CHT analysis for a Graphics Card with heat single source

Project – 1 CHT Analysis on a Graphics Card.

Objective: To perform Conjugate Heat transfer Analysis on a graphics card which consists 1.GPU which is the only heat source and is in contact with the 2.cooling fins which loose the heat to incoming air stream via convection, 3.other electronics components and a PCB on which everything is soldered.

Geometry Setup:

• Instead of suppressing geometry for all the components inside the enclosure all the bodies are kept as it is as there are more than one bodies with different material properties. Using the share topology option it is made sure that the meshing can be continuous.

Analysis Setup:

• An enclosure is used to limit the domain. It represents a wind tunnel in which the card is placed for the analysis. The GPU material is assumed as gold while the fins are to be of copper as it conducts heat better. All other components as well as PCB is considered to have the material FR-4 which is more of a grade of materials which are dielectric glass epoxy. PCBs or Printed Circuit boards have circuits etched in the designed circuit diagram and then are doped with conductive material such as copper. The boards are naturally insulators and are widely used in electronic devices.
• The incoming air has a temperature of 15 degree Celsius and a velocity of 7 m/s.
• As the GPU is very small at 64mm3 by volume compared to generally available ones which are 300-400 mm3 we only consider 1/5th the power they consume. A normal high end GPU draws anywhere between 40 – 60 Watts for our case we have considered a power consumption of 9.6 Watts. Hence, the Energy generated with no losses comes out : 9.6 / 64*10-9 = 1.5e8 W/m3 .
• The GPU is the only energy source for the case as a result both conduction of heat from the GPU to both cooling fins and the PCB is expected. The comparatively cooler incoming air takes some of the heat away via Convection, cooling down the components keeping them within working conditions. The whole system is generally programmed to monitor this temperature and shut down if the temperatures reach the higher limits to prevent any sort of permanent damage to the GPU and other necessary components.
• Generally a fan or any other sort of cooling system gets turned on to keep things under control as these high end GPUs are susceptible to working at higher temperatures as the GPUs have millions of transistors performing millions of calculations at a time. It is normal for a modern day GPU to reach temperatures upto 130 degree Celsius during intensive operations such as high quality rendering. Efficient and effective cooling is necessary for the long life of the component as well as prolonged intensive usage.

Case 1 – Baseline Mesh:

• To start with the analysis the most basic mesh is used as a part of grid dependency test to get an idea of what kind of results are to be expected with minimal complexity. The element/cell size is gradually decreased there after till no significant difference in results is noticed.
• For this case an element size of 10mm is selected with default options. About 86k cells are formed and cover the domain.

Iterations:

• The solutions converge at around 125 iterations and the residuals start reaching steady state. As the number of elements/cells is pretty small the solutions converges quickly as there are fewer equations to solve.

Temperature Distribusion Plot:

• A lot of dispersion can be seen specially around the graphics card because of the coarse mesh. Maximum recorded Temperature is 396.02 Kelvin at the Processor. The cooling by convection can be seen as the incoming air steam absorbs some heat energy and gets hotter which is portrayed in the wake of the fins. Majority of heat is conducted to the fins as the coefficient of conductivity for copper is higher than aluminium.

https://drive.google.com/open?id=1pzKURVhQUUfpZRZDd8dulNXWs_Kelhwx

Velocity and Pressure plot:

• As per Bernoulli’s principle high velocity low pressure can be seen over the Graphics card as the flow gets converged past the separated boundary layer only to diverge afterwards.
• Low pressure zones are seen behind all sorts of obstructions such as the fins and electronic components as vortices are formed there due to boundary layer separation at sharp corners.

https://drive.google.com/open?id=1686uZmTKCoI2ExZA2PyBpCdTFGgG-ehu

Wall Heat Transfer Coefficient:

• Functioning of the fins can be seen in this plot as almost 80% of the fin surface transfers heat to the air stream via convection at a peak value of 1791 W/m2k. The dispersion persists here and the boundaries are not well defined.

Case 2 – Refined Mesh:

• For this case the mesh is refined using sizing option on the graphics card with element size of 1mm for better connectivity and quality at the boundaries. The overall mesh is made finer with smaller element size of 1.65mm. As the fins do not have much width the inflation option cannot be used without ruining the mesh. The local sizing helps with better coverage and interpolation of results at the boundary layer around the parts such as fins, electronic components etc. To cope with the student license limitations the element/cells size is controlled by the element size at 503501 elements.

Iterations:

• As there are close to half a million cells, a lot of equations are needed to be solved namely 7 per cell for a steady state analysis hence it takes a lot of time for the limited computation power to reach convergence. A steady state is reached at about 200 iterations and the solution can be assumed to have converged.

Temperature Plot:

• The conjugate heat energy transfer can be seen in this case very clearly. The processor conducts a lot of heat to both fins and the PCB which can be seen the hot zone near the GPU where the maximum temperature of 370.24 Kelvin is recorded. The cool incoming air takes some of the heat away from the PCB which can be seen by the comparatively cooler parts of the PCB.
• The highly conductive Fin material absorbs the heat from the GPU and looses it to the atmospheric air in our case the incoming stream by convection. This keeps the rest of the components as well as the GPU itself from overheating and getting damaged as a result. The heat energy absorbed by the air is portrayed by the hot streams in the wake of the model.

https://drive.google.com/open?id=1gurD8xJ4omfdSAHxLJj6-t7IO5NnL6Wb

Velocity and Pressure Plot:

• A very clear representation of the plots can be seen in this case as a result of finer mesh. The boundaries are well defined and the High and low intensity zones are visible. The Bernoulli’s principle can be seen in work here as the stagnation points have the peak pressure while the high velocity flow over the model has lower pressure.
• Boundary layers and their velocities can be seen and distinguished.

https://drive.google.com/open?id=1iJGZqnsI3JVkweDSlU0fsFX3YZyJjCma

Streamlines:

• Formation of vortices and the boundary layers can be seen better with streamlines. The pressure difference due to stagnation of the fluid on one edge and the accelerated flow over corners and sharp edges is usually the reason why the vortices are formed. The turbulence helps with the heat transfer and is beneficial in our case.
• The first point of contact are the front most edged of the PCB which forms a rectangle can be seen to generate two opposite spin vortices at the left and right end. The same happens for the fins and eventually even bigger vortices are created. The widest part of the component is behind the fins where the flow is already disturbed and the vortices eventually collapse as they loose all the energy.

Wall Heat Transfer Coefficient:

• A more realistic figure of 536.5400 W/m2k is recorded for this case which can be seen mostly at the fin boundaries.

Conclusion:

• A stark difference in the results is noticed between the two cases as the cooling is a surface phenomenon for this case - Convection. The poor coverage of fin surface and as a result the neighbouring domain affects the results for the 1st case where the maximum temperature figures as well as CHT figures are greatly exaggerated.

Case 1: Maximum temperature: 393 Kelvin with hotspots near the fin-GPU and GPU-PCB interfaces:

Case 2: Maximum temperature: 371 Kelvin with hotspots near the fin-GPU and GPU-PCB interfaces:

• It can be concluded from this analysis that for the assumed inlet stream conditions there isn’t any problem in terms of thermal damages as the cooling effect is more than sufficient. We assumed the materials which are not very realistic in the case of the GPU, Which are silicon chip with millions of transistors etched on them and generally do not have the conductivity as high as gold. Even with a highly conductive material representing GPU and a very high energy consumption of 9.6 Watts for a very tiny chip, the maximum temperature recorded was 370 Kelvin, which as seen in the temperature plot is recorded at the GPU only. The fins help limiting the heat energy transfer by loosing it to the incoming air. This keeps the temperatures down at other electronic components and prevents any kind of damage to them.
• As mentioned before High end GPUs are designed to work and endure temperatures as high as 120 degree Celsius with over-clocking and are accompanied with adequate cooling which is monitored and controlled by the CPU. For our case the usage of copper cooling fins is very effective and efficient than other methods such as liquid cooling. To make the cooling even more effective utilisation of an GPU fan or more than one fans with smaller sizes can give even better results and is advisable.

Drive link to folder with all the files:

https://drive.google.com/open?id=1bi1ODoRBKX7Ew8FxAljHPP_jE6HXcg-x

PPT Report:

https://drive.google.com/open?id=1J-V0R8Kxs0esInzkxw2eCcnfeikZGyID