1. Computer Systems are Different!
Frans Kaashoek and Robert Morris
6.033 Spring 2009

2. Composibility via static discipline
Be tolerant of inputs and strict on outputs

3. Moore’s law cost per transistor transistors per die
Left graph is sketch from Moore’s 1965 notebook (also 1965 article).
Right graph from 1965 article.
transistors per die
“Cramming More Components Onto Integrated Circuits”, Electronics, April 1965

4. Transistors/die doubles every ~18 months
291 Million for Conroe which is fewer than previous chip, but less power.

5. Lithography: the driver behind transistor count
Components/area O(x2) with feature size
Total components O(a) with die area
Switching rate O(x) with feature size
See article by Hennessy and Jouppi

6. CPU performance Microprocessors overtook everything around 1990
See Patterson and Hennessy

7. DRAM density

8. Disk: Price per GByte drops at ~30-35% per year
100 Gbyte for $50 -> 50 cents per GB

9. ENIAC 1946 Only one 5000 adds/sec 20 10-digit registers
18,000 vacuum tubes
124,500 watts
Not really stored program
Illustrate graph trends with selected computers from history.
First general purpose programmable electronic computer.
See Patterson and Hennessy

10. UNIVAC (Universal Automatic Computer)
1951
46 sold
2000 ops/sec
1, digit words (mercury)
5000 tubes
$1.5 million
First American commercial computer. Stored program.

11. IBM System/360-40 1964 1.6 MHz 16-256 KB core $225,000 Family of six
32-bit
Time-sharing
First upwardly-compatible family.

12. Cray 1: supercomputer 1976 80 sold 80 MHz 8 Mbyte SRAM 230,000 gates
$5 million
Most famous supercomputer, still faster than any microprocessor even in 1990.

13. DEC PDP-8 (1964) 4096 12-bit words 60,000 sold $18,000
First successful minicomputer.
bit words
$18,000
60,000 sold
330,000 adds/sec

14. Apple II 1977 1 MHz 6502 microprocessor 4 to 48 Kilobytes RAM $1300
Basic, Visicalc
First successful personal computer.

15. IBM’s wrist watch 2001 Linux and X11 74 Mhz CPU 8 Megabyte flash
8 Megabyte DRAM
Wireless

16. Software follows hardware
Millions of lines of source code
Point: we CAN build complex stuff, taking advantage of h/w opportunities
Credit: Hari Balakrishnan

17. Cheap  Pervasive

18. Pervasive  qualitative change
Number crunching
log (people per computer)
Word processing
Communication
So if we look at the computing spectrum today, each of the classes exist simultaneously. In a room we can put 100s of teraflops and many petabytes of computing.
Our productivity, document preparation, and personal information management fits in our pocket and a new class is emerging that will provide a means of streaming information to and from the physical world like we have never seen before.
Embedded
Sense/control
year
Slide from David Culler, UC Berkeley

19. Latency improves slowly
Moore’s law (~70% per year)
Improvement wrt year #1
Speed of light (0% per year)
DRAM access latency (~7% per year)
Year #

20. Heat is a problem

21. Recent Intel CPU Clock Rates
Pentium 4 HT
Pentium 4
Pentium III
mHz
PentiumPro
Pentium
486

22. The Future: will it be painful?
# gates increasing even though clock rate isn’t… How to use lots of gates to improve performance?
Maybe force big changes in how we program (or may not).
AMD Barcelona Quad-core chip

23. What went right? Unbounded composibility General-purpose computers
Only need to make one thing fast
Separate arch from implementation
S/W can exploit new H/W
Cumulative R&D investment over years