Week 6 - CHT Analysis on a Graphics card

Aim: To perform CHT analysis on graphic card.

Objectives:

1) To learn the procedure for CHT analysis in ANSYS Fluent

2) To understand various meshing concepts and methods

3) To infer various physical parameters from the simulation

Introduction:

     A CHT (Conjugate Heat Transfer) analysis is useful where there are multiple modes of heat transfer in our computational domain, like conduction and convection. Having such a great number of applications for this type of simulation makes it important to be studied in depth.

Assumptions:

1) The air flow profile from inlet is considered to be of uniform value, instead of a profile of different values.

2) Air flow resistance due to cabinet's back mesh is considered to be 0.

3) The heat producing chip is small in size compared to real life graphic cards, so the heat generation capacity is reduced accordingly.

4) The physical constraints to the flow due to side walls are considered to be zero. (The temperature gradient along side walls is found to be negligible, so they are replaced with symmetry boundary conditions)

Input calculations:

1) Inlet velocity calculations:

According to Ref[1], the inlet velocity of fan reached to a maximum of 2 m/s in experimental study. So, the inlet velocity here is kept to be 2 m/s.

2) Heat generation:

An basic 2Gb graphic card will consume 75 W of energy and 99.9% of it will be given out in heat (Ref[2]). It will have the GPU of 40 x 40 x 2 mm. Accordingly,

75/(0.04 x 0.04 x 0.002) =  23437500 W/m^3 = 2.3437e7 W/m^3

The GPU chip used here is 8 x 8 x 1 mm, but we can keep the heat generation capacity as same, so the above value is used for calculations.

 

Materials & their properties: (Ref [3],[4])

1) Fins --> Aluminium

Standard aluminium is available in Fluent database.

Density = 2719 Kg/m^3

Heat capacity = 871 J/KgK

Thermal conductivity = 202.4 W/mK

 

2) Processer --> Silicon

Since the silicon is not available in Fluent's solid database, I had to define it

Density = 2000 Kg/m^3

Heat capacity = 710 J/KgK

Thermal conductivity = 150 W/mK

 

3) PCB --> FR4 (glass-reinforced epoxy laminate material)

Since the FR4 is not available in Fluent's solid database, I had to define it

Density = 1850 Kg/m^3

Heat capacity = 950 J/KgK

Thermal conductivity = 0.29 W/mK

 

4) Fluid --> Air

Standard air is available in Fluent database.

Density = 1.225 Kg/m^3

Heat capacity = 1006.43 J/KgK

Thermal conductivity = 0.0242 W/mK

Viscosity = 1.7894e-05 Kg/m-s

 

This report has 3 cases of meshing:

(The geometry and setup is discussed only in case 1 as it remains same for other cases)

 

Case 1:

1) Geometry

The enclosure and graphic dimensions are:

After importing such geometry, the shared topology is enabled in fluent and by using shareprep function, a conformal geometry is made.

2) Mesh settings:

Global mesh setting of element size = 10 mm

1) PCB body sizing
element size = 1 mm
capture curvature - yes, 18 deg

2) Processor body sizing
element size = 1 mm
capture curvature - no

3) fins body sizing
element size = 1 mm
capture curvature - no

Global number of elements = 201920

3) Setup:

Steady state analysis

Solver - pressure based

Turbulence model --> realizable k-e model

Inlet velocity = 2 m/s

Heat generation rate = 2.3437e7 W/m^3

Side walls --> symmetry

Number of iterations = 500

The solution converged around 325 iterations.

 

4) Results:

(It might appear in some images that contours are plotted on interior, but pay attention if its a shadow or a real interior volume. There seemed to some problem with naming of interiors and walls but the end results are plotted where they should be.)

1) Temperature contour - Global

 

 

The max temperature reached by system is 53.175 Deg C. The PCB doesn't conduct the heat to a great extent so it remains around 33 Deg C. Also, there appears to be a local hotspot in centre of the fin base.

  Temperature contour - Local ;

It can be seen that fins at the front are cooler than the ones behind. This happens because the cold air first hits the front fins, it gets heated up and goes to next fin. Hence, the next fins can't give it enough heat as difference in temperature is lower in between solid and fluid

The outer fins are more cooler than inner fins. This is due to velocity of air passing from sides is more than that of centre.

 Temperature Streamline ;

 Temperature volume rendering ;

 Temperature contour on a plane at y = 0

Processor temperature ;

The temperature is highest at the region little back from centre. This is due to more cooling from the front fins.

A potential hotspot is also present below processor, on the PCB. So one must avoid placing components on that place.

 

 

Velocity contour

The flow has a dead (wake) region behind the fins. It should be seen correctly through a vector plot.

 Velocity vector plot

There are 3 recirculation regions fromed and their velocity is less than 0.3 m/s. So, it is important to look at flow over this region using streamlines.

Temperature - Streamline - recirculation region

It is seen that the components behind the fin plate receive a high temperature(38 Deg C) air with low velocity (<0.3 m/s). If any component generates heat over that region, it would not cool down faster. If a component doesn't generate heat, the air would heat that component. So, in either way the heat will stay in that region, causing problems to the components present there. So, the processor must be placed at last in the flow direction.

 Heat flux plot

1) The heat flux seems to be more from front end. This is due to higher temperature difference in solid and fluid. As the air moves forward, it takes less heat, so the region appears blue.

 2) The heat flux distribution over a fin is interesting. The corners and centres have a higher contact with air, so they give up more heat.

 

Heat flux - processor

The heat flux is shown around the walls of processor. It is more from the top because the fins take more heat than the pcb plate.

Wall heat transfer Coefficient (WHTC)

The distribution seems to averaged to a light blue region but there are local high spots at fin ends and processor slot. This error might be due to mesh at that location. To find average coefficient, we use function calculator:

Reason for the high heat transfer coefficient:

A good mesh is said to have element quality near 1. Here, the fins and fin ends have quality < 0.4 and in some places <0.2. This must have caused the high WHTC values.

Also, a good mesh must have skewness near to 0 and must have average skewness around 0.33.

In current case, average is near to 0.5 and fins have higher than 0.6. In fin ends the skewness is near to 0.9. So the mesh needs to be more fine.

Reasons to refine the mesh:

1) The recirculation region is not clearly visible

2) The WHTC has high values at some locations.

3) The element quality and skewness is not good.

 

Case 2:

1) Mesh:

Global mesh setting of element size = 5 mm

1) PCB body sizing
element size = 1 mm
capture curvature - yes, 18 deg

2) Proc body sizing
element size = 0.5 mm
capture curvature - no

3) Fins body sizing
element size = 0.5 mm
capture curvature - no

Global number of elements = 422289

Results:

(The results obtained have similar explanations, so the results given below won't have exclusive reasons accompanied)

1) Residuals:

Temperature plots:

Velocity Plots:

The streamlines show the formation of two giant recirculation regions, formed due to air flowing from sides to middle of fin structure's back.

 

Heat flux;

Wall heat transfer coefficient;

 

Case 3:

1) Geometry:

In order to refine the mesh and keep the element numbers to a minimum, an enclosure is made surrounding the graphics card. An important step while doing this is selecting all three bodies(fins, processor, PCB) and then applying enclosure feature in ANSYS Fluent. The outer enclosure is deleted and then created again by selecting inner enclosure. The above given image gives information of dimensions needed to make the enclosures. The dimensions are used based on the lengths of volume surrounding graphics card where there is a gradient in temperature.

 

2) Meshing:

Global controls:
     Element size = 5 mm
     Capture proximity = yes
     Proximity size = 0.1 mm

1) Multizone - outer enclosure

2) Body sizing - inner enclosure
    Element size = 4 mm
    Capture curvature - yes

3) Body sizing = fin body
    Element size = 0.4 mm
    Behaviour - hard
    Capture curvature - no

    Capture proximity - no

4) Body sizing - processor body
    Element size = 0.4 mm
    Behaviour - hard
    Capture curvature - no
    Capture proximity - no

5) Body sizing - PCB
    Element size = 2 mm
    Capture curvature - yes
    Capture proximity - no

6) Pinch - on that edge
    tolerance = 0.1 mm

Element number = 1852356

The pinch control is applied on the edge where the WHTC increases suddenly. It acts like a virtual topology. Even after we apply pinch control, the geometry doesn't change but the mesh changes. The reason for using this:

The gap between fin and base's side is 0.1 mm. In order to resolve the length that small using element size, we would need element size = 0.05 mm. Doing so would increase the element number greater than 10 million. Other option is to locally change the size using sphere of influence method, but after doing so the high WHTC phenomenon spread to whole region of influence (tried before). So it became important to ignore the gap.

Ignoring the gap might seem a bit wrong but it isn't because the temperature, temperature gradient, heat flux, velocity plots were uniform in that region without even using pinch control.

For verifying whether the pinch control works correct, we have applied pinch control only on the top surface. The bottom surface has some gaps along the edges where the processor fits. There no pinch control is applied.

The pinched element has quality more than 0.9.

 

 Results:

Residuals;

The solution converged at around 425 iterations.

Temperature plots:

 

Here we can observe 5 distinct recirculation zones with their temperature. The largest one has temperature of 40 Deg C.

In this plot we can see the streamlines entering from front and getting heated gradually.

A potential hotspot:

Velocity plots:

 Wall heat flux;

 Wall Heat Transfer Coefficient:

 

As expected, the pinch control worked and we got uniformity on top surface but the bottom surface didn't have a smooth transition.

Final Comparison:

The trend can be observed using values from the table:

Conclusion:

1) The CHT analysis is done for graphic card

2) Effect of various mesh parameters on simulation result is studied.

3) Use of intermediate meshing method is studied.

 

CHT analysis on graphic card animation ;

https://youtu.be/ScgZIQn_iBo