Lecture 1b Technology Trends Pradondet Nilagupta , KU (Based on lecture notes by Prof. Randy Katz & Prof. Jan M. Rabaey , UC Berkeley)

Original Big Fishes Eating Little Fishes

Massively Parallel Processors 1988 Computer Food Chain Mainframe Work- station PC Mini- computer Mini- supercomputer Supercomputer Massively Parallel Processors

Massively Parallel Processors Mini- supercomputer Mini- computer Massively Parallel Processors 1998 Computer Food Chain Mainframe Work- station PC Server Supercomputer Now who is eating whom?

Why Such Change in 10 years? Function Rise of networking/local interconnection technology Performance Technology Advances CMOS VLSI dominates TTL, ECL in cost & performance Computer architecture advances improves low-end RISC, Superscalar, RAID, … Price: Lower costs due to … Simpler development CMOS VLSI: smaller systems, fewer components Higher volumes CMOS VLSI : same dev. cost 10,000 vs. 10,000,000 units Lower margins by class of computer, due to fewer services

1985 Computer Food Chain Technologies ECL TTL MOS Mainframe Work- station PC Mini- computer Mini- supercomputer Supercomputer

Technology Trends: Microprocessor Capacity “Graduation Window” Alpha 21264: 15 million Pentium Pro: 5.5 million PowerPC 620: 6.9 million Alpha 21164: 9.3 million Sparc Ultra: 5.2 million Moore’s Law CMOS improvements: Die size: 2X every 3 yrs Line width: halve / 7 yrs

Memory Capacity (Single Chip DRAM) year size(Mb) cyc time 1980 0.0625 250 ns 1983 0.25 220 ns 1986 1 190 ns 1989 4 165 ns 1992 16 145 ns 1996 64 120 ns 2000 256 100 ns

CMOS Improvements Die size 2X every 3 yrs Line widths halve every 7 yrs 25 Die size increase plus transistor count increase 20 15 Transistor Count 10 5 Line Width Improvement Die Size 1980 1983 1986 1989 1992

Future DRAMs (Spectrum 1/96) Year Smallest Die Size Bits/Chip Feature m sq. mm (Billions) 1995 .35 190 .064 1998 .25 280 .256 2001 .18 420 1 2004 .13 640 4 2007 .10 960 16 2010 .07 1400 64

Memory Size of Various Systems Over Time 128K 128M 2000 8K 1M 8M 1G 8G 1970 1980 1990 1 chip 1Kbit 640K 4K 16K 64K 256K 4M 16M 64M 256M DOS limit 1/8 chip 8 chips-PC 64 chips workstation 512 chips 4K chips Bytes Time

Technology Trends (Summary) Capacity Speed (latency) Logic 2x in 3 years 2x in 3 years DRAM 4x in 3 years 2x in 10 years Disk 4x in 3 years 2x in 10 years

Processor frequency trend Frequency doubles each generation Number of gates/clock reduce by 25%

Processor Performance Trends Microprocessors Minicomputers Mainframes Supercomputers 0.1 1 10 100 1000 1965 1970 1975 1980 1985 1990 1995 2000 Year

Performance vs. Time 0.1 1.0 10 100 Performance (VAX 780s) 1980 1985 Mips 25 MHz 0.1 1.0 10 100 Performance (VAX 780s) 1980 1985 1990 MV10K 68K 780 5 MHz RISC 60% / yr. uVAX 6K (CMOS) 8600 TTL ECL 15%/yr. CMOS CISC 38%/yr. o | MIPS (8 MHz) 9000 (65 MHz) uVAX CMOS Will RISC continue on a 60%, (x4 / 3 years)? 4K

Processor Performance (1.35X before, 1.55X now) 1.54X/yr

Performance Trends (Summary) Workstation performance (measured in Spec Marks) improves roughly 50% per year (2X every 18 months) Improvement in cost performance estimated at 70% per year

Instructions/Second/$ vs. Time 1999 1997 1995 1993 1991 1989 1987 1985 1983 1 10 100 1000 10000 100000 Mainframe Super-mini Workstation Instructions/second/$ PC

Processor Perspective Putting performance growth in perspective: IBM POWER2 Cray YMP Workstation Supercomputer Year 1993 1988 MIPS > 200 MIPS < 50 MIPS Linpack 140 MFLOPS 160 MFLOPS Cost $120,000 $1M ($1.6M in 1994$) Clock 71.5 MHz 167 MHz Cache 256 KB 0.25 KB Memory 512 MB 256 MB 1988 supercomputer in 1993 server!

Where Has This Performance Improvement Come From? Technology? Organization? Instruction Set Architecture? Software? Some combination of all of the above?

Performance Trends Revisited (Architectural Innovation) 1000 Supercomputers 100 Mainframes 10 Minicomputers Microprocessors 1 CISC/RISC 0.1 1965 1970 1975 1980 1985 1990 1995 2000 Year

Performance Trends Revisited (Technology Advances) Logic Speed: 2x per 3 years Logic Capacity: 2x per 3 years Leads to: Computing capacity: 4x per 3 years If can keep all the transistors busy all the time Actual: 3.3x per 3 years

Performance Trends Revisited (Microprocessor Organization) Bit Level Parallelism Pipelining Caches Instruction Level Parallelism Out-of-order Xeq Speculation . . .

What is Ahead? Greater instruction level parallelism? Bigger caches? Multiple processors per chip? Complete systems on a chip? (Portable Systems) High performance LAN, Interface, and Interconnect

Hardware Technology 1980 1990 2000 Memory chips 64 K 4 M 256 M-1 G 1980 1990 2000 Memory chips 64 K 4 M 256 M-1 G Speed 1-2 20-40 400-1000 5-1/4 in. disks 40 M 1 G 20 G Floppies .256 M 1.5 M 500-2,000 M LAN (Switch) 2-10 Mbits 10 (100) 155-655 (ATM) Busses 2-20 Mbytes 40-400

Software Technology 1980 1990 2000 Languages C, FORTRAN C++, HPF object stuff?? Op. System proprietary +DUM* +DUM+NT User I/F glass Teletype WIMP* stylus, voice, audio,video, ?? Comp. Styles T/S, PC Client/Server agents*mobile New things PC & WS parallel proc. appliances Capabilities WP, SS WP,SS, mail video, ?? DUM = DOS, n-UNIX's, MAC WIMP = Windows, Icons, Mouse, Pull-down menus Agents = robots that work on information

Computing 2001 Continue quadrupling memory every 3 years 1K chip in 72 becomes 1 gigabit chip (128 Mbytes) in 2002 On-line 12-25 Gigabytes; $10 1-Gbyte floppies & CDs Micros increase at 60% per year ... parallelism ญ100 Radio links for untethered computing

Computing 2001 Telephone, fax, radio, television, camera, house, ... Real personal (watch, wallet,notepad) computers We should be able to simulate: Nearly everything we make and their factories Much of the universe from the nucleus to galaxies Performance implies: voice and visual Ease of use. Agents!

Applications: Unlimited Opportunities Office agents: phone/FAX/comm; files/paper handling Untethered computing: fully distributed offices ?? Integration of video, communication, and computing: desktop video publishing, conferencing, & mail Large, commercial transaction processing systems Encapsulate knowledge in a computer: scientific & engineering simulation (e.g.. planetarium, wind tunnel, ... )

Applications: Unlimited Opportunities Visualization & virtual reality Computational chemistry e.g. biochemistry and materials Mechanical engineering without prototypes Image/signal processing: medicine, maps, surveillance. Personal computers in 2001 are today's supercomputers Integration of the PC & TV => TC

Challenges for 1990s Platforms 64-bit computers video, voice, communication, any really new apps? Increasingly large, complex systems and environments Usability? Plethora of non-portable, distributed, incompatible, non-interoperable computers: Usability? Scalable parallel computers can provide “commodity supercomputing”: Markets and trained users?

Challenges for 1990s Platforms Apps to fuel and support a chubby industry: communications, paper/office, and digital video The true portable, wireless communication computer Truly personal card, pencil, pocket, wallet computer Networks continue to limit: WAN, ISDN, and ATM?

A glimpse into the future Silicon in 2010 Die Area: 2.5x2.5 cm Voltage: 0.6 - 0.9 V Technology: 0.07 m 15 times denser than today 2.5 times power density 5 times clock rate

Source: Richard Newton What is the next wave? Source: Richard Newton

The Embedded Processor What? A programmable processor whose programming interface is not accessible to the end-user of the product. The only user-interaction is through the actual application.

Examples: Sharp PDA’s are encapsulated products with fixed functionality - 3COM Palm pilots were originally intended as embedded systems. Opening up the programmers interface turned them into more generic computer systems.

Some interesting numbers The Intel 4004 was intended for an embedded application (a calculator) Of todays microprocessors 95% go into embedded applications SSH3/4 (Hitachi): best selling RISC microprocessor 50% of microprocessor revenue stems from embedded systems

Some interesting numbers (cont.) Often focused on particular application area Microcontrollers DSPs Media Processors Graphics Processors Network and Communication Processors

Some different evaluation metrics Components of Cost Area of die / yield Code density (memory is the major part of die size) Packaging Design effort Programming cost Time-to-market Reusability Power Cost Flexibility Performance as a Functionality Constraint (“Just-in-Time Computing”)