This document contains legal disclaimers for any presentations made by Intel. It states that any forward-looking statements are subject to risks and uncertainties, and actual results may differ from expectations. It also notes where to find reconciliations for any non-GAAP financial measures used and disclaims any warranties or intellectual property claims. Finally, it says performance tests were conducted on specific systems and components, and results may vary depending on hardware and software configuration.
1 of 31
Download to read offline
More Related Content
Bending The Curve
1. Legal Information
Todays presentations contain forward-looking statements. All statements made that are not historical facts are subject to a number of
risks and uncertainties, and actual results may differ materially. Please refer to our most recent Earnings Release and our most recent
Form 10-Q or 10-K filing for more information on the risk factors that could cause actual results to differ.
If we use any non-GAAP financial measures during the presentations, you will find on our website, intc.com, the required reconciliation
to the most directly comparable GAAP financial measure.
INFORMATION IN THIS DOCUMENT IS PROVIDED AS IS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY
INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS
ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS
FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY
RIGHT.
Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate
performance of Intel products as measured by those tests. Any difference in system hardware or software design or configuration may
affect actual performance. Buyers should consult other sources of information to evaluate the performance of systems or components
they are considering purchasing. For more information on performance tests and on the performance of Intel products, reference
www.intel.com/software/products.
Software and workloads used in performance tests may have been optimized for performance only on Intel
microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components,
software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other
information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that
product when combined with other products.
2. Bending The Curve
Daryan Dehghanpisheh
May 24, 2012
Source:
All trademarks and brands are registered property of their respective companies.
3. A Shameless Plug
Join in the discussion!
All trademarks and brands are registered property of their respective companies. 3
4. The So What?
We have Knowns, unknowns, and known unknowns
Radical performance is here. And, affordable for most.
Dealing with data requires a whole new strategy.
The Cost Curves of the business are bending.
All told: A new breed of competitor lurks.
4
6. What I hear about, all the time.
Source: 6
1. Google Images search, public domain
All trademarks and brands are registered property of their respective companies.
7. These Are Not Black Swans
We Knew the Storm Was Coming
Source: 7
1. Google Images search, public domain.
All trademarks and brands are registered property of their respective companies.
8. And Yet, There Was Still Carnage
Halted Due to Technical Issues
Source: 8
1. Google Finance, chart.
All trademarks and brands are registered property of their respective companies.
10. Cause and Effect?
Performance Unleashed
Cost Per:
Year GFLOP GB Mic
2007 $2.48 $48.00 $20,000,000.00
2012 $0.48 $0.92 $150,000.00
Decrease >5X >50X >130X
Massive Capabilities. Small Investments.
Source: 10
1.The Aggregate, 2003-08-22, retrieved 2012-02-09
2. Adam Stevenson, Yann Le Du, and Mariem El Afrit. "^ Adam Stevenson, Yann Le Du, and Mariem El Afrit. "High-performance computing on gamer PCs." Ars Technica. March 31, 2011.." Ars Technica. March 31, 2011.
11. Optimization Is Relative
Latency Registers
Arb GETCO (1 Cycle)
Flow based Caches
execution
arbitrage Breakwater (2-20 Cycles)
Micro-market Memory
linear models BATS (100 Cycles)
Remote Memory
Machine Learning Models Automat (10,000 Cycles)
Stochastic Models of order book Flash Drives
dynamics
Citadel
(100,000 Cycles)
Spinning Disks
Statistical Arbitrage
Rentec (10,000,000 Cycles)
It is a blend of art, science...and strategy.
13. Affordable Low(er) Latency
Latency vs Message Rate (Mpps)
15
Latency (袖sec)
10
5
0
1.0 1.5 2.0 2.5 3.0 3.5 4.0
Westmere 3.5 6.7 6.8 13.2
SandyBridge 2.5 2.6 2.9 3.2 3.8 4.7 7.7
Moores Law Brings Lower Latency to the Masses More Easily
Source:
1. Solarflare: http://www.solarflare.com
All trademarks and brands are registered property of their respective companies.
14. LATENCY LOWER IS BETTER
The Latency Spectrum
Parallelism; # of Threads
15. LATENCY LOWER IS BETTER
The Latency Spectrum
PARRALELISM; # of THRAEDS
Apps that are truly single threaded and acutely latency
sensitive represent an extremely small footprint of code.
16. Our Collective Challenge
In 2012 all PCs & Servers shipping are Multi-Core, Multi-Threaded
Source: PC Semiconductor Market Briefing: Re-Architecting the PC and
Units = M the Migration of Value, June 2008
450
400
350
300
250
200
150
PERFORMANCE
Multi-core Needs 100
Parallel Applications 50
0
GHz Era Multi-core Era 2007 2008 2009 2010 2011 2012
TIME Single Core Dual Core Triple Core Quad Core
Hex Core Octal Core 12 Core 16 Core
const str newNormal = parallel;
16
17. Setting The Pace: Intel x86 ISA
Next:
Leapfrog performance with wide Future Extensions
Performance / core
Hardware FMA
vectorization, ISA extensions.
Memory Latency/BW
Many Other Features
Traditionally:
Ivy Bridge
Improved performance with upcoming
Half float support
Intel速 micro architectures Random Number
Generator
Intel Xeon E5 Series
Intel Xeon 56xx Series Intel速 AVX
Intel速 SSE4
Up to 2X FLOP Output
Intel AES-NI^-Cryptographic
Acceleration 3-Operand instructions
Memory latency, BW
Fast Unaligned support
Core
Intel Architecture Goes Well Beyond Speeds and Feeds
^AES-NI - Advanced Encryption Standard New Instruction
AVX Advanced Vector Extensions
18. Intel Xeon E5 - AVX Instructions
Newest in a long line of processor
instruction innovations
Increases floating
point operations per clock up to 2X1
for technical, financial, scientific &
content creation applications
Performance to
solve your most complex
problems
AVX Can Provide >1.8X Improvement to SSE Code
Source: 18
1. Intel Corporation: http://www.intel.com
All trademarks and brands are registered property of their respective companies.
19. Intel Fused Multiply Add (FMA)
Computes (ab)c with no intermediate rounding.
Can speed up many common floating-point computations
Get twice as many arithmetic ops per instruction
Circuitry does not need to normalize intermediate result.
Accurate computation of small residuals
Varieties supported
types: float, double
widths: scalar, 128-bit vector, 256-bit vector
negations: ab+c, (ab)+c, abc, (ab)c,
And vector versions with alternating signs:
akbkck on even indices akbk+ck on even indices
akbk+ck on odd indices akbkck on odd indices
19
21. A Cambrian Explosion of Data
230M
50M Tweets sent
250M
100M Photos uploaded to Facebook
2B
7B Videos watched
New Types of Data. New Types of Trading Opportunities.
Source: 21
1. Public sources, Company Metrics from public filings: http://finance.google.com/(symbol)
All trademarks and brands are registered property of their respective companies.
22. Creating New Data Challenges
Intelligent Manageable Personalized
New Technologies and Methods are Required
22
26. Alpha From Tweets
Unstructured Graph Data
>2X Better Returns than Major Indexes During Test Period
Source: 26
Correlating Financial Time Series with Micro Blogging Activity, Eduardo J. Ruiz, Vagelis Hristidis, Carlos Castillo, Aristedes Gionis, Alejandor Jaimes, February 8 -2012, Seattle Washingon, US Copyright 2012
27. Analysis (And Execution) In the Cloud
8 People, 4 Trades
>$X00M Gain All from Cloud
Cloud Architectures
Much Higher Alpha. Much Lower Costs.
Source: 27
All trademarks and brands are registered property of their respective companies.
28. Fully Optimizing For Success
Performance Increase
Optimization Choices
Much Higher Alpha. Much Lower Costs.
Source: 28
All trademarks and brands are registered property of their respective companies.
29. How Long Until Youre Optimized?
Days? Months? Years?
New Hardware, New Tools
New Bios New OS / Java, Optimized
Cores, Sockets New Network 1588., thread
Settings Compilers Application
Instructions checker
If time is money, how much of both are you losing?
Start today. Start with Intel.
29
30. One Last Plug
Join in the discussion!
Source: 30
All trademarks and brands are registered property of their respective companies.
31. Thank you.
Source: 31
All trademarks and brands are registered property of their respective companies.