Conjugate Heat Transfer Analysis on a graphics card.

Aim

To perform CHT analysis on a basic graphic card to study the temperature characteristics of components and determine the maximum temperature attained by the processor in ANSYS Fluent

1. Introduction

A graphic card or a video card is a specialized electronic circuit designed to accelerate the creation of images displayed in the output monitor. With the advancements in computers and processing hardware, the graphic processing unit has become an integral part of the electronic industry and finds its application in computers as dedicated or integrated GPU. As is the trend where computers are expected to be compact and portable, this has opened up new challenges for the developers. One of the most important challenges is the heat generated inside the integrated systems.

A GPU consists of three major components- base, processor and fins. The base is made up of fiberglass in which the circuits of transistors and capacitors are integrated. The processor which is the core device pumping heat into the system as it runs is also integrated with the base. The processor is made up of several materials but the major contribution is accounted by silicon material. The fins integrated on top of the processor are designed to conduct heat from the processor and dissipate it into the incoming air flow through convection. Hence, a suitable material for the fins is aluminum alloys. Air flow is induced into the system using a blower and the power output of this blower is crucial for the sustenance of the graphic processor unit.

With the advances in GPU, a commercial grade unit such as NVIDIA GeForce GTX 1080 has processing speed of 10 Gbps at 1.7 GHz. It is to note that the dimensions of GPU are 4.5"x10.5". Due to this small volume, the heat flux generated is very large. This is quantified by defining the Thermal design power. TDP for a GTX 1080 is 180 W. Due to the optimum working temperature range of the materials such as silicon in processors being less than 85 ^oC, the cooling system has to pump air at room temperature to dissipate the enormous amount of heat generated.

A physical prototype to test the effectiveness of the cooling system can be considered an unnecessary expenditure of time and money. Hence, more and more manufacturers rely heavily on using computational fluid dynamics software packages. As the simulation involves heat transfer between a solid and fluid at interface, the analysis is termed conjugate heat transfer (CHT). The given study performs a CHT analysis on a basic GPU unit. The study shows the importance of inlet air velocity in effective heat transfer in the case and determines the maximum temperature attained by the processor.

2. Methodology

The following steps are carried out in the given study:

1. The CAD model of the graphic card along with enclosure is imported in Spaceclaim to clean the design and share topology amongst the components using SharePrep feature command.
2. The design model is meshed with baseline mesh and solved for three different inlet fluid (air) velocities- 1 m/s
3. The setup involves assigning appropriate material selection for the different components, and setting up the heat source in the processor solid component
4. The solution is run for steady state, pressure based solver and the temperature characteristics are studied.
5. Based on the observations, the study is extended for refinement of the mesh and the results are compared with baseline mesh
6. The refined mesh is used to determine temperature characteristics when inlet air velocity is 1 m/s, 3 m/s and 5 m/s
7. The study concludes by determining the maximum temperature of the processor and the hotspots generated in the graphic card unit.

3. Pre-processing and solver settings

3.1 Geometry

Fig 1- Geometry of graphic card

Fig 2- Components (a) fins, (b) processor and (c) base

Dimensions of processor = 0.008 x 0.008 x 0.001 m

Using Shareprep tool, the common edges and faces of the components are shared and interference check is performed to verify no conflicting bodies are present in the model. This results in conformal mesh generation.

3.2 Mesh generation

For baseline mesh, the element size of 10 mm is considered and the mesh is generated as shown in figure

Element size = 1e-002 m

Fig 3 -Mesh

Fig 4 – Mesh metrics

3.3 Setup:

Type: Pressure-based

Time: Steady State

Viscous model: K-Ɛ turbulence model

3.3.1Material selection

Using the general composition of parts of a GPU, we define:

Fins- Aluminum

Base- FR-4 Fiberglass

Processor- Silicon

As Si and FR-4 are not available in Fluent Library, user-defined library created with below properties:

Fig 5- User-defined material- Fr-4

Fig 6- User-defined material- Silicon

For fluid, we assign material: air.

After defining the materials, respective cell zones are allocated in Fluent.

3.3.2 Processor heat source

For the component-processor, an energy source term is defined with heat flux calculation as follows:

High-end processor TDP = 5 W

Volume of processor= 64e-009 m³

Heat flux = 7.8125 e007 W/m³

The calculated heat flux is allocated to the processor cell zone.

3.4 Boundary conditions

Inlet air velocity = 1 m/s

Outlet: pressure-outlet

Sidewall type: Symmetry

Bottom wall type: Wall

4. Results

The case is run for 200 iterations and the results are noted.

4.1 Convergence

Fig 7- Plot of residuals vs iterations

From the plot of residuals against the iterations, it can observed that the solution attains convergence at iteration= 150.

4.2 Temperature characteristics

The solution is imported to CFD-Post and the temperature characteristics are plotted.

Fig 8- Temperature contour plot

The contour plot highlights the temperature distribution in the different components of GPU in a global range. To observe the effect of processor heat flux on the fins and base components, the contour plot along and across the unit are plotted.

Fig 9- Temperature contour plot in axial plane

Fig 10- Temperature contour plot in cross sectional plane

It can be observed from the figure that heat from the processor is conducted into the fins. Due to the air flow, heat convection occurs at the fins and the temperature drops at the fin surfaces. At the base, the heat flux from the processor increases the temperature at the contact surface. However, due to the low conductivity of base (FR-4 material), the temperature distribution is such that there is a hotspot generated at the interface of processor and base. Away from the hotspot, there is gradual decrease in the temperature across the surface of the base component.

However, due to the coarse mesh, the temperature distribution inside the solid components and the effect of air cooling at the edge is not captured with much accuracy.

4.3 Air flow vector plot

Fig 11- Vector plot along the air flow

To observe the air flow velocity across the unit, we use vector plot. It can be observed that the incoming air at 1 m/s underdoes recirculation at the wake region of the fins. This wake region decreases the linear velocity components. The physical significance of this wake is the decrease in heat convection at regions behind the fins. The air at a higher temperature carries less heat and the recirculation may result in heating of capacitors and transistors. A solution for this is the design optimization of fins to delay flow separations and place minimal components after the fins along the direction of air flow.

It can be noted that the wake region is not uniform and consistent. This is due to the coarse mesh generated. Thus, the study proceeds further to refine the mesh and address the two major drawbacks of coarse mesh:

1. Unsatisfactory temperature profile at the interfaces and edges
2. Inaccurate wake regions and velocity recirculation.

5. Mesh refinement

The mesh is refined by reducing the element size from 10 mm to 5 mm. A comparison of the refined mesh with the baseline mesh is shown in figure.

Fig 12- Mesh generated for fins (a) baseline (b) refined

Fig 13- Mesh generated for processor (a) baseline (b) refined

Fig 14- Mesh generated for base (a) baseline (b) refined

Fig 15- Mesh generated for enclosure body (a) baseline (b) refined

Fig 16- Refined mesh metrics

From the metrics, the refined mesh can be accepted for simulation of the given case.

5.1Comparison of results of refined mesh

The temperature contours and air flow vector plot are generated for a flow velocity of 1 m/s.

Fig 17- Temperature contour along axial plane

Fig 18- Temperature contour along cross sectional plane

Fig 19- Vector plot along air flow

It can be noted from the contour plots that the refinement of mesh has resulted effective capture of the wake region and the temperature gradient at the edges and interfaces are refined. Thus, the refined mesh can be used to conduct further simulation study.

6. Effects of variation in flow velocity over the graphic card unit

The study is extended to calculate the temperature characteristics for air velocity 3 m/s and 5 m/s. The results obtained are compared to study the effect on heat dissipation.

6.1 Convergence

Fig 20- Plot of residuals against iteration

It can be noted that the convergence of solution is obtained for the highest velocity at iteration = 225 whereas the solution coverges at iteration = 450 for the lowest velocity of 1m/s

6.2 Temperature contour of GPU

Fig 21- Total temperature contour plot for the three cases

A brief comparison of the temperature contour plots for the three different velocities suggest a decrease in the maximum temperature of the unit. This is due to improved forced convection of heat from the graphic card interface into the fluid domain.

It can also be observed that for lower velocities, the heat transfer between the fins and air is significant in regions close to the frontal area. As the velocity increases, this heat transfer is spread throughout the fin surfaces and the temperature of the system is bound to decrease.

6.3 Velocity vector plots

Fig 22- Vector plot along air flow for the three cases

A brief study of the velocity vector plots indicate that the wake region formed when varying the velocity from 1 m/s to 5 m/s is similar.

6.4 Maximum temperature of processor

The maximum temperature of the processor for the three cases is tabulated below:

Table 1- Comparison of results for three different air flow velocities

Air flow velocity in m/s	1	3	5
Convergence iteration	450	450	225
Maximum temperature of processor (in K)	430.59	364.49	347.5
Maximum velocity of air (m/s)	1.25	3.72	6.2

It can be noted that with an increase air flow velocity at the inlet, there is a decrease in the maximum temperature of the processor. Considering the maximum allowable temperature range for the processor to be less than 95^oC (393 k), the blower selected for cooling system must generate air flow velocity of 3m/s.

6.5 Heat transfer coefficients

Fig 23- Wall heat transfer coefficient contour plot for (a) GPU (b) Processor

The figure shows the wall heat transfer coefficient contour plot for the GPU and the processor. From the plot, we can observe the maximum wall heat transfer coefficient to be 576.4 W/m².K.

6.6 Temperature hotspots

The temperature contour of the individual components (for a particular case of velocity= 5 m/s) is studied to find the various hotspots in the given case.

Fig 24- Temperature contour for components- base, fins and processor

From the figure, it can be seen that temperature hotspots can be found in the individual components. This must be addressed during the design stage. In fins, the temperature hotspot can be seen at the point of contact with processor. Thus, surface area of fins need to be maximum at this region to enhance forced convection by air and reduce the temperature.

In case of base, the low conductivity of fiberglass results in temperature hotspot at region in contact with processor. Thus, the electronic components such as transistors and capacitors should not be placed in this region of the base.

7. Conclusion

The study successfully performs a CHT analysis on a graphic card unit to determine the maximum temperature achieved by the processor for different air flow velocities. Using this, the optimum velocity for sustainable design is concluded. The study further discusses on the temperature contour and hotspots generated. Also, vector plots are generated to determine the wake region and the recirculation of air flow.