2. Objective of this session
Will help you develop & architect
new breakthrough ideas
There is a transition in the hardware
industry
Look this is an opportunity for you
This session will help you discover &
seize those opportunities
3. The age of Multicore,
what does it mean for you?
4. Inevitable future
“Reduced cost is one of the
big attractions of integrated
electronics, and the cost 1.72 Billion
Transistors advantage continues to increase
Per Die as the technology evolves toward Transistors
the production of larger and larger on a chip
1010 circuit functions on a single
Today
1965 Data (Moore) semiconductor substrate.”
Itanium™ 2
Microprocessor Gordon Moore, 1965 Processor Family
Itanium™ 2 Processor
Itanium™ Processor
Pentium® 4 Processor
Pentium® III Processor
Pentium® II Processor
Pentium® Processor
486™ Processor
105 386™ Processor
80286
8086
8080
8008
4004
More transistor budget, transistors per $
(Moore's law)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
10. By 2010, all PCs shipping will be multi-core
450
400
350
300
250
200 By 2010, all PCs shipping will be multi-core
150
100
50
0
2007 2008 2009 2010 2011 2012
Single Core Dual Core Triple Core Quad Core
Hex Core Octal Core 12 Core 16 Core
11. Directions: Shift to multi-core &
many-core Multi core; Many core
Quad-Core
Dual-Core
Single-Core 2006/07
2006
Previous
Today, Your Home PC can run 16 threads,
all simultaneously !!!
12. Multi & Many, Heterogeneous
12 Cores 24 Cores 144 Cores
Cache
C1 C2 Small
Core
Cache Large Core
Small
C3 C4 Core
13. The age of Multicore,
what does it mean for you?
14. Looking for the opportunity to „wow‟ your
users?
Performance Through
Multi-Core
Your apps
Performance
can do
things that
wasn’t
possible
before!
Performance Through
frequency
2006
- +
Your opportunity to add surprising
capabilities in your apps
16. end-user value
human values happiness, togetherness, spirituality, …
human needs
& motivations
user experiences
tasks
solutions
systems
platforms
subsystems
ingredients
16
17. end-user value
human values happiness, togetherness, spirituality, …
human needs
& motivations
user experiences
what we make
possible
tasks
solutions
systems what we collectively
platforms
make (ecosystem)
subsystems
what Intel makes
ingredients
17
18. What will be characteristics of a future
multicore app?
Entertainment “RMS” Applications
Recognition
TIPS Learning &
Mining
IPS = Instruction per second
Travel Synthesis
Performance
RMS Personal Media
Creation and
GIPS
Management
3D &
Video
Tera-scale
MIPS Mult-
Media Multi-core
Text
KIPS
Single Core
Health
Kilobytes Megabytes Gigabytes Terabytes
Dataset Size
19. Opportunities for the Future
Immersive
Improved Experience
Productivity
Breakthrough
Innovation
20.
21.
22.
23.
24.
25. Recognition Mining Synthesis
What is a tumor? Is there a tumor here? What if the tumor progresses?
It is all about dealing efficiently with complex multimodal datasets
26. Emerging “Killer Apps” (R)
“What is it?”
Recognition Modeling and identifying
using multi-modal data
Speech recognition
combining
speech analysis
and lip reading
Source: Intel
Nefian, et. al, “Dynamic Bayesian networks for audio-visual speech
recognition,” Journal of Applied Signal Processing, 2002
27. Emerging “Killer Apps” (M)
“Where is it?”
Mining Search for a similar
instance
Source: Intel
28. Emerging “Killer Apps” (S)
“What if?”
Synthesis Creating new model
instances
Source: Intel Source: InTrace Source: Stanford
29. iRMS Loop Illustration
Analytically Correct, Muscle- Physics-Based Deformable Tissue
Video Input Feature Tracking
Activated Human Head Model (Finite Element Method)
Facial Muscle Activations:
Compact motion
representation,
well suited for modeling
and synthesis
User Interaction:
Modified Muscle
Activations
User Interaction:
Video Output Modified Physical
Model
Source: E. Sifakis, I. Neverov and R. Fedkiw, “Automatic Determination of Facial Muscle Activations from
Sparse Motion Capture Marker Data”, ACM SIGGRAPH, 2005 (to appear)
30. Educational Simulation…Synthesis
“Literacy in the computer age means being able to make dynamic models of the ideas
that you're thinking about and trying to explain to others and arguing about. In other
words, simulations. In the future, children will grow up learning how to make ideas
that actually function on computers and use them as part of their discourse and
thinking processes.” – Alan Kay of Viewpoints Research Institute
Visualize and simulate complex natural, physical systems Playing and learning
- NASA Goddard Scientific Visualization Studio - LEGO Mindstorms / LEGO Digital Designer
Usage Trends Key Technologies
• Graphical “literacy” is part of curriculum • Advanced Real-Time Graphics
• Computer Vision
• Visual simulations with “what if”
analyses for finances, home
• Machine Learning
renovations, storms • Physics
• RMS, RTA
34. Necessity for Terascale performing apps
1 TFlops
800 GFlops
600 GFlops
400 GFlops Driving Factors:
•Doubling number of slices (128,
200 GFlops 256, 512, 1024, 2048 over an
approximate 10 year period )
100 GFlops
•Rotational speed
(25% reduction in time per image)
10 GFlops
Image Reconstruction •Number of detector channels
1 GFlops •Number of views per “rotation”
2004 2006 2008 2010 2012
Medical Imaging: Needs 1TFlops by 2010
Source: Intel Digital Health Group
35.
36. The age of Multicore,
what does it mean for you?
37. The secret
to leverage opportunities
Parallelism
Parallelism
Parallelism
Highway animation
38. Simplifying Parallelism
Design Code Optimize Validat
e
Across multiple Actionable Correctness
Applications for programming performance
Parallelism models with data guidance
and task
parallelism
39. Parallel Developer Tools
Defining the developer experience for constructing parallel
applications
• Design and modeling tools to enable developers
Design to start with zero parallelism debt
• Debug across multiple programming models, with
Debug data and task-focused visualizations
• Actionable performance guidance for
Optimize understanding and optimizing parallel
applications
• Tools for developers and testers to validate
Validate correctness and cope with inherent non-
deterministic execution
Integrate/Tool/Encapsulate/Raise
40. Parallelism vs. Concurrent
Concurrent processing: Parallel processing:
What: independent requests decompose one task to
(most server applications) enable concurrent execution
How:
“Arbitrate “ownership” of the “Start concurrent searches
Example: nodes” …”
Simulating isolation of
Implemen threads
Scheduling tasks
t:
Multi-threading, Asynchronous, …
47. Summary
Discover possibilities with multi core
–Innovate software capabilities by
leveraging multi core compute power
– (Examples: Google Desktop search, .NET WPF)
“Parallel thinking” software
–Now is the time to start your transition to parallel
computing.
–If you aren’t parallel, you can’t fully utilize multi-
core processors.
48. Conclusions
Create opportunities with the multicore
era:
–create new trends & experiences
– E.g. Research s/w for Visual/3D web
Teach multi-core:
–develop your intuitions for parallelism
www.intel.com/software
Notas do Editor
With TPL’s structured parallelism, refactoring the code is easy. If you squint, you almost recognize a “parallelfor” statement. C# 3.0 (and various other languages) is expressive enough to make such code concise to write using lambda expressions. Note that size is the exclusive upper-bound for Parallel.For, and “i” is the lambda parameter fed in to the body of the loop.