPU core is conducted, convected, and radiated away from the server chassis.
📊 Comparison Table: Conduction vs. Convection vs. Radiation¶
| Property | Conduction | Convection | Radiation |
|---|---|---|---|
| Medium needed | Solid contact | Fluid (air or liquid) | None (vacuum works) |
| Speed | Slow through solids | Moderate to fast | Speed of light |
| Dominant in servers | Inside chips and heat sinks | Airflow across components | Minor, but present |
| Example | Heat moving from GPU to heat sink | Fan blowing air over fins | GPU backplate warming chassis wall |
| Controlled by | Thermal paste, material choice | Fan speed, airflow design | Surface finish, emissivity |
🕵️ Why This Matters for Engineers¶
- Thermal design power (TDP): Every component has a TDP rating — the maximum heat it can generate. You must ensure your cooling system can remove that heat.
- Hotspots: Poor conduction (e.g., dried thermal paste) can create localized hotspots that damage hardware.
- Airflow management: Blocked vents or missing fans break convection paths, leading to overheating.
- Radiation awareness: In dense GPU clusters, radiation between closely packed servers can add to the thermal load.
Simple rule for new engineers: If a server feels hot to the touch, conduction and convection are working. If it feels very hot, radiation is also contributing — and you likely need better cooling.
✅ Summary¶
- Conduction moves heat through solid contact (chip → heat sink).
- Convection moves heat through fluid movement (fan → air → exhaust).
- Radiation moves heat through electromagnetic waves (hot surface → cooler surface).
- All three happen simultaneously inside every server.
- Understanding these basics helps you troubleshoot overheating, plan cooling layouts, and design efficient AI infrastructure.