compute power with energy-...
TRANSCRIPT
![Page 1: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/1.jpg)
1
Compute Power with Energy-Efficiency
Partnerships, Standards and the ARM GPU Perspective
Jem DaviesARM Fellow, VP of Technology,Media Processing Division, ARM
![Page 2: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/2.jpg)
2
What does ARM do? ARM® was founded in 1990 Initially designed CPUs Over twenty years, have covered CPUs, GPUs, DSPs,
audio processors, video processors, tools, fabric IP and physical IP
ARM’s partners make SoCs containing many heterogeneous compute engines
Heritage of low-power Not just mobile, but across multiple segments > 6 Bn ARM-based chips in 2010
![Page 3: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/3.jpg)
3
ARM also does GraphicsGrowing the Media
Processor Licensing BaseGrowing Shipments in Mobileand Non-Mobile Applications
46 licenses for graphics and video 7 licenses added in Q1 2011
More Mali™ based chips shipping into mobile and consumer electronics devices
Samsung announced that Mali-based Exynos delivered 5 times more graphics performance than their previous design
STMicroelectronics announced 10 major STB design wins for Mali-based STi7108
46
Licenses Partners RoyaltyPayers
40
6
Pre-2007, 248
14
2007
2008
2009
2010 11
Q1 2011 7
![Page 4: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/4.jpg)
4
History - the Passage of Time What lessons from history can we learn?
![Page 5: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/5.jpg)
5
Microprocessor Forum 1992
![Page 6: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/6.jpg)
6
Count the Architectures (11)
ARM
MIPS29K
PA
NVAX
N32x1688xxx
Alpha
x86
i960
SPARC
x86x86 x86
![Page 7: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/7.jpg)
7
The Survivors
ARM
x86
![Page 8: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/8.jpg)
8
What Were the Lessons? Good architectures will succeed
Power did matter It has come to matter more
Too much variety can be a bad thing For developers
The Ecosystem Really matters Developers need to be able to support a range of platforms
![Page 9: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/9.jpg)
9
The Eras of Computing
1960 1970 1980 1990 2000 2010 2020
Uni
ts
1M
10MMainframe
Mini
1st Era
100M
1 BillionPC
DesktopInternet
2nd Era
100 BillionThe Internet of Things
10 BillionMobile Internet
![Page 10: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/10.jpg)
10
The Eras of Computing
1960 1970 1980 1990 2000 2010 2020
Uni
ts
1M
10MMainframe
Mini
1st Era
100M
1 BillionPC
DesktopInternet
2nd Era
100 BillionThe Internet of Things
10 BillionMobile Internet
![Page 11: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/11.jpg)
11
Industry Changes in Requirements
Evolution of the industry-driving metric
![Page 12: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/12.jpg)
12
Industry Changes in Requirements
Evolution of the industry-driving metricFunctionality
Up to 1980sSupercomputers &
mainframes
![Page 13: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/13.jpg)
13
Industry Changes in Requirements
Evolution of the industry-driving metricFunctionality
Up to 1980sSupercomputers &
mainframes
Functionality$
1990sThe personal
computer
![Page 14: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/14.jpg)
14
Industry Changes in Requirements
Evolution of the industry-driving metricFunctionality
Up to 1980sSupercomputers &
mainframes
Functionality$
1990sThe personal
computer
FunctionalityPower × $
2000sNotebooks
![Page 15: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/15.jpg)
15
Industry Changes in Requirements
Evolution of the industry-driving metricFunctionality
Up to 1980sSupercomputers &
mainframes
Functionality$
1990sThe personal
computer
FunctionalityPower × $
2000sNotebooks
FunctionalityEnergy × $
2010sMobiles &mobility
![Page 16: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/16.jpg)
16
ARM’s Strategic Direction
FunctionalityEnergy×$
![Page 17: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/17.jpg)
17
ARM’s Strategic Direction
The bad news: this is a veryhard metric to optimize for
The good news: if you crack it, you “own”the simpler metrics as well…
FunctionalityEnergy×$
![Page 18: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/18.jpg)
18
Energy-efficiency Has Always Mattered Segment differentiation matters less than you think Mobile worries about warm hands/ears And “You can never be too thin”
Sensors want 10 years from a button cell battery Everybody hates fans Desktops struggle to: Dissipate more than ~150W out of a single chip package Supply more than ~300W from a PSU onto a PCI card
Servers struggle: To get more than ~10kW into a rack To get more than ~500kW of heat out of a shipping container Spending more on power and cooling than on hardware To buy more power from the grid
![Page 19: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/19.jpg)
19
What’s Coming?
Why should developers care?
![Page 20: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/20.jpg)
20
Moore’s Law Is Not Dead Some version of Moore’s
law will continue to be true for this decade
But it is getting less and less relevant
In the past, the fabs and Moore’s law gave us PPA (Power, Performance and Area) improvements for free But not any more…
![Page 21: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/21.jpg)
21
Expectations of The Future
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
28nm (2010)
20nm (2012)
16nm (2014)
14nm (2016)
10nm (2018)
7nm (2020)
Freq
uenc
y (G
Hz)
Expectation: 20% frequency uplift per node
![Page 22: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/22.jpg)
22
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
28nm (2010)
20nm (2012)
16nm (2014)
14nm (2016)
10nm (2018)
7nm (2020)
Freq
uenc
y (G
Hz)
The Bitter Reality?
Expectation: 20% frequency uplift per node
Current best prediction of core-type performance
Note:Normalized to fixed
Leakage / um
![Page 23: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/23.jpg)
23
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
28nm (2010)
20nm (2012)
16nm (2014)
14nm (2016)
10nm (2018)
7nm (2020)
Freq
uenc
y (G
Hz)
The Bitter Reality?
Expectation: 20% frequency uplift per node
Current best prediction of core-type performance
Note:Normalized to fixed
Leakage / um
![Page 24: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/24.jpg)
24
Silicon Generations: Shrink and Add
32nm45nm
Expectation: shrink and addnew functionality in about same area
![Page 25: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/25.jpg)
25
The Creation of Dark Silicon
Year
Node
Area-1
Peak freq
Power
45nm
2008
1
1
1
22nm
2014
4
1.6
1
11nm
2020
16
2.4
0.6
Exploitable Si(in 45nm power budget)
25%
(4 x 1)-1 = 25%
10%
(16 x 0.6)-1 = 10%
Source: ITRS 2008
![Page 26: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/26.jpg)
26
The Creation of Dark Silicon
Year
Node
Area-1
Peak freq
Power
45nm
2008
1
1
1
22nm
2014
4
1.6
1
11nm
2020
16
2.4
0.6
Exploitable Si(in 45nm power budget)
25%
(4 x 1)-1 = 25%
10%
(16 x 0.6)-1 = 10%
Source: ITRS 2008
Source: ITRS 2008Lack of power scaling severely limits
system options!
![Page 27: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/27.jpg)
27
So What Can We Do? We can have more transistors
We just can’t power them all at the same time
We need to use these extra transistors in new ways Multicores Many-cores Domain-specific processors
It all points to heterogeneous processing And aggressive power management
Computing to be done in the most efficient place
![Page 28: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/28.jpg)
28
Graphics
![Page 29: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/29.jpg)
29
The Perfect Domain-specific Use-case? GPUs have traditionally been the poster-boys for domain-specific computing Everyone accepts graphics on CPU makes little sense From performance perspective and from energy-efficiency
3D computer graphics used to be very fixed-function Transform, lighting, texturing, rendering...Hardware followed those fixed functions
Modern APIs provide more flexible, programmable models GPUs now used for more general-purpose computing So where does this all end up?
![Page 30: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/30.jpg)
30
Domain-Specific vs. General Purpose Do GPUs merge with CPUs? Graphics remains one of the very few problem spaces that have very high levels of
parallelism: Do <foo> on every vertex in the frame (10s - 100s of kvertices/frame) Do <bar> on every pixel in the frame (millions of pixels/frame)
Data-level parallelism through thread-level parallelism
Throughput-oriented architectures (and implementations) will continue to have great advantages: Performance Energy
CPUs remain “easier” to program
![Page 31: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/31.jpg)
31
Innovative ‘Midgard’ GPU architecture Tri-pipe - for performance and flexibility
Scalable and high-end performance Superb graphics quality and performance Up to 2 Gpix/s
Powerful general-purpose computing Up to 68 GFLOPS
Multicore scalability up to 4 cores
Resource efficiency Frugal use of memory and bus-bandwidth Dynamic power management
Broad API and OS support OpenCL™ Full Profile/Renderscript OpenGL® ES and Open VG™ Microsoft DirectX™ to v11
Common driver for all Midgard GPUs
Mali-T604 for Visual Computing
![Page 32: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/32.jpg)
32
It’s All About The System
CoreLink™ CCI-400 Cache Coherent Interconnect128 bit @ 0.5 Eagle frequency
Quad Cortex-
A15
CoherentI/O
device
128b
Mali-T604Graphics
asyncbridge
asyncbridge
128b 128b
MMU-400 MMU-400
128b 128b
Huxley DMCDMC-400
ACE
ACE ACE-Lite + DVM
ACE-LiteACE-LiteACE-Lite
ACE-Lite
NIC-400
Other Slaves
Other Slaves
128b
NIC-400
LCDVideo
Quad Cortex-
A15
128b
ACE
ACE
AXI4
AXI4
Configurable: AXI4/AXI3/AHB/APB
Configurable: AXI4/AXI3/AHB
DDR3/LPDDR2
WideIO DRAM
ACE-LiteACE-Lite
PHYPHY
GIC-400
ACE-Lite + DVM ACE-Lite + DVM
128b
MMU-400
![Page 33: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/33.jpg)
33
GPU Computing Observations Beware of vested interests Companies that designs CPUs, GPUs, fabric can afford a more balanced view
Raw GFLOPS numbers tell you (nearly) nothing Most GPUs will only run well (efficiently, fast) with very high numbers of
simultaneous threads (high parallelism) Most GPUs are designed for high throughput Most GPUs will run badly with few threads or requirements for low latency Remember the system (memory, energy...) Understand performance vs. bandwidth vs. latency
There is still a shortage of highly-parallelisable algorithms Remember Amdahl’s law
![Page 34: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/34.jpg)
34
Amdahl’s Law is Alive and Well100% parallel
95% parallel
90% parallel
75% parallel50% parallel
0
4
8
12
16
20
24
28
32
0 4 8 12 16 20 24 28 32
Max
imum
spe
edup
Number of cores
Speedup on parallel processors is limited by thesequential portion of the program
Sequential portion need not be largeto constrain speedup significantly !
![Page 35: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/35.jpg)
35
Standards
![Page 36: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/36.jpg)
36
Complexity Out-of-order superscalar CPUs introduces some complexity Multicore CPUs/GPUs – more complexity Memory consistency model Threading models
Heterogeneous computing - more complexity Some GPU compute engines are: Complex Badly described Difficult to reason about
Most graphics developers want to create visually stunning content, not argue about the finer points of computer science
![Page 37: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/37.jpg)
37
Complexity, Abstraction, Standards If we do not abstract away from this complexity
and present a simpler world to the developer… Either they won’t use these new systems Or, they’ll use it and get it wrong
Either way, it will be our fault,and they will hate us for it
If we all provide different abstractions… They will hate us for it
We need standard(s) And a very small number of them
![Page 38: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/38.jpg)
38
Standards
And if you don’t like any of the current standards, if you wait for a while, there will be new ones along…
"The great thing about standards is that you have SO many to choose from."
- Andrew Tannenbaum
![Page 39: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/39.jpg)
39
Before Creating New Standards… Check that you are solving the right problem The problem changes Check that there isn’t a “Good enough” solution already
There are many solutions to the same problem Not all have the same costs Some costs are well hidden Some costs are not borne by “you” Some costs only become clear over time
Remember the ecosystem
![Page 40: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/40.jpg)
40
Developers and Developer Communities ARM and AMD know a bit about developer communities
Content remains king!
Development is really expensive and very hard
Re-use of content is an economic imperative For developers…
But it’s not enough that developers don’t hate us! They have to make money as well
![Page 41: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/41.jpg)
41
Business Model Benefits Business models like ARM’s means everybody makes money Which encourages that developer community What goes around,
comes around
![Page 42: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/42.jpg)
42
Different-sized Pies
![Page 43: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/43.jpg)
43
Remember…
Being Right is Easy
Making Money is the Tricky Bit
![Page 44: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/44.jpg)
44
Final Thoughts To provide increased performance we must be clever Not just rely on Moore’s Law
GPU computing is here already Heterogeneous computing is next Computing in the most energy-efficient way is the real problem Solve that, and everything else will be easy
When we introduce new complex stuff… It must make/save money It must be easily (and widely) usable Or it won’t get used!
And then we wouldn’t make any money
![Page 45: Compute Power with Energy- Efficiencydeveloper.amd.com/wordpress/media/2013/06/Compute_Power_with... · Compute Power with Energy-Efficiency ... 3D computer graphics used to be very](https://reader033.vdocument.in/reader033/viewer/2022042801/5aa7f39a7f8b9a6d5a8d06e2/html5/thumbnails/45.jpg)
45
Thank you
Questions?