Transcript Testbed Thrust Overview

The Evaluation and Development of an Efficient Cooling
System for High Performance Computing Applications
Raaghul Senthilkumar and Ronik Sheth
CURENT, University of Tennessee
Final Presentation
July 16, 2014
Knoxville, Tennessee
Presentation Outline
• Research Goal
• Background
• Methods
• Results
• Discussion
2
Research Goal
• Early models of Microsoft’s Xbox 360
suffered from such problems stemming
from an improperly designed cooling
scheme.
• The research goal was to evaluate the
cooling system of an Xbox 360 console
and to determine what caused the
console to fail.
3
Background
• Basic Understanding of Hardware
• Thermal Management
• Introduction to Red Ring of Death
4
Basics of Computer Hardware
CPU
GPU
5
Integrated Circuits
• A integrated circuit consists of
millions of transistors
• Transistors have two possible
states
6
Central Processing Unit (CPU)
•Xenon processor
Manufactured by IBM
Single Die
•Multi-Core Processor
Three Cores
Two Threads
•90 nm Process
7
Graphics Processing Units (GPU)
• Xenos Processor

Manufactured by AMD

Two Dies

eDRAM

Unified Shader Architecture
• 90 nm Process
• 48 Unified Shaders

Combined Pixel and Vertex Shaders

More Efficient
8
Thermal Management
• Passive Cooling
 Air Cooling
 Heat Sink
• Active Cooling
 Fans
 Liquid Cooling
9
Heat Sinks
• Passive Cooling
• Transfers thermal waste to
the air
• Heat Pipes
• Surface Area
Figure 1: Heat Sinks
Upper Left: 1st Gen CPU; Upper Right: 2nd Gen CPU;
Bottom Left: 1st Gen GPU; Bottom Right: 2nd Gen GPU
10
Fans
• Active Cooling
• Air Cooling
11
Intro To Red Ring of Death
• The infamous Red Ring of Death was
caused by flaws in the design of the
cooling scheme.
• Overheating weakened the soldering
connections underneath the chips
12
Methodology
• Different combinations of the heat sinks were then implemented to see
which combination keeps the Xbox 360 running without failure.
• Two different cooling fans were also used to see which fan kept
temperatures low.
• Thermal sensors were used to measure the temperature of the integrated
circuits.
13
Results
Heat of the Components Over Time with Standard Fans
T
e
m
p
e
r
a
t
u
r
e
60
Ends with Red
Ring from
Overheating
50
40
30
2nd Generation GPU
20
10
1st Generation CPU
(˚C)
0
0
0.5
1
3
3.5
Time (minutes)
14
Results
Heat of the CPU over Time with Improved Fans
1st Generation
T
e
m
p
e
r
a
t
u
r
e
2nd Generation
45
Ends with Red Ring
from Overheating
40
35
30
Plateau
25
(˚C)
20
0
0.5
1
5
8.5
10
Time (minutes)
15
Results
Heat of the GPU over Time with Improved Fans
1st Generation
T
e
m
p
e
r
a
t
u
r
e
2nd Generation
70
2nd generation
dissipates heat
better
60
50
40
Plateau
30
20
10
(˚C)
0
0
0.5
1
5
7
10
15
20
Time (minutes)
16
Discussions of Thermal Management
• The Xbox 360 manages thermal waste through a system of passive and
active cooling.
• When designing an effective cooling scheme, the implementation of an
efficient ventilation system is necessary. Ventilation enables the heat to be
transferred out from the device.
17
Results
Operating Loss (in millions)
1400
1200
$1.06 billion
warranty
extension
1000
800
600
Operating Loss (in millions)
400
200
0
FY06Q2 FY06Q3 FY06Q4 FY07Q1 FY07Q2 FY07Q3 FY07Q4
Fiscal Quarter
18
Discussion of Economics
• The liabilities stemming from the repairs of Xbox 360 would lower the
profitability of the Xbox 360.
• A properly designed cooling scheme increases would have allowed for
more powerful chips.
• By shortening the “Time to Market”, they were able to reach the market
before the Playstation 3 and keep the cost of development lower.
19
References
• http://www.dailytech.com/New+Xbox+360+Cooling+Hardware+Found+in+
Europe/article7667.htm
• http://www.dailytech.com/Microsoft+Insider+Xbox+360+Team+Knowingly+
Launched+Faulty+Hardware/article10434.htm
• http://www.ibtimes.com/ps4-vs-xbox-one-ps4-9-million-sold-microsoftresponds-sonys-success-1622396
• http://www.microsoft.com/investor/EarningsAndFinancials/Earnings/Press
ReleaseAndWebcast/FY14/Q3/default.aspx
• http://techreport.com/news/13974/microsoft-insider-xbox-360-was-rushedto-market
• http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=1624324&q
ueryText%3Dxbox+360
• Jeff
20
Acknowledgements
Thanks to Terence Randall, Chris Crowder, and Dr. Kamrul Islam
This work was supported primarily by the Engineering Research Center
Program of the National Science Foundation and the Department of Energy
under NSF Award Number EEC-1041877 and the CURENT Industry Partnership
Program.
21
22