Transcript powerpoint

Microprocessor
Architecture meets
Maryann, the Steam
Shovel
Bob Colwell
Intel Corporation
®
R
ISCA, June 2000
1
1
Comp Arch is not out of ideas
 The
more they buy the faster we dig
– we can keep making faster computers
– memory is falling way behind, but there are
ways to deal with that
OoO execution, value prediction, caches, prefetch,
trace caches, memory renaming, better compilers,
better apps, data streaming, multithreading…
®
R

But are they buying because we’re digging?

Or are they tolerating us for lack of better
alternatives?
2
2
A Message From Nature
When the best improvements are only
worth a few percent
and they are incredibly complex
and they interact with each other
and they make other metrics worse (thermals)
and users can’t see the difference anyway
Then you are on thin ice of diminishing
returns -- find a new game to play
®
R
3
3
Other Possible Good Goals
 Thermals
 Cost
 User
interface
 Maintainability
 Dependability
 Mobility, battery life
 Boot time
 I/O response time
®
R
4
4
The Problem of Scale
 Big
Systems do not scale from small
 Must design differently
– Root causes: system complexity, physics
– Human body, telephone system, internet
– No single points of failure
– End-to-end checking, judicious redundancy
– Overall system “understands” user’s intent
 Where
is this point?
 Will uP’s hit it? Have they already?
®
R
5
5
What’s needed
 Need
to develop new uArch technology
–
–
–
–
Robust computing in face of SER
Robust in face of design flaws
Robust in face of determined hackers
Use all those transistors to really improve
machines, not just make them trivially faster
– Extra credit: robust in face of SW bugs
– Let performance increases come from process
Match what people want from computing
®
R
6
6
A little light reading
Normal Accidents, Charles Perrow, Princeton University Press, 1984, 1999
An Investigation of the Therac-25 Accidents, Nancy Leveson & Clark Turner, IEEE Computer V 25, N 7, p 18
The Invisible Computer, Donald Norman, MIT Press, 1998
The Limits of Safety, Scott D. Sagan, Princeton University Press, 1993
To Engineer Is Human, Henry Petroski, St. Martin’s Press, 1982
The Challenger Launch Decision, Diane Vaughan, The University of Chicago Press, 1996
Searching For Safety, Aaron Wildavsky, Transaction Publishers, 1991
Why Buildings Fall Down, Matthys Levy and Mario Salvadori, Norton & Co., 1992
Why Things Bite Back: Technology and the Revenge of Unintended Consequences, Edward Tenner, Knopf, 1996
Beyond Engineering: How Society Shapes Technology, Robert Pool, Oxford University Press, 1997
Flying Buttresses, Entropy, and O-Rings: The World of an Engineer, James L. Adams, Harvard Press, 1991
Societal Risk Assessment: How Safe Is Safe Enough?, Richard Schwing and Walter Albers, Plenum Press, 1980
Controlling Technology, Stephen H. Unger, Wiley Interscience, 1994
Inside the Sky, William Langewiesche, Vintage Books, 1998
On The Right Lines? The Limits of Technological Innovation, Stephen Potter, St. Martin’s Press, 1987
The Hubble Wars, Eric J. Chaisson, HarperCollins Publishers, 1994
The Blunder Book, M. Hirsh Goldberg, Quill, 1984
The Control of Nature, John McPhee, The Noonday Press, 1989
The Civilized Engineer, Samuel C. Florman, St. Martin’s Press, 1987
The Mythical Man-Month, Frederick P. Brooks Jr., Addison-Wesley, 1979
What Engineers Know and How They Know It, Walter G. Vincenti, Johns Hopkins University Press, 1990
Fluid Concepts and Creative Analogies, Douglas Hofstadter, Basic Books, 1995
Computing Calamities: Lessons Learned from products, projects, and companies that failed, Robert L. Glass,
Prentice-Hall, 1999
®
R
7
7