3. Introduction
• Background
• Average Power Usage Effectiveness (PUE)
• Average PUE of 2.9 for U.S. servers in 2013
• Power required for cooling is a major
concern
• Research on how to improve heat sink
performance
• Motivation
• Improve heat sink thermal efficiency
• Decrease PCH operating temperature
3Erik Jacobs
5. Procedure
1. Develop server model
2. Incorporate model into ANSYS Icepak
3. Experimental testing
4. Perform Computational Fluid Dynamics (CFD) simulation
5. Compare simulation & experimental results
5Erik Jacobs
6. Previous work
• Solid model built in Solidworks, imported to Icepak
• Parameter-based testing
• CFD simulations with conformal meshing
6Erik Jacobs
Figure 1.4: Solid model of server Figure 1.5: Area surrounding PCH
(Hard drive not shown)
PCH Location
8. Experimental testing
• Test was run on the Winterfell 2OU hybrid cooling server
8
External radiator fan and
bread board
Server Power Supply
Winterfell 2OU Hybrid
Cooled Server
Sample Heat sinks and
TIM Injector
Figure 2.1: Experimental
Test Setup
Adam McAvene
9. Icepak Model
9
Figure 2.2: Isometric View of CAD-
constructed CFD model
Figure 2.3: Isometric View of Icepak-
native CFD model
Adam McAvene
10. Meshing
10
• Non-conformal meshing implemented
• Reduce “mesh bleeding”
• Refine specific object without affecting other areas
• Significantly reduced testing time
Figure 2.4: Example of Conformal mesh of
thin objects
Figure 2.5: Example of Non-conformal mesh
of thin objectsAdam McAvene
12. Miniature Partitions
• 55 x 8 x 1 and 19 x 8 x 1 mm^3
• Redirect airflow leaving the heat sink
12
Figure 2.10: Square Pin Heat Sink With Mini
Partitions
PCH Located Underneath
Heat Sink
Binh Tran
13. 13Binh Tran
Without Mini Partitions
13
• Flow leaves from gaps between square pins
• Air does not travel (and convect) through entire length of heat sink
Figure 2.11: Cross-Sections of Heat Sink/PCH
Temperature Without Partitions
14. 14
With Mini Partitions
• Flow exit regions become much smaller
• Increased convection near exit regions
Figure 2.12: Cross-Sections of Heat Sink/PCH
Temperature With Partitions
Binh Tran
18. 18
Simulation Results vs. Experiment Results
• Mean PCH Temperature (100%
CPU Utilization)
• Simulation: 61.86 ± 1.47
• Experimental : 61.83
• Worst case % Error: 2.4%
55.04
60.13
61.87
62.01
61.83
64.05
50
52
54
56
58
60
62
64
66
68
70
MeanPCHTemperature(°C)
CPU Utilization %
PCH
Figure 3.1: Extruded Fin Experimental results
for different power utilizationsRyan Hart
19. Baseline Test Results
19
• Cross-Cut fin lowest
temperature for base
heat sinks
• Extruded fin, elliptical,
and circular pins have
roughly same range of
temperature
Figure 3.2: Maximum PCH Temperature for
Heat Sinks without Modifications
Ryan Hart
20. Mini Partition Test Results
20
• Lowers cross-cut fin,
elliptical and square pin
temperatures
• High reduction in square pin
PCH temperature
• Negligible or
detrimental effect on
hexagonal and circular
pin fins
• Possible interference with
flow mixing
Figure 3.3: Maximum PCH Temperature for
Heat Sinks Mini PartitionsRyan Hart
21. Vapor Chamber Test Results
21
• Lower PCH
temperatures on all
designs
• Heat spreading increases
potential convection
• Higher temperature
drop in cross-cut,
extruded fin and
circular pin heat sinks
• greater effect for
denser heat sinks
Figure 3.4: Maximum PCH Temperature for
Heat Sinks with Vapor Chamber basesRyan Hart
22. Partitions + Vapor Chamber
22
Figure 3.5: Maximum PCH Temperature for Heat
Sinks with both Mini Partitions and Vapor Chambers
Ryan Hart
24. Final Design
24Damone Norwood
• Cross-Cut Fin
• Thermal Improvement
• No Modification- 2°C
• 2.9% Temp. decrease
• VC + Mini Partition - 4.3°C
• 6.2% Temp. decrease
Figure 3.6: Final Heat sink Design
25. Summary
• Modeled server in Icepak
• Implemented non-conformal meshing
• Experimental Testing to validate CFD model within 5% error
• Developed final design & server improvements
• PCH temperature reduction of 6%
25
Damone Norwood
27. Works Cited
27
[9] Patil, Dhanraj Arun. “CFD Modeling and Parametric Study of
Vapor Chambers as Heat
Spreaders for High-Power Electronic Devices” Master’s
Dissertation, Department of Mechanical Engineering,
University of Texas at Arlington, Arlington, TX, 2015
28. Further Work
• Additional Physical Model Testing
• Recirculating Flow
• Ducting and Vapor Chamber
• Further Design Considerations
• Quantify power savings from cooling
• Both (internal fan and coolant)
26Damone Norwood
Figure 3.3: Winterfell 2OU Hybrid Cooling
Server Rack
Editor's Notes
The primary objective of this semester was to improve upon the design of the heat sink from our previous semester, as well as investigate other methods of improving heat transfer
The PCH is a computer chip that controls data paths and support functions used with Intel CPUs
Our heat sink is located directly on top of the PCH
Since our project is simulation based, we had to recreate the physical server as accurately as possible to a digital model. So to build the 3D model, we used a set of calipers and measured each of the components that were deemed necessary. We then sketched the parts with these dimensions in solidworks so that the assembled server model could be imported into icepak to start running simulation tests. The main qualifications for a component to be critically necessary for our simulation, is how they would affect the temperature of the flowing air and how they would restrict affect fluid flow around the heat sink
the results we found from our new round of CFD testing would be tested against an experimental test of the server.
we rebuilt our CFD model from the previous semester to run more efficiently, and accommodate iterative changes to the heat sink.
In the previous semester, all components with the exception of the internal fan, the PCH and the heat sink were modeled in CAD software and imported to Icepak as a STEP file . We discovered that the specific geometry of several CAD generated parts created issues with Icepak’s meshing procedure
so for this semester, we replaced nearly all of the CAD generated components with Icepak-native objects. This both simplified the meshing procedure within Icepak and allowed us to properly implement non-conformal meshing
In addition to the extruded fin and circular pin fin design we simulated last semester, we ran simulations on additional heat sink designs this semester. All of these designs are currently manufactured by Alpha Novatech, and are made out of Al-6063. This is an improvement
Make Partitions more visible
In addition to the partition, we also modeled our heat sinks’ bases as vapor chambers. A vapor chamber acts in a similar way to a heat pipe, but can be much thinner and occupy a much smaller cross section.
In our CFD simulation, we adapted our Vapor Chamber model from a prior experiment that focused on Vapor Chamber simulation, and modified the base to fit the same cross section as our heat sink.
The baseline case outperforms both the square pin (S1555) and the hexagonal pin (N54) heat sinks,
had roughly equivalent performance with the elliptical (W54) and our circular pin fin (22x22) design, and
was outperformed by the strip fin (Z54) design.
Separate each bar in graph
Use descriptive names rather than Code names (Square pin rather than S1555)
Vapor Chamber heat sinks decreased the maximum heat sink temperatures for all models of the heat sink. This is due to the heat spreading in the base, letting conducted heat travel faster to the edges of the heat sink.
Better performance in the denser heat sinks, possibly because conduction was a limiting factor in these heat sinks before.
include experimentation on heat sink performance at higher heat fluxes (1 W/cm^2 or more).
In addition, an in depth study of the pressure and volumetric flow characteristics of the server would be useful for improving server performance and validating CFD pressure readings within the server.
Quantify vapor chamber (fan power) savings