Conjugate Heat Transfer Analysis on a graphics card:

The objective of this project is to study and analyze the use of fins to maintain the temperature of a graphics processor within it's limits. A graphics card is a computor hardware responsible for rendering images to the monitor. It is required when smooth and high quality visuals are desired by the computor user. There are two types of graphics card mainly integrated graphics which is built into the mother board and discreet graphics which is an extra component added to the motherboard. Here we are interested in the discreet graphics card or graphics processing unit.

As the gpu takes on heavy load for better visuals, it generates more heat and that affects the gpu performance and very high temperatures can damage the gpu. Hence, heat sinks and fans assemblies are common for gpu to pull the heat away from the processor. High-end graphics card have powerful fans and cooling system for optimal performance. But for low-end, weak graphics card, only a simple heat sink, usually metallic fins are provided.

Here we are analyzing a low-end graphics card with metallic fins attached as heat sink. The processor being a heat source will generate heat and the fin attached on top of it will help carry the heat from processor to the air flowing over the fin. The simulation shall be carried out in three different cases with different air flow velocity over the GPU.

Geometry:

The above geometry is a rough modelof a graphics card with a PCB board, cooling fins and processor inbetween them. The model is kept inside a rectangular enclosure.

Meshing:

The model was first given a default meshautomatically generated by fluent to test the simulation. Later a refined mesh was made. The graphics card, although a hardware with countless components, was grouped into 4 basic components and named as;

1. Processor
2. Fin
3. PCB board

• Baseline mesh:

With a total 84389 elements, the baseline mesh showed goodmesh metrics.

• Refined mesh:

For refining the mesh, the following tweaks were made on the baseline mesh;

1. The maximum element size was set to 5mm on the whole setup.
2. As the gpu processor is a rectangular block, multizone was enabled with hexa mesh.
3. Body sizing on the processor and fins as we are interested in heat exchange between them in this project. The max element size on processor and fin were set to 0.2mm and 0.5 respectively.

The total element count was now 486684.

• GPU mesh:

• Cross-sectional view:

Mesh metrics of refined mesh:

SETUP:

Absolute pressure based steady state solver with energy equation enabled and K-epsilon turbulance model was used to carryout the simulation.

The details of materials used are shown below;

Component	Material	Density(kg/m3)	specific heat(J/kg-K)	Thermal conductivity (w/m-K)
Processor	Steel	8030	502.48	16.27
Fin	Aluminium	2719	871	202.4
Board	FR-4	1900	1150	0.29
Cooling fluid	Air	1.225	1006.43	0.0242

Since heat conduction between solid bodies were involved, shell conduction was enabled. The heat generated by the processor was given as 2e+7 w/m3.

Boundary conditions:

Inlet air temperature: 293 K

Inlet air velocity: 1-5 m/s

Outlet pressure: 0 Pa

Results:

Convergence was observed when residual lines became horizontal and unchanging with each successive iterations.

Case1: Inlet velocity= 5m/s

• Baseline mesh:

Temperature distribution inside the enclosure:

• Refined Mesh:

Temperature distribution inside the enclosure:

The temperature distribution of the refined mesh results are much smoother and defined than that of baseline mesh. A closer look at the temperature distribution over different GPU components are shown below,

• Processor

• PCB

• Fin

• Velocity contour of airflow over the GPU:

• Heat transfer co-efficient over the model:

Similarly, the simulation was carried out for two more cases with two different inlet velocities.

Observations:

Inlet velocity(m/s)	Max temperature(K)	Max heat transfer coefficient(w/m2-k)
5 (baseline mesh)	312.14	346.83
5	306.8	468
3	310.9	375
1	326.15	214.25

Inferences:

After successfully simulating the cooling of a working GPU, the following points were inferred from the results and observations;

• The maximum temperature attained by the gpu was lowest for the highest inlet velocity case. The greater velocity of air provided more mass transfer and hence greater cooling for the processor.
• The heat transfer coefficient was also the highest for the same case as the aluminium fins are taking away more heat from the processor than the rest of the cases as it is being cooled at a faster rate.
• The potential hot-spots were observed at the processor-PCB interface and fin-Processor interface as these parts are in direct contact with the heat source. The low conductivity of the PCB material makes it a dominant hotspot.