Page 1
Mobile CPU's Rise to Power: Quantifying the Impact of Generational Mobile CPU Design
Trends on Performance, Energy, and User Satisfaction
Matthew Halpern Yuhao Zhu Vijay Janapa ReddiDept. of Electrical and Computer Engineering
The University of Texas at Austin 1
Page 5
Snake 2000
Snake Simulator 2016
2
Page 7
Touchscreen
3
Mobile Device
Page 8
Touchscreen Cellular WiFI Bluetooth
3
Mobile Device
Page 9
Touchscreen Cellular WiFI Bluetooth
Camera
3
Mobile Device
Page 10
Touchscreen Cellular WiFI Bluetooth
GPSCamera Sensors Battery
3
Mobile Device
Page 11
System-On-A-Chip
Touchscreen Cellular WiFI Bluetooth
GPSCamera Sensors Battery
3
Mobile Device
Page 12
System-On-A-Chip
4
Page 13
System-On-A-Chip
4
Page 14
System-On-A-Chip
CPU
4
Page 15
System-On-A-Chip
CPU
GPU
4
Page 16
System-On-A-Chip
CPU
GPU
Accelerators
4
Page 17
System-On-A-Chip
CPU
Network-On-ChipGPU
Accelerators
4
Page 18
System-On-A-Chip
CPU
Network-On-ChipGPU
Accelerators
CPU
4
Page 19
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 20
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 21
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 22
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 23
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 24
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 25
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 26
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 27
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
5
Page 28
Nor
mal
ized
ARM
Mar
kets
hare
(%)
0
20
40
60
80
100
Year2009 2010 2011 2012 2013 2014 2015
ARM11 A8 A5 A9 A15 A7 A53 A57
Mobile CPU design is fast-paced.
5
Page 29
Conventional Research Scope
6
Page 30
HardwareSoftware
Conventional Research Scope
6
Page 31
HardwareSoftware
Bottlenecks
Conventional Research Scope
6
Page 32
HardwareSoftware
Performance
Bottlenecks
Conventional Research Scope
6
Page 33
HardwareSoftware
Performance
Bottlenecks
Conventional Research Scope
6
Page 34
HardwareSoftware
Performance
Bottlenecks
Conventional Research Scope
6
Page 35
HardwareSoftware
Perf
orm
ance
Bottlenecks
Expanding the Research Scope
7
Page 36
HardwareSoftware ProcessorApplications
Perf
orm
ance
Bottlenecks
Expanding the Research Scope
7
Page 37
HardwareSoftware ProcessorApplications
Perf
orm
ance
Bottlenecks
Expanding the Research Scope
Conventional Research Scope
7
Page 38
HardwareSoftware ProcessorApplicationsEnd-Users
Perf
orm
ance
Bottlenecks
Expanding the Research Scope
Conventional Research Scope
7
Page 39
HardwareSoftware ProcessorApplicationsEnd-Users
Perf
orm
ance
BottlenecksFe
atur
es
Expanding the Research Scope
Conventional Research Scope
7
Page 40
HardwareSoftware ProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
ce
BottlenecksFe
atur
es
Expanding the Research Scope
Conventional Research Scope
7
Page 41
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
ce
BottlenecksFe
atur
es
Expanding the Research Scope
Conventional Research Scope
7
Page 42
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
ce
Pow
er B
udge
ts
BottlenecksFe
atur
es
Expanding the Research Scope
Conventional Research Scope
7
Page 43
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
cePower Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Expanding the Research Scope
Conventional Research Scope
7
Page 44
Our Scope
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
cePower Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Expanding the Research Scope
7
Page 45
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
8
Page 46
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
1. Has mobile CPU efficiency improved?
8
Page 47
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
1. Has mobile CPU efficiency improved?2. Have mobile CPU advancements
improved end-user satisfaction?
8
Page 48
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
1. Has mobile CPU efficiency improved?2. Have mobile CPU advancements
improved end-user satisfaction?3. How has the rest of the mobile device
evolved around the CPU?8
Page 49
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
9
Page 50
Characterize how the interactions between mobile CPU, end-user, and
mobile device have changed over time through real-world measurement
9
Page 51
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
10
Page 52
Samsung Galaxy
S6
Samsung Galaxy
S5
Samsung Galaxy
S4
Samsung Galaxy
S3
Samsung Galaxy Nexus
Samsung Galaxy
S
2015201420132012201120102009
Motorola Droid
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
10
Page 53
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 54
Year 2009 2010 2011 2012 2013 2014 2015
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 55
Year 2009 2010 2011 2012 2013 2014 2015uArch A8 A8 A9 A9 A15 A15 A57
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 56
Year 2009 2010 2011 2012 2013 2014 2015uArch A8 A8 A9 A9 A15 A15 A57
Process 65 nm 45 nm 32 nm 28 nm 28 nm 28 nm 14 nm
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 57
Year 2009 2010 2011 2012 2013 2014 2015uArch A8 A8 A9 A9 A15 A15 A57
Process 65 nm 45 nm 32 nm 28 nm 28 nm 28 nm 14 nmFreq 0.6 GHz
GHz1 GHz 1.2 GHz
GHGHz1.4 GHz
GHz1.6 GHz
GHz2.1 GHz
GHz2.1 GHz
GHz
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 58
Year 2009 2010 2011 2012 2013 2014 2015uArch A8 A8 A9 A9 A15 A15 A57
Process 65 nm 45 nm 32 nm 28 nm 28 nm 28 nm 14 nmFreq 0.6 GHz
GHz1 GHz 1.2 GHz
GHGHz1.4 GHz
GHz1.6 GHz
GHz2.1 GHz
GHz2.1 GHz
GHzCores 1 1 2 4 4 (+4) 4 (+4) 4 (+4)
11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 59
Year 2009 2010 2011 2012 2013 2014 2015uArch A8 A8 A9 A9 A15 A15 A57
Process 65 nm 45 nm 32 nm 28 nm 28 nm 28 nm 14 nmFreq 0.6 GHz
GHz1 GHz 1.2 GHz
GHGHz1.4 GHz
GHz1.6 GHz
GHz2.1 GHz
GHz2.1 GHz
GHzCores 1 1 2 4 4 (+4) 4 (+4) 4 (+4)L1 I$ 32 KB 32 KB 32 KB 32 KB 32 KB 32 KB 48 KBL1 D$ 32 KB 32 KB 32 KB 32 KB 32 KB 32 KB 32 KBLLC 256 KB 512 KB 1 MB 2 MB 2 MB 2 MB 2 MB
DRAM 256 MB 512 MB MB
1 GB 1 GB 2 GB 2 GB 3 GB11
Capturing Real-world Mobile CPU Trends from Off-the-Shelf Smartphones
Page 60
Has mobile CPU efficiency improved?
12
Page 61
Performance
Power
Energy
Has mobile CPU efficiency improved?
12
Page 62
Substantial Performance Improvements
13
Page 63
Spee
dup
1
3
5
7
9
11
13
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC
Substantial Performance Improvements
13
Page 64
Spee
dup
1
3
5
7
9
11
13
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC
Substantial Performance Improvements
13
Page 65
Spee
dup
1
3
5
7
9
11
13
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC
Substantial Performance Improvements
13
Page 66
Spee
dup
1
3
5
7
9
11
13
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC
Substantial Performance Improvements
What are the key architectural contributors?
13
Page 67
Speedup = IPC Speedup x Clock Speedup
14
Page 68
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Speedup = IPC Speedup x Clock Speedup
14
Page 69
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Speedup = IPC Speedup x Clock Speedup
14
Page 70
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Speedup = IPC Speedup x Clock Speedup
14
Page 71
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
A8
Speedup = IPC Speedup x Clock Speedup
14
Page 72
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
A832-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 73
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue
A832-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 74
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-IssueIn-Order
A832-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 75
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-IssueIn-Order
A8 A932-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 76
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-IssueIn-Order Out-of-Order
A8 A932-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 77
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-IssueIn-Order Out-of-Order
A8 A9 A1532-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 78
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order
A8 A9 A1532-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 79
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order Aggressive Out-of-Order
A8 A9 A1532-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 80
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order Aggressive Out-of-Order
A8 A9 A15
Aggressive Mem Hierarchy
32-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 81
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order Aggressive Out-of-Order
A8 A9 A15 A57
Aggressive Mem Hierarchy
32-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 82
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order Aggressive Out-of-Order
A8 A9 A15 A57
Aggressive Mem Hierarchy
32-bit 64-bit
Speedup = IPC Speedup x Clock Speedup
14
Page 83
Spee
dup
1
2
3
4
5
Year2009 2010 2011 2012 2013 2014 2015
Frequency IPC
Dual-Issue Triple-IssueIn-Order Out-of-Order Aggressive Out-of-Order
A8 A9 A15 A57
Aggressive Mem Hierarchy
32-bit 64-bit
Aggressive core scaling techniques have provided mobile CPUs substantial performance improvements.
Speedup = IPC Speedup x Clock Speedup
14
Page 84
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
Excessive Power Consumption
15
Page 85
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
Excessive Power Consumption
15
Page 86
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
Excessive Power Consumption
15
Page 87
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
Excessive Power Consumption
15
Page 88
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm
Excessive Power Consumption
15
Page 89
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm
Excessive Power Consumption
15
Page 90
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm
Excessive Power Consumption
15
Page 91
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 32 nm
Excessive Power Consumption
15
Page 92
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 32 nm
HKMG
Excessive Power Consumption
15
Page 93
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 28 nm32 nm
HKMG
Excessive Power Consumption
15
Page 94
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 28 nm 14nm32 nm
HKMG
Excessive Power Consumption
15
Page 95
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 28 nm 14nm32 nm
HKMG
Memory Activity
Excessive Power Consumption
15
Page 96
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 28 nm 14nm32 nm
HKMG
Single-core Thermal Design Point
Memory Activity
Excessive Power Consumption
15
Page 97
Pow
er (m
W)
0
500
1000
1500
2000
2500
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
In-Order Out-of-Order Aggressive Out-of-OrderA8 A9 A15 A57
65 nm 45 nm 28 nm 14nm32 nm
HKMG
Single-core Thermal Design Point
Memory Activity
Mobile CPUs designs are beginning to approach a power wall.
Excessive Power Consumption
15
Page 99
Nor
mal
ized
Ener
gy
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
16
Page 100
Nor
mal
ized
Ener
gy
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
16
Page 101
Nor
mal
ized
Ener
gy
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
16
Page 102
Nor
mal
ized
Ener
gy
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
Peak Energy Efficiency
16
Page 103
Nor
mal
ized
Ener
gy
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Year2009 2010 2011 2012 2013 2014 2015
Coremark SPEC Bleh
Peak Energy Efficiency
Mobile CPU performance improvements are in an era of energy efficiency stagnation.
16
Page 104
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
cePower Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Incorporating the End-User
17
Page 105
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Incorporating the End-User
17
Page 106
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Incorporating the End-User
17
Page 107
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Incorporating the End-User
?17
Page 108
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
Incorporating the End-User
17
Page 109
Have mobile CPU advancements improved end-user satisfaction?
18
Page 110
Have mobile CPU advancements improved end-user satisfaction?
1. Is single-core performance necessary?
18
Page 111
Have mobile CPU advancements improved end-user satisfaction?
1. Is single-core performance necessary?2. Is multi-core performance necessary?
18
Page 112
Have mobile CPU advancements improved end-user satisfaction?
1. Is single-core performance necessary?2. Is multi-core performance necessary?3. Does graphics performance matter more
than CPU performance?
18
Page 113
Studying user satisfaction requires users
19
Page 114
Studying user satisfaction requires users
19
LOTS^
Page 115
Leveraging the Crowd to Achieve Scale
20
Page 116
Leveraging the Crowd to Achieve Scale
Over 25,000 participants!
20
Page 117
Survey Design
21
Page 118
Survey Design
21
Page 119
Survey Design
21
Page 120
Survey Design
21
Page 121
Emulating the Mobile CPU Evolution
2009 2010 2011 2012 2013 2014
Year
22
Page 122
Emulating the Mobile CPU Evolution
D
2009 2010 2011 2012 2013 2014
Year
22
Page 123
Emulating the Mobile CPU Evolution
D S
2009 2010 2011 2012 2013 2014
Year
22
Page 124
Emulating the Mobile CPU Evolution
D S N
2009 2010 2011 2012 2013 2014
Year
22
Page 125
Emulating the Mobile CPU Evolution
D S N S3
2009 2010 2011 2012 2013 2014
Year
22
Page 126
Emulating the Mobile CPU Evolution
D S N S3 S4
2009 2010 2011 2012 2013 2014
Year
22
Page 127
Emulating the Mobile CPU Evolution
D S N S3 S4 S5
2009 2010 2011 2012 2013 2014
Year
22
Page 128
Emulating the Mobile CPU Evolution
D S N S3 S4 S5
2009 2010 2011 2012 2013 2014
Year
1
Core
s
22
Page 129
Emulating the Mobile CPU Evolution
D SN
S3 S4 S5
2009 2010 2011 2012 2013 2014
Year
12Co
res
22
Page 130
Emulating the Mobile CPU Evolution
D SN
S3
S4 S5
2009 2010 2011 2012 2013 2014
Year
1234
Core
s
22
Page 131
Emulating the Mobile CPU Evolution
D SN
S3 S4
S5
2009 2010 2011 2012 2013 2014
Year
1234
Core
s
22
Page 132
Emulating the Mobile CPU Evolution
D SN
S3 S4 S5
2009 2010 2011 2012 2013 2014
Year
1234
Core
s
22
Page 133
Emulating the Mobile CPU Evolution
D SN
S3 S4 S5
2009 2010 2011 2012 2013 2014
Year
1234
Core
s
22
Page 134
Emulating the Mobile CPU Evolution
D SN
S3 S4 S5
1234
Core
s
422 729 1036 1497 1958 2457
S5 Clock Frequency
22
Page 135
Emulating the Mobile CPU Evolution
D SN
S3 S4 S5
1234
422 729 1036 1497 1958 2457
S5 Clock Frequency
S5 C
ores
Ena
bled
22
Page 136
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 137
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 138
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 139
Putting the Pieces Together
CPU Frequency
CPU Cores
GPU Frequency
Record User Parametrized Replay Post Survey Crowdsource
23
Page 140
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 141
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 142
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 143
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 144
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 145
Putting the Pieces TogetherRecord User Parametrized Replay Post Survey Crowdsource
23
Page 146
Do we need single-core performance?
24
Page 147
Do we need single-core performance?
24
Page 148
Do we need single-core performance?
24
Page 149
Do we need single-core performance?
24
Page 150
Do we need single-core performance?
24
Page 151
Do we need single-core performance?
24
Page 152
Do we need single-core performance?
24
Page 153
Do we need single-core performance?
24
Page 154
Do we need single-core performance?
24
Page 155
Do we need single-core performance?
24
Page 156
Do we need single-core performance?
24
User satisfaction is latency-critical. Single-core CPU performance enhancements have been crucial to the end-user.
Page 157
Do we need multi-core performance?
25
Page 158
Do we need multi-core performance?
25
Page 159
Do we need multi-core performance?
25
Page 160
Do we need multi-core performance?
25
Page 161
Do we need multi-core performance?
25
Page 162
Do we need multi-core performance?
25
Page 163
Do we need multi-core performance?
25
Page 164
Do we need multi-core performance?
25
Page 165
Do we need multi-core performance?
25
Page 166
Do we need multi-core performance?
25
Page 167
Do we need multi-core performance?
25
Multi-threading is being used for user-critical functionalities. Multiple CPU cores can provide benefit to the end user.
Page 168
Does graphics performance matter more than CPU performance?
26
Page 169
Does graphics performance matter more than CPU performance?
26
Page 170
Does graphics performance matter more than CPU performance?
26
Page 171
Does graphics performance matter more than CPU performance?
26
Page 172
Does graphics performance matter more than CPU performance?
26
Page 173
Does graphics performance matter more than CPU performance?
26
Page 174
Does graphics performance matter more than CPU performance?
26
Page 175
Does graphics performance matter more than CPU performance?
26
Page 176
Does graphics performance matter more than CPU performance?
Even amongst applications that make use of the GPU and other accelerators, end-users are sensitive to CPU performance.
26
Page 177
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
cePower Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 178
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 179
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 180
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 181
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 182
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
Satisfaction Perf
orm
ance
Power Consumption
Pow
er B
udge
ts
BottlenecksFe
atur
es
At the Mercy of Power Constraints
27
Page 183
How has the rest of the mobile device evolved around the CPU?
28
Page 184
Sharing the Power Budget: Device-level
29
Page 185
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Sharing the Power Budget: Device-level
29
Page 186
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Sharing the Power Budget: Device-level
29
Page 187
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Sharing the Power Budget: Device-level
29
Page 188
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Sharing the Power Budget: Device-level
29
Page 189
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Sharing the Power Budget: Device-level
29
Page 190
Pow
er (w
atts
)
0
2
4
6
8
Year
2009 2010 2011 2012 2013 2014 2015
Display RadioCPU: Single-core CPU: Multi-core
Thermal Throttling
Sharing the Power Budget: Device-level
29
Page 191
Mobile SoC: Unsustainable By Design
30
Page 192
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
Mobile SoC: Unsustainable By Design
30
Page 193
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
Mobile SoC: Unsustainable By Design
30
Page 194
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
Mobile SoC: Unsustainable By Design
30
Page 195
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
Mobile SoC: Unsustainable By Design
30
Page 196
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
Mobile SoC: Unsustainable By Design
30
Page 197
Repo
rted
TDP
(wat
ts)
0
3
6
9
12
Phone
Galaxy S5 Galaxy S6
CPU: A15 CPU: A7 GPU Other
3.5 W TDP Budget
Mobile SoC: Unsustainable By Design
30
Page 198
HardwareSoftware Mobile DeviceProcessorApplicationsEnd-Users
SatisfactionPe
rfor
man
cePower Consumption
Pow
er B
udge
ts
CapabilitiesFe
atur
es
Tying It All Together
31
Page 199
A Call to Action
Mobile Device
Processor
Applications
End-Users
32
Page 200
A Call to ActionUse metrics that incorporate end-user
Mobile Device
Processor
Applications
End-Users
32
Page 201
A Call to ActionUse metrics that incorporate end-user
Mobile Device
Processor
Applications
End-Users
Identify user-critical application segments
32
Page 202
A Call to ActionUse metrics that incorporate end-user
Understand application characteristics
Mobile Device
Processor
Applications
End-Users
Identify user-critical application segments
32
Page 203
A Call to ActionUse metrics that incorporate end-user
Understand application characteristics
Deviate from desktop scaling and embrace the era specialization
Mobile Device
Processor
Applications
End-Users
Identify user-critical application segments
32
Page 204
A Call to ActionUse metrics that incorporate end-user
Understand application characteristics
Deviate from desktop scaling and embrace the era specialization
Consider thermal and energy constraints at the mobile-device level
Mobile Device
Processor
Applications
End-Users
Identify user-critical application segments
32
Page 206
8
6
4
2
0
Cor
e C
ount
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Year
213
29
25
21
Cac
he S
ize
(KB)
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Year
L1: Both L2: Mobile L2: Desktop L3: Desktop
5
4
3
2
1
0Clo
ck F
requ
ency
(GH
z)
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Year
Desktop-like CPU Scaling
Clock Scaling Resource Scaling Core Scaling
34
Page 207
Other Applications
Page 208
Application Selection Criteria
Page 209
http://www.anandtech.com/show/9780/taking-notes-with-ipad-pro/2
Apple SoCs
Page 210
http://www.anandtech.com/show/9686/the-apple-iphone-6s-and-iphone-6s-plus-review/4
Apple CPUs
Page 211
http://techreport.com/r.x/2014_8_11_Nvidia_claims_Haswellclass_performance_for_Denver_CPU_core/denver-block.jpg
Page 212
Crowdsourcing Considerations
Page 213
Crowdsourcing Considerations
Page 214
Crowdsourcing Considerations
Task Design
Page 215
Crowdsourcing Considerations
Task Design▹ Open-ended question
Page 216
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers
Page 217
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive
Page 218
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive▹ $0.10 / task
Page 219
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive▹ $0.10 / task
▹ $8.00 / hour wage
Page 220
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive▹ $0.10 / task
▹ $8.00 / hour wageData Integrity
Page 221
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive▹ $0.10 / task
▹ $8.00 / hour wageData Integrity▹ Scale of trials (> 50 trials / configuration)
Page 222
Crowdsourcing Considerations
Task Design▹ Open-ended question
▹ Well-defined answers Worker Recruitment and Incentive▹ $0.10 / task
▹ $8.00 / hour wageData Integrity▹ Scale of trials (> 50 trials / configuration)
▹ Validation keyword prevents scripters
Page 224
Other CPUs and Benchmarks: Perf
11
9
7
5
3
1
Nor
mal
ized
Spe
edup
D S N S3 S4 S5 S6Smartphone Model
SPEC Coremark Sunspider Geekbench Stream
Stock IP Custom IP
Page 225
Other CPUs and Benchmarks: Power
2.5
2.0
1.5
1.0
0.5
0.0
Dyn
amic
Pow
er (W
)
D S N S3 S4 S5 S6Smartphone Model
Page 226
Other CPUs and Benchmarks: Energy
1.0
0.8
0.6
0.4
0.2
0.0
Nor
mal
ized
Ene
rgy
D S N S3 S4 S5 S6Smartphone Model