space: the final frontier 1966-1969 the future of computer architecture babylonian dynasty 1894-1595...

24
Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research Sept. 2014 In Honor of Yale Patt’s Dodranscentennial LXXV The Cell Phone !!! Thumb’s Up! 3600 Years

Upload: shamar-wollam

Post on 14-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Space: The Final Frontier1966-1969

The Future ofComputer

Architecture

Babylonian Dynasty1894-1595 BC

Phil EmmaChief Scientist

IBM TJ Watson ResearchSept. 2014

In Honor of Yale Patt’sDodranscentennial

LXXV

TheCell

Phone!!!

Thumb’s Up!

3600 Years

Page 2: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Why did God give man opposable thumbs?

What can we do because we have thumbs?Give Up?

Hitch-hiking?

Live long, and prosper!(Vulcans)

Playing the guitar?

Riding the subway?

Page 3: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Thumbs are for sports… …to hone our hunting skills.

Page 4: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

So what are thumbs for?

But we don’t USE our thumbs in most sports, and we generally don’t use them for hunting!

Babylonian Dynasty1894-1595 BC

Space: The Final Frontier1966-1969

Groovy !Excellent !

BillBill TedTed KirkKirk

Page 5: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

For Texting !!!

Page 6: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

1) We are too pessimistic in forecasting when many technologies will happen.

2) And we are almost always wrong about how, exactly, they will manifest.

Using your thumbs to Text on an iPhone?

OMG LOL !!!

Kirk: Circa 1970

Kirk: Circa 2010

Trends

More on thumbs…

Groovy !Talking into a

hand-held WIRELESSTelephone?

Jeepers!

DILLIGAF?

40 YearsLater…

Page 7: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

0 1 2 “Many”Thumbs were first used for counting. That’s why only people and monkeys are known for using Binary.

Two

Aborigines felt no need to count, and while they all had words for “one” and “two,” only some made it to “three.” The Walpiri, for example, only has words for “one,” “two,” and “many.”

The Piraha of the Amazon have also been cited as a using a “one-two-many” system of counting.

This is why our first few generations of computers have used the Binary number system.

The Binary Number System & Counting

Page 8: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

BillBill TedTedAs we changed to Sexagesimal (Base 60 – As we changed to Sexagesimal (Base 60 – used by the Sumerians in the 3used by the Sumerians in the 3rdrd Millenium BC, Millenium BC, and then the Babylonians), Two Thumbs UP, and then the Babylonians), Two Thumbs UP, in addition to “Many,” was used as a sign of in addition to “Many,” was used as a sign of ultimate approval, or of “Excellence;” also it’s ultimate approval, or of “Excellence;” also it’s meaning in Decimal (Base 10) today.meaning in Decimal (Base 10) today.

Babylonian Dynasty1894-1595 BC

Steve-OSteve-O

Excellent !

Page 9: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

I am thinking about something much more important than bombs.I am thinking about computers. - John von Neumann, 1946

Automatic Computing Engine - Turing

DifferenceEngine

- Babbage

ENIAC – Eckert & Mauchley IAS Machine – von Neumann

Page 10: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Philosophical Breakthroughs

• Babbage – Automatic calculation by machine. Manipulating numbers can be done mechanically – without “thought.”

• Turing – Theory of computation.• Eckert & Mauchley – Doing many fast

calculations on a very large scale.• von Neumann – A stored-program

computer. The program can modify itself while running: what the program does depends on the data.

Page 11: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

How Many View Moore’s Law

Time

log (whatever)

The Good-O

ld D

ays

Now

The World Ends

Page 12: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Moore’s Law – My Interpretation

Time

log (whatever)

CMOS Roadmap:No real architectural innovation required.Keep scaling the Technology!

Now

The Birth of Architectural Innovation !!!

HAPPYBIRTHDAY

YALE

Page 13: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Where we’ve been since von Neumann (1946)

ProgrammingBinaryAssemblyHLLProc.-OrientedObject-Oriented

ArchitectureScalarCacheSuperscalarBranch Pred.Multithreading

ApplicationsCalculationsLinear SystemsText ProcessingOSDatabaseTransactions

ArchitectureRegistersStacksVirtual MachinesVLIW

What dimensions have failed to evolve?

Page 14: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

MetaCode

Macro A

Macro A

Macro A

Macro A

Macro B

Macro B

Macro C

Macro C

DynamicTranslation

AssemblyCode

A, B, and C are NOT visible to the program (to the ISA). They can be defined as best fits any specific implementation.

Legacy code – bound to the ISA.

DynamicTranslationto a Meta-

Architecture

PC

MetaCache

IP

Page 15: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

BA TA BA TA BA TA

BA TA BA TA BA TA

TakenBranch

P31

To P32 & P33

ToIFetch

Patt Predictor Sub-engine

IP

Page 16: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

P31

P32

P33

P32 & P33Inputs

P31 & P33Inputs

P31 & P32Inputs

3 I-FetchPredictions

FA1

FA3

FA2

Target Address(following misprediction)

Patt Predictor

Patt Sub-engine

Patt Sub-engine

Patt Sub-engineIP

Page 17: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Three Big Ones• Environmental

– What’s the temperature? (Regulate)– How much energy & power? (Regulate)– What’s running with me? (Optimizing)– Cost of operation? (Dynamic Adjusting)

• Moving Beyond Scalar Operands– Vectors, Matricies– Records, Linked Lists– Objects

• Abstract Computing, Acceleration– Direct manipulation of surfaces &

objects– Sensory operation

• Hardness & shape• Odor & Taste

– Estimation & approximate computing– The Brain???

Duh!

Page 18: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

0…………………………………………………………………………………..n

The NEW Definition of the von Neumann Bottleneck !

To the program, storage looks like this. Why? Because different levels of storage know nothing about the physical layouts of other levels:

L1 L216 BytesRead Req.

8 Bytes½ Speed

PROC

L3

L4 L4 L4 L4

DDR Memory Cards

L3

Memory

A physical 3D system like this

would simplify the OS dramatically.

Why does the OS juggle 100s of jobs today?

Etc.

Page 19: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Matrix Multiplication in 3D

X =A B CFor Row = 1 to n doBegin

For Col = 1 to n doBegin

C[Row,Col] := A[Row] X B[Col];End;

End;

If I have multiple planes of data (in 3D), I can deal with multidimensional data directly.

This is a NEW Machine-level Instruction!

Miss

Miss

ONLY2 Misses in 3D!

ALL Scalar Code

Is Gone!

Only 2 Registers Required: They’re purely “Structural,”

i.e., it’s a 2D structure.

This is >> Order of Magnitude Speedup, and uses FEWER registers

The Cache understands “Row” vs “Column”

How Could 3D Enrich the ISA?

Page 20: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Making the Architecture More Powerful

Why doesn’t the ISA or the cache “understand” these structures?

Rec

RecRec

RecRecRec Rec

Rec

Rec Rec

Rec Rec

……

Header

RightTree

LeftTree

ManyMBytes

Fields that we’ll touch

1 Record

DynamicallyDefined

Cache Block

Page 21: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Topography(looking down)

One Layer of this Slice Stored as a Binary

Image on the 2D Plane Corresponding to a

Fixed Altitude

1sss

0s everywhereelse

???

Manipulation of Topologies

Facial Recognition

Topographical Maps

Page 22: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

Power vs Frequency

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

Normalized Frequency

No

rmal

ized

Po

wer

Power vs. Frequency

Frequency (Normalized to 4 GHz)

Po

wer

(N

orm

aliz

ed)

50%

25%

Energy Efficient

Operation

High SpeedOperation

Slope = 3.5

Slope = 1.1

L3P

PL3

PL3

L3P

L3P

PL3

PL3

L3P

L3P

PL3

PL3

L3P

L3P

PL3

PL3

L3P

p7p7

3D Core-on-Core3D Core-on-CoreStudyStudy

11

22

Page 23: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

What is “Architecture” ?The purpose of “Architecture” is to create a good interface between applications and technology – with a classical understanding of applications, and a contemporary grasp of the technology.

Both evolve. And each drives the other.

Architecture requires broad technical knowledge, coupled to a lively imagination, and what I’ll call an “Artistic Sense.” You can calculate parts of what you need, but sheer “facts” are insufficient.

Success: Is it a Blessing or a Curse?

Page 24: Space: The Final Frontier 1966-1969 The Future of Computer Architecture Babylonian Dynasty 1894-1595 BC Phil Emma Chief Scientist IBM TJ Watson Research

“A great building must begin with the immeasurable, must go through measurable means when being designed, but in the

end must be unmeasured.” – Louis Kahn

“Architecture should have little to do with problem solving - rather it should create desirable conditions and opportunities

hitherto thought impossible.” – Cedric Price

“If you want to get rich from writing, then write the sort of thing that’s read by persons who move their lips when

reading to themselves.” – Don Marquis

“I want a man who is kind and understanding. Is that too much to ask for in a millionaire?” – Zsa Zsa Gabor

What Kind of Research Should we Do?

Academia