Year 2000 Computer Technology Forecast

as of May, 1996

I. Executive Overview: Personal Computers in the Year 2000
II. Historical Baseline
III. Last Year (1995) State of the Art
IV. This Year (1996)
V. Next Year (1997)
VI. Year After Next (1998)
VII. Three Years from Now (1999)
VIII. Four Years Hence (2000)

I. Executive Overview:
Personal Computers in the Year 2000:
    The following projections are culled from the trade literature. Of course, whatever else is true, the next five years won't happen exactly as is foretold here.

Year 2000 Microcomputer Speeds:
    311 SPECint92s in May, '96®1,000 to 3,000 SPECint92s and
    245 SPECfp92s in May,'96®1,000 to 3,000 SPECfp92s
in Dec., 2000.

Year 2000 On-Chip Multimedia-Extension Speeds:
    311 SPECint92s in May,'96® 1,500 to 24,000 SPECint92s (MMx) by Dec., 2000.

Year 2000 DRAM (Dynamic Random Access Memory):
    DRAM should cost $1 to $2/MB (MegaByte).
    If DRAM is much more expensive than $2 a megabyte, progress in the computer field will be impacted, and other memory technologies and/or vendors may enter the marketplace.
    Most new ($1,000) computer systems will probably ship with 128 to 256 MB of RAM.

Year 2000 Hard Drives:
    IBM has promised 90 MB hard drives by the year 2000.
    Most new computer systems will probably ship with 10 to 16 GB hard drives.

Year 2000 CDs:
    650 MB in '95® 4.3, 8.6 or 9.4, 17.2,and perhaps 80 GB (gigabytes) by 2000.
    The technology exists to ship 17 GB CDs, with 80 GB CDs running in the lab. What happens next will undoubtedly hinge upon cost factors and standardization issues. CDs are rapidly becoming the cheap distribution medium of choice for software and data. However, there is enthusiasm within the industry for the new 4.7-to-17 GB format. If 80 to 100 GB optical drives appear, they will probably be in response to IBM's 90GB hard drives.
    Most new computer systems will perhaps ship with 4.3 GB CD ROMs.

Year 2000 Graphics:
    17", 1600 X 1200-pixel displays will probably be the standard. 10.4" to 17" LCD displays may be their competitors. Twenty or twenty-one-inch monitors might be bundled with high-end systems. These displays might also be designed for true 3-D stereo viewing, with or without glasses. Virtual reality hardware such as powergloves, motion seats, and strap-on, total-immersion displays might be available at Madison Books and Computers, if not at Walmart. Large screen projection systems may become more popular because of their ability to display HDTV quality imagery from cable, direct satellite systems, broadband telephone lines, and computer-generated sources. They may be used for interactive entertainment, virtual reality simulations, virtual shopping, the virtual office, and videoconferencing.
    High end systems may ship with stereo microphones, low-cost color video cameras, and Pantone color calibration.

Year 2000 Telecommunications:
    6 Mb/s (Megabits per second) downlinks, 225 to 650 Kb/s uplinks.
    Wideband telephone and cable lines and/or satellite links should be inexpensively available to our homes and offices here in Huntsville, providing at least 6 megabaud downlinks and at least 225 to 650 kilobaud uplinks to other locations and to online services such as the Internet. (BellSouth is about to get some stiff competition from Cable Alabama.)
    Most new computer systems will probably ship with MPEG-2 or MPEG-4 data compression/decompression hardware and perhaps with 6 Mb/s modems. Conventional 33.6 kilobaud modems and fax machines could be obsolescent (on their way to obsolete)

Windows 2000:
    A highly-sophisticated, artificial intelligence-based Windows 2000 with a 3-D virtual interface will probably support limited context sensitive voice dictation (currently available on IBM PCs), limited natural language processing (currently under development by Microsoft for HELP queries), videoconferencing, virtual reality support, stereo 3-D , surround-sound home theater, network access software, and intelligent agents that, like a good spouse, adapt to your ways.
    One development that might occur over the next few years is the creation of an entirely new interface based upon artificial intelligence, natural language processing, voice input and output, and 3-D virtual reality settings, including video animations. A virtual office might be featured containing file cabinets, a videophone, an alarm clock, an appointment calender, a to-do list, a secretary to screen phone calls and visitors, to take dictation, and to remind you of appointments, a fax machine, names and addresses, and so forth. You might walk out the door of your office and down the hall to someone else's office.

Year 2000 Applications:
    Applications will include such activities as online library access (e.g.—for school children), voice dictation, natural language processing (perhaps with some AI dialogue capabilities), information retrieval of various kinds, online banking and shopping, videoconferencing and whiteboarding, network social interactions, entertainment, edutainment, video rentals, role and game playing in virtual reality settings, and many other functions that are hard to foresee.

Historical Baseline
    It may be instructive to consider what has happened in computer technology over the past thirty years.
    In 1965, the IBM 7094 II was the lion king of the computer world. It boasted 144 kilobytes of 1.1 microsecond (1,100 nanosecond) RAM (core memory), no disk drives, an operating speed of, perhaps, 0.5 to 0.75 SPECint92s, and perhaps 200,000 floating point operations (200 kiloflops) per second. Corrected for inflation, its RAM memory cost about $4,500,000 and its total cost, including tape drives and printers, was about $13,500,000.
    IBM's smallest minicomputer, the IBM 1620 Mod I, offered 10 kilobytes of 40,000 nanosecond RAM, no disk, floating point speeds of about 0.0005 SPECint92s (500 fixed-point calculations per second), 0.060 kiloflops (60 floating point calculations) per second, and 7 trigonometric calculations per second at a cost of about $500,000. It had an online card reader/punch, a console typewriter, and an offline printer. It could carry out 100 floating point multiplies in 4 seconds and could calculate 100 trigonometric functions in 15 seconds.
    Its accelerated upgrade, the 1620 Mod II, afforded 30 kilobytes of 10,000 nanosecond RAM, and delivered 5 times the computational speed at a cost of about $1,350,000. These computers obeyed Grosch' Law, which stated that computer performance increased as the square of the cost.
    By comparison, 1995's hottest desktops deliver perhaps 200 SPECint92s and 60 megaflops with 16,000-to 64,000 kilobytes of 60-nanosecond RAM, 1-to-2 gigabytes of disk storage, a high-resolution color graphics display, and an online printer for $2,500-to-$4,000. They can calculate 100 floating point multiplies in perhaps 6 microseconds. As such, they are 200-300 times faster the 7094, offer 100 to 400 times as much 60-nanoseond RAM at 1/6000th to 1/4000th the cost. On a price/performance basis, today's computers offer better than a 1,000,000:1 price/performance improvement over those of 30 years ago (50,000,000 times the price/performance ratio of the IBM 1620 Mod I and 30,000,000 times the price/performance figure of the 1620 Mod II). This translates into a doubling in price/performance every 18 months, or a 10-folding every 5 years—exactly the same rate of improvement as semiconductor memory densities. At the same time, processor chip densities, which ran 1 transistor per chip in 1965, reached the 1,000,000 to 10,000,000 transistor per chip range in 1995.

Last Year (1995) State-of-the-Art
1995 PCs:
    At the beginning of 1995, the 90 Mhz Pentium and the 80 MHz 601 PowerPC were the fastest PCs available. By the end of 1995, 133 MHz Pentiums and 150 MHz 604 PowerPC's could be purchased for about $3,000. The 150 MHz 604 at year's end was more than 2 1/2 times as fast as its 80 MHz 601 predecessor at the beginning of 1995. Also, the 200 MHz P6 hit the marketplace at 2.8 times the integer speed of the 90 MHz Pentium and 3.5 times its floating point performance.
    During 1995, magnetic disk prices nearly halved, with 9 GB disks dropping from about $3,500 to about $2,100. Iomega introduced its $200 100 MB ZIP drive and its $600 1 GB Jazz drive. CDs went from 300 KB/sec. (2X) transfer rates to 1.2 MB/sec. (8X) transfer rates. Magneto-optical disks transitioned from 1.3 GB to (less expensive) 4.3 MB disks. DRAM prices remained unchanged and may even have risen slightly. 1995 was the year that the Internet went big-time.

This Year (1996)
The Seven Dwarfs
    DEC will introduce a 300 MHz a chip, and a 333 MHz version generating 400 SPECint92s and 600 SPECfp92s, and perhaps, a 400 MHz version good for 480 SPECints and 720 SPECfps during 1996. The MIPS R10000 is promising 300 SPECint92s and 600 SPECfp92s, the UltraSPARC I affords 260 and 410 SPECs, respectively, and the HP8000 should run at 360 and 550 SPECs. The Motorola-IBM-Apple consortium is eating dust at about 225 SPECint92s, while Cyrix and NexGen/AMd are hoping for 300 SPECint92s out of their K6 chips.

Snow White's (Intel's) Pentiums
    Intel plans to introduce Native Signal Processing (NSP) on the Pentium P55C, mounted on the "Garcia Board", in the latter half of 1996. Native Signal Processing allows the Pentium's 64-bit wide registers to simultaneously process two 32-bit, four 16-bit, or eight 8-bit integers. Overflows are handled by placing the largest integer (all 1's) in any sub-register that has an overflow. This technique will double (32-bit), quadruple (16-bit), or octuple (8-bit) signal processing speeds, permitting voice dictation, arcade quality graphics, and MPEG-2 (Motion Picture Experts Group 2) video capabilities to be designed into mainstream applications (since software designers can now count on these enhanced signal processing capability being present in all future 80X86 microprocessor chips). Pentium chips with clock speeds of 150 MHz, 166 MHz, and perhaps 180 MHz clock speeds should be on the market by year's end.

Snow White's P6
    Native Signal Processing will also be available on Intel's new Pentium Pro (P6) chips. These will be available with 200 MHzÆ 233 MHzÆ 266 MHz clock speeds. The first generation, 150 MHz chips embodied 0.6 µ design rules, leading to a large die size and equating to high prices. The second generation of chips is now being manufactured, using 0.35 µ design rules, and is priced at $1,325 (compared to the Pentium's introductory price of $875). NexGen/AMD (AdvancedMicro Devices) is also planning 180 MHz 80686 (K6) chips fabricated in 0.35 µ silicon, with 6 million transistors and, at 300 SPECint92s, running faster than Intel's P6 was expected to run when NexGen made that claim, with volume production beginning in 1997. These will also possess multimedia instructions and AMD claims speeds up to 6 Gigops(!?) for these instructions. (We note, however, that Intel's currently-available 200 MHz P6 delivers 311 SPECint92s and 245 SPECfp92s and is available today rather than in 1997. This may remove some of the luster from the very pricey HP-8000 and MIPS R10000 workstations, since they are slower than the 200 MHz P6 and they are not even on the market yet. This will be expecially true if later in the year, Intel drops the other shoe and boosts the P6 clock rate to 266 MHz, thereby raising the SPECint count to 513 and the SPECfp reading to nearly 400. Has Intel blown its competitors out of the water?) The P6 might well be expected to appear in two and four processor configurations, since it is designed from the ground up for multiprocessor operation. (Intergraph is providing up to four clusters of four P6 processors for a total of 16 processors.) It is touted for high-end servers, as were the original Pentium Processors, but is being marketed for desktop systems as well, at a price of about $3,500.

The PowerPC
    IBM hopes to introduce a beefed up version of Motorola's 620 PowerPC chip. In the meantime, a 150 MHz 604e PowerPC chip is available, delivering performance somewhat below (225 SPECints) that of the 150 MHz P6 from Intel. Apple has teamed with a small startup called "Exponential" to try to triple the performance of the 604 over that of a 133 MHz Pentium by developing a bipolar chip with a 400 MHz clock speed. This should deliver 600 SPECint92s. It had better. Apple has bet the farm on the PowerPC.
    Prototypes are anticipated late this year and production is projected to begin next year (1997).

The Common Hardware Reference Platform
    Apple and IBM will begin to market their PowerPC CHRP (Common Hardware Reference Platform), which can run several popular operating systems, including Windows, DOS, and the Apple OS. Apple will introduce its Copeland operating system.

The Internet Browser
    Can an Internet browser be built for $500? Of course it can! The Commodore 64 only cost $200 back in 1982. A modified word processor with a floppy drive would be more than adequate for text inquiries on the Internet. A basic processor with 2.5 to 4 MB of RAM, a 13" color graphics monitor, a small disk drive, a keyboard, a mouse, and a 14.4 kilobaud modem might be fielded for $500. Apple's Pippin, which includes 603 chip, 6 MB of RAM, a 4X CD, and uses an enhanced TV set for display might come in at a price not too far above $500.

Dynamic Random Access Memory (DRAM)
    The worldwide DRAM shortage is expected to continue, with DRAM prices hovering around $30/megabyte. By year's end, most new PC's will probably be sold with 16 megabytes of RAM. A shift to Synchronous Dynamic Random Access Memory (SDRAM) may develop over the year, along with a policy of using main RAM for video refresh.

Flash Memory
    Flash memory is electrically-erasable, programmable, "read-only" memory. Like magnetic media, it is non-volatile and therefore, could function as a plug-in memory for portable electronic equipment. Its only limitation is that it has a lifetime measured in millions of read-write cycles and might eventually wear out. Flash memory will continue to expand in the niche market of portable computer systems and other devices.

Magnetic Disks
    By year's end, most new PC's will probably come with 1.6 to 2 gigabytes of magnetic disk storage. MPEG-2 decoders will probably appear during the year. Speaker-independent, continuous-speech voice recognition may be available during the year.
    At some point, 2:1 on-the-fly disk data compression/decompression may become practical (requires super-fast hardware). Data compression/decompression is currently available at a 150 kB/sec. rate for magnetic tape back-up.

CD's (Compact Disks)
    CD capacity is forecast to jump to from 650 megabytes to 4.3 or 4.7 gigabytes during the year, as digital video disks (DVD’s) enter the marketplace.

    Seventeen-inch monitors should gradually become more prevalent, as fifteen inch monitors continue to crowd out fourteen inch monitors on new computers. Color, flat-screen, liquid crystal displays (LCDs) are appearing on the market at about twice the price of CRT's.
    Three-D graphics formats are expected to become standardized, and computer-based games should begin to rival dedicated Sega Saturn, Nintendo Ultra, and Sony gaming units. Videoconferencing (VC) hardware for POTS (Plain Old Telephone Systems) and for ISDN (Integrated Services Digital Network) services will become more available, at prices ranging from $250 for a Connectix color system to $2,500 for a top-of-the-line Picteretel VC. Because of the Internet frenzy, virtual reality, as well the demand for wideband data communications, will rapidly become more significant market forces, especially in the business world.
     The 4 gigops video signal processor from Philips should supplant the 2 gigops digital signal processor from Texas Instruments as the Fastest Graphics Chip in the West.

    33.6 kilobaud modems will be available this year, now that a 33.6 kilobaud communications standard has been established. In the meantime, the widening distribution of ISDN, the advent of cable modems, and the approaching Asynchronous Digital Subscriber Line (ADSL) lines from the Baby Bells should begin to cut a little into the market for POTS modems.

Networking and Telecommunications
    It has been announced that ultra-fast cable modems and interactive Internet access will become available in the Huntsville area from Cable Alabama. In the face of this dangerous competition, BellSouth may be expected to at least promise ADSL service in the Huntsville area, providing 6 megabaud downlink service and 225 to 650 kilobaud uplink data rates. Competition may be expected to force the Bell Operating Companies to accelerate their conversion to wideband communcations. Otherwise, their kingdoms will be divided amongst the Medes and the Persians.

    Online shopping, banking, and a myriad social and business transactions will continue to expand, as color images, virtual reality simulations, and live video are more readily transmitted through higher speed communications lines. Advertisers may begin to provide an advertising revenue base for Internet upgrades, as the Internet expands to accommodate real-time image and voice transmission. Library functions, especially for children, will probably be satisfied more and more over the Internet. Who needs to buy an encyclopedia, a thesaurus, an unabridged dictionary, or other books when they are instantly available over the Internet? Video rentals may begin over the telecommunications lines rather than from video rental stores—i.e., the video rental stores will go online. Of course, none of this will take place overnight but will occur over a several year period.

Next Year (1997)
The Seven Dwarfs, or However Many Are Left By This Time
    MIPS' H4 and HP's Precision Architecture-9000 processors are slated to sample during the fall of 1997 and to proffer 1,000 SPECint92s and 2,000 SPECfp92s. (Wow! A supercomputer on a chip!) If history repeats itself, these processors will come in at perhaps 700-800 SPECint92s and will slide into 1998. The 300 MHz UltraSPARC II will put out 440 SPECint92s and 660 SPECfp92s, making it a toy in this speed-demon world. Only its Visual Instruction Set (Native Signal Processing) can hope to save it. DEC is going to be hard-pressed to retain its role as the purveyor of the world's fastest microprocessors. No doubt DEC has something up its sleeve.

Meanwhile, Back at the Castle
Snow White's Pentium
    The Pentium should reach clock speeds in the 200-250 MHz range during the year, unless it is retired from service at its 180 MHz clock rate. There should be competition from second sourcers such as NexGen/Advanced Micro-Devices and Cyrix.

The P6
    The P6 might also appear in a 333 MHz version during 1997. By now, it may be replacing the outdated Pentium processor as the mainstream micro of the day. Multiprocessor versions may become popular for high-end network server applications.

The P7:
    Intel's P7 chip is scheduled for its debut in Spring, '97. If the past is prologue, it may well drift into late '97 or early '98, though well worth the delay. Plans call for a very-long-instruction-word (VLIW) format that would deliver a breakthrough in performance, although this might be difficult to deliver. It might well be introduced with a 300 MHz clock speed and might deliver 600 SPECint92s. Or it might execute in parallel much faster than that. It will have 128-bit wide data paths, compared to 64 bits for the P5 and P6, and will probably incorporate 10-15 million transistors. AMD plans to introduce its 400 MHz, 700 SPECint92s, K7 chip, fabricated with 10-15 million-transistors using 0.18 µ technology.
    1997 ought to see 0.35 µ fabs and 0.25 µ pilot fabs (and obviously, at least one 0.18 µ AMD fab) come online.

The PowerPC
    IBM, Apple, and Motorola will have to crank up their PowerPC clock rates to remain competitive with Intel.
    Apple's effort to leapfrog the Pentium-P6 family may bear fruit by now and furnish 400 MHz clock speeds, yielding 600 SPECint92 performance or better. We can probably look for Native Signal Processing instructions.
    Apple might offer the Daystar Gemini 2-to-4 processor system, perhaps using the revamped 620 chip.

    The 1.2 terops Cray and the 1.8 terops Intel Touchstone supercomputers are supposed to go into operation next year (1997).

The Internet Browser
    With the anticipated decline in RAM prices, an Internet browser with 8 MB of RAM, a 14" monitor, a 500 MB hard drive, a keyboard, a mouse, and a 28.8 kilobaud modem ought to be readily available for $500. Alternatively, the 1996 browser ought to be producible for about $300. Of course, with what will be available on the Internet at this time, users may be willing to shell out the money for the best hardware they can afford. Still, there may be commercial functions (e.g., online shopping, banking, E-Mail, voice-mail, word processing) for which the $300 to $500 units would be genteel sufficient.

Magnetic Disks
    Magnetic disk drive capacities may move into the 25-35 GB range as IBM boosts disk densities by a factor of 3 to 4. Most new computers will probably ship with 2-to-4 gigabyte drives.

    Memory prices should begin to erode, dropping to $12 to $15 a megabyte by year's end. The 256 megabit RAM chip may sample toward the end of next year. Minimum RAM requirements will probably run 32 megabytes, with 64 megabytes recommended.

Flash Memory
    Flash memory may move into the mainstream by this time. So far, flash memory has been "Always a bridesmaid, never a bride". If flash memory doesn't make it big in 1997, it might prove to be "a flash in the pan".
Ferroelectric RAM:
    Like flash memory, ferroelectric RAM is non-volatile, but unlike flash memory, it may be cycled billions of times, rendering it a candidate for mainline RAM. It will initially be expensive and less dense than DRAM. Next year, it will probably be in the prototyping phase. It could be attractive because it doesn't consume power or generate heat the way dynamic RAM chips do, making it well-suited to battery-operated devices.
    It is scheduled to debut next year in a Hewlett-Packard laptop.

    So much depends upon marketing decisions that it is hard to predict, but CDs may reach their double-sided or double-layered, technically feasible level of 8.6 GB.

    Multi-megabit cable and telephone services should expand rapidly as they compete with each other for turf and as the appetite for high bandwidth grows. ADSL lines ought to be cheaply available here in Huntsville as BellSouth and Cable Alabama duke it out. Voice, video, interactive virtual reality scenarios, and game and role-playing might be operating via the Internet, not to mention independent videoconferencing, online shopping, and banking.

    33.6-kilobaud modems will probably still be prevalent, although cable, ISDN, and ADSL will be pointing the way to their eventual obsolescence. Wideband communications will probably carry a cost premium, and many computer users may not be prepared to fork over the extra money for extra service.

    Three-D graphics and virtual reality should be in high gear mext year, using your 333 MHz, 2 gigops, 440 Mflops (million floating point operations per second) P6, or with specialized digital signal processors processors in the multi-gigops range. Computer graphics faces a several-fold bump-up in speeds, as RAMBUS and Native Signal Processing hit the mainstream. Most new mid-to-upper range systems will probably come with 17" monitors. True 3-D stereo displays might enter the marketplace by this time. Thirteen inch LCD displays may be cost-competitive with 17" CRTs, and, with slightly higher resolutions, may be well suited to 3-D stereo displays. As LCD displays become relatively cheap, large-screen projection displays may become popular, particularly for videoconferencing. Four projectors could be mosaicked together to provide HDTV resolutions in conjunction with the wideband lines that are becoming ubiquitous. Digital Video Disks, coupled with MPEG2 decoding capability, ought to make possible high-resolution video movies in the HDTV range. Want to make some money? Start recording old movies on optical disks at HDTV-level resolution using MPEG-2 data compression. High speed access to graphics information sources will probably be one of the forces driving this market. Next year's computer graphics will probably rival arcade game displays.

    Continuous speech recognition, natural language processing, videoconferencing, virtual reality 3-D displays, and data compression/decompression interfaces to wide-band communications links may be standard in new computers hitting the market. Computers might be used as telephones and digital TV displays, perhaps in conjunction with large-screen projection systems. All of the online applications for the computer appliance will continue to expand. Travel will be reduced as we begin to videoconference and continue to access information online. Using the built-in NSP hardware, desktop videoconferencing systems should be available next year for prices ranging from $200 to $1,000.

Year after next (1998):
The Seven(?) Dwarfs
    By 1998, the H4 and the HP 9000 should be rolling off the production line and may hit their announced targets of 1,000 SPECint92s. In any case, the bar will have been raised to the 1,000 SPECint92 and 2,000 SPECfp92 level of processing speeds, forcing competitors to ante up or cash in their silicon chips. By now, Native Signal Processing feature probably be a standard feature on new processors. Presumably, the UltraSPARC and DEC product lines will be matching or exceeding these performance specifications, while Cyrix, Motorola, and NexGen/AMD will be targeting the 80X86 specifications. 2,000 to 2,500 SPECint92 processors should be in the pipeline for the 1999-2000 time frame, perhaps using multiple processors and/or 1,000 MHz clocks.

Back at the Castle with Snow White
The Pentium
    By now, the Pentium should be sharing the fate of the 486 processor family, with 200 to 333 MHz P6's taking over.

The P6
    This chip, now in its heyday, should be in its third generation, using 0.25 µ design rules. It should be sufficiently cheap to permit the use of two to four processors for small servers and perhaps for offloading digital signal processing. Clock speeds should have climbed to 400 MHz, with performance in the 750 SPECint and 455 SPECfp range. The chip might feature an onboard digital signal processor to help keep up with the Seven Dwarfs.

The P7
    The P7 will be relatively expensive, possibly appearing first in 0.35 µ silicon with a large die size. It will initially be targeted for high-end servers, but a second generation chip fabricated in 0.25 µ silicon might logically be expected by year's end. It might even be produced in 0.18 µ silicon although that would be moving pretty quickly. One would expect it to match or exceed the 400 MHz clock rate of the AMD K7. Its performance is hard to predict since it depends upon a VLIW design breakthrough. Its success depends perhaps in large part upon the development of software that can make proper use of the parallelism and pipelining capabilities of the hardware. However, with 1,000/2,000-SPEC µ's coming from other chipmakers, can Intel be far behind?
    Because of unanticipated delays, we might forecast that AMD doesn't begin production of its K7 chip until the latter part of the year and initially delivers at 800 SPECin92 s.

The PowerPC
    IBM, Apple, and Motorola took a RISC (Reduced Instruction Set Computer) and so far, it hasn't paid off very well for IBM. They bet that Intel couldn't match RISC speeds with its CISC (Complete Instruction Set Computer) architecture, and yet, couldn't afford to abandon its 80X86 instruction set. They bet wrong. Now they are going to have to hustle to catch up with Intel. However, they will have to shoot for chips that will match the MIPS H4 and the HP 9000 RISC chips, since that's going to set the new baseline for microprocessor performance. Apple's gamble on Exponential includes a down-the-road attempt at 700 MHz clock speeds. Either these high clock speeds or multiple processors would seem to be needed to remain competitive against Intel. Achieving that in 1998 would probably keep the Apple-IBM-Motorola Troika competitive.

    Although it is very difficult to predict supply and demand, a glut of new fabs should have come online, leading to excess world wide capacity. Hopefully, this will cause DRAM prices collapse, dropping to, perhaps, the $4 per megabyte range. If this happens, it will come in the nick of time, as RAM requirements continue to swell. New computers may ship with 64 megabyte to 128 megabyte RAM complements, driven by rapidly growing RAM requirements for voice dictation, teleconferencing (e.g., over the Internet), home entertainment, and online access.
    If price drops in DRAM don't occur, then other memory technologies such as ferromagnetic, holographic, and even flash memory may step up to the plate to satisfy the demand.

Flash Memory:
    Flash memoryshould be available in the $4/megabyte (MB) price range, perhaps in the form of small, non-volatile solid state disks, and as plug-in cards for laptops and many other devices.

Holographic Memory
    Holographic memory has the potential to store terabytes of information with microsecond access times. Its status year after next is hard to know.

Hard Drives
    New systems might ship with 4 to 6 GB hard drives, with 25 to 35 GB drives dropping in price.

Optical Disk Memory
    At some point over the next few years, laser disk memories should reach 17.2 GB, using double-layered, double-sided disks. Whether that will be 1997 or the year 2000, it is coming soon to a CD player near you.
    In the meantime, erasable optical storage in the 9 to 18 GB range should be available and not too expensive.

    Graphics trends will undoubtedly consist of more of the same. Driven by video games, interactive drama, edutainment, and yes, probably pornography, displays will continue to improve.

    3-D stereo displays should surely be available in this time frame, either with or without glasses. The first experimental HDTV broadcasts are expected to take place year after next, and these should dovetail with the computer activities that will then be underway. Seventeen inch monitors should occupy the position that fifteen inch monitors enjoy now. Twenty inch monitors may be cheap enough to be bundled with some high end systems. Surround sound may be widespread. During this period, computers might double as early HDTV receivers, although the aspect ratios, the small screens, and perhaps, the interlace characteristics don't favor this stopgap solution. TV sets should logically interface with home computers, which may use them as monitors to permit "window wall" operation. HDTV sets will contain their own digital processors. By now (year after next), the hardware (high-capacity optical or magnetic disks) should be in place to show video movies at HDTV or near-HDTV resolutions, using desktop monitors or high-grade projection displays. This may afford the first major exhibition of HDTV capabilities. Also, the hardware may be available to permit transmitting HDTV-caliber imagery over standard telephone lines or video cable. Want to make some money? Just store some good movies on Digital Video Disks at higher-than-VCR resolutions.

Entertainment and Virtual Reality
    As TV displays become digital, perhaps using large screen projection systems, the line between computer monitors and TV displays will probably blur further than it already has. Virtual reality hardware other than visual and audio output might be selling at Walmart—perhaps tactile, kinesthetic, head, and direction-of-gaze sensors, together with powerglove and motion-seat output devices. A camouflaged fan might simulate the wind on your face. With the high-speed graphics processors that will be available for home use, virtual reality, coupled with over-the-net game and role playing, should be a seductive form of entertainment and edutainment. If done properly and if it sells well, one might learn history and other subjects while having a rip-roaring good time.

Videoconferencing and Whiteboarding
    Small-screen videoconferencing should be taking place over office LANs and wideband lines. Large-screen videoconferencing systems would seem to be a logical, not-terribly-expensive approach to videoconferencing for both the home and the office. Office videoconferencing equipment might run $200 or less, since the computer should already come equipped with everything but the color camera.

    The applications will probably consist of a continuation of the applications described above. (The unforeseen applications of these computers are difficult for me to foresee.)

Three-to-Four Years From Now (1999)
Intel's P6
    If the past is a valid precursor, one might expect the P6 to be the aging dowager of the 80X86 line, with a third-generation, 0.18 µ P7 appearing before the end of 1999. By this time, the P6 might be available in a 500 MHz edition. One alternative strategy to extend the life of the P6 might be to incorporate two to four P6 processors in one chip, since the P6 is designed for multiprocessor operation. Two processors would require 11,000,000 transistors (assuming complete independence), while four processors might demand 22,000,000 transistors. These multiprocessor chips would provide a maximum of 1,930 and 3,860 SPECint92s, and 1,440 and 2,880 SPECfp92s respectively. The first (11,000,000-transistor) option would appear to be economically feasible by 1999. Another strategy might be that of packaging one or more onboard digital signal processors to handle graphics, natural language processing, speech recognition and dictation, data compression/decompression, and other computationally intensive functions. Of course, many complex factors such as cooling, interconnections, packaging, bus structure, and cache memory complications must surely enter into such design tradeoffs.

Intel's P7:
    We might expect the P7 to be be crowding out the P6 now as the workhorse of the Intel-compatible PC world. Its clock speed might hit 500 MHz and it might provide 900-1,000 SPECint92s, with the possibility that a VLIW implementation might reach considerably higher throughputs. Bus speeds of 200 MHz have been forecast for the year 2000 time frame.

Intel's P8
    Assuming that there will be a P8, it might appear by the end of 1999, using 0.18 µ design rules, and might carry, perhaps, 30 to 40 million transistors. Clock speeds of 600 MHz might be in place by this time, with chips that deliver more than 1,000 SPECint92s (and perhaps much more).

The PowerPC
    Given some architectural improvements culled from it competitors, a 700 MHz PowerPC might keep The Troika competitive through 1999.

    DRAM should be selling for $2 to $4 a megabyte, and new computer systems might be shipping with 64 to 128 megabytes of RAM, 5 GB hard drives, and 18.8 GB CDs.

Flash Memory, Ferroelectric Memory, Holographic Memory
    It is very difficult to forecast the future of these memories, since they are in a research phase at this time. Flash memory should be in widespread use in portable devices. The future of ferroelectric memory may depend upon the need to restrain power consumption and heat generation. Holographic memories are truly a dark horse, since they are currently a long way from an engineering prototype, much less a production model.

Hard Drives
    IBM's 90 gigabyte hard drives might appear by this time, with 8 to 10 GB hard drives shipping in new systems.

    This is anybody's guess. 80 GB CDs were running in the lab in 1995. The Japanese have established a program to produce a terabyte optical drive. No obstacles to progress in optical drives is foreseen for the next 15 years so the technology should be there. And if magnetic drives reach 90 GB, optical drives had better keep pace.

    Nothing anticipated here but steady progress. Laser holographic displays (which require enormous computational resources) might enter the picture for proof of principle and specialized applications. Other emerging display technologies such as plasma panels, Texas Instruments' Digital Video Mirror Device, laser-optical and field emission displays might be penetrating specialized markets. Displays with resolutions exceeding 1600 X 1200 should be possible by mosaicking LCD projection screens. HDTV-resolution imagery may be making inroads in our standards of high quality video even without direct broadcasts.
    Large-screen videoconferencing would seem to be quite attractive as a complement to desktop video interactions. All of this will be an integral part of virtual reality technology, which will be the enabling technology for various kinds of workaday activities. Videoconferencing, the examination of items in online catalogs, virtual travel, the virtual office, virtual walkthroughs, interactive video entertainment, and the operation of your computer may all be carried out as embodiments of virtual reality. More and more realistic simulations will whet everyone's appetite for continued improvement of the verisimilitude of these experiences. High-resolution displays that support 3-D stereo vision, coupled with specialized sensory feedback equipment will heighten the sense of realism. A lot of money is waiting to be made here. Computer game companies may lead the way to the mass marketplace, and might carve out a future for themselves by manufacturing this ancillary equipment. However, as the market expands, competition will heat up.

    Online role-playing and socialization over the Internet may become important applications of this new technology. Online banking, shopping, studying, and programmed learning might be ways in which this is used. (The online fax has become a valuable tool for the office, as was the dedicated fax machine before it.) The virtual office in which you operate your office computer from home or from a motel room and videoconference with your co-workers over the telephone may become more prevalent. Voice dictation may gradually gain a foothold, although it will have to become fast, accurate, and cheap. Again, all of this will be evolutionary rather than revolutionary. Tape recorders have been available for decades and yet we're still reading books and typing memos and reports. (The more things change, the more they remain the same.)

Four years from now (2000):
The Seven Dwarfs:
    No announcements have been made by these worthies, but CPUs running in the 3 to 5 gigops range would seem to be necessary to stay the course. Intel is making a play for the workstation business, winner take all, so what happens next should be interesting. DEC has stated that it plans to 1,000-fold its computer speeds over the next 25 years. To maintain its pace, it will have to field microprocessors running in the 2 gigops range in the year 2000, which should be do-able in this time frame.

Intel's P7
    Intel's P7 might now be the flagship of the 80X86 fleet, running at 600 MHz, while the P8 finds its way into high end workstations and desktop systems.

Intel's P8
    Intel's hypothetical P8 should pump out 1,500 to 3,000 SPECint92s and a similar number of megaflops (floating point operations per second) through some combination of parallelism and elevated clock speeds. It might incorporate on-chip digital signal processors. (We know that 2 gigops processors are feasible using a 50 MHz clock because Texas Instruments is currently selling them for $500.)

    Intel should be delivering its 10 terops processor at this time. This should provide computational speeds at the lower end of the estimated envelope for human-class mentation.

Hard Drives
    New computers should be shipping with 10 to 16 gigabyte hard drives, and not a minute too soon, as software expands to take up all the available room. Ninety gigabyte hard drives may be available for the more demanding customer.

    If the computer revolution is to continue revolving, DRAM should retail for $1 to $2 a megabyte. The 1-gigabit chip should be sampling by the end of this year. Typical systems might be shipping with 128 to 256 MB of RAM early in the year, increasing to 256 to 512 MB by the end of the year. (Of course, this assumes dramatic reductions in DRAM prices. Right now, 512 MB of DRAM would cost at least $15,000.)

Optical Disks
    CDs might weigh in at the 80 to 100 GB level (unless they tarry for a while at the 17.2 GB level). With dual HDTV tracks for 3-D vision, surround sound, interactive alternative scenarios, and perhaps, separate tracks for tactile and kinetic outputs, it shouldn't be too difficult to fill an 80 GB disk. We may all be clamoring for additional disk capacity. CDs will be driven by the commercial marketplace, while rewritable optical disks might be more responsive to hard drive capacities. At the same time, it will take a while for Digital Video Disks to penetrate the marketplace.

    Basically, we're talking about bandwidth. Bandwidth will always be relatively limited and expensive. However, by this time, we should be operating broadband, with digital voice communications. It's anybody's guess how wide telephone/cable bandwidths will be and who will be providing them. Low earth orbit satellite systems will be in operation by this time, and GPS navigational systems may be selling in upscale car and trucks. First generation automowers may be showing up in the neighborhood.

Virtual reality hardware
    2-megapixel, 17'-21" displays will probably be widespread. 1600 X 1200 resolutions are probably the entry-level displays for the 3-D stereo vision that is so important to the photorealism desired for virtual reality.

    LCD active-matrix displays might carve out a niche as desktop CRT replacements and in projection systems . They will probably still be too expensive and too small for direct-view TV. Plasma display panels, Texas Instrument's Digital Video Mirror, or some other emerging display technology such as field-emission, laser, plasma panel, or electroluminescent displays may be on trial in early HDTV sets. 2.5-4 megapixel resolution might be in the offing for higher resolution applications such as medical imaging (if it isn't already available). (Current 700, 800, and 900-line projection TV displays should be able to match or exceed HDTV requirements.) HDTV with surround sound should help popularize the home theater concept. Digital VCRs, DVDs, and recordable DVDs may bring HDTV into the home independently of broadcasts. Also, the large-screen HDTV display may be pressed into service for virtual reality entertainment, using interactive adventure or romantic movies, for networked game and role-playing, and for "window wall" videoconferencing.

    If desktop PC progress were to continue as it has in the past, a simple linear extrapolation of the past to 2010 would yield:
Year Attribute Conservative Expensive
2010 CPU Speed 150 gigops 300 gigops
2010 RAM 8 GB 32 GB
2010 Disk 60 GB* 500 GB*
2010 Accelerators 1.2 terops 9.6 terops
* - Disk storage capacity is an area of technology that is most likely to exceed expectations.

    Carrying this naive Moore's Law extrapolation through to the year 2025 would yield:
2025 CPU Speed 150 terops 300 terops
2025 RAM 2 terabytes 8 terabytes
2025 Disk 50 terabytes 200terabytes
2025 Accelerators 1.2 petops 9.6 petops
    These numbers are at the high end of the envelope for a human-caliber artificial intelligence).

    However, I have chosen to assume that progress may slow down somewhat, and to attempt to formulate a realistic set of projections that stand a good chance of coming to fruition. Engineering managers for both Intel and Motorola have stated that traditional computer technology will "hit a brick wall" in 2005 to 2010. They are suggesting that quantum-level devices, with an accompanying shift in our computer architectural paradigm will be able to overcome these limitations and permit us to continue making microprocessors smaller and faster. Here is a summary at what might be reasonable through the year 2010:

  Year  Speed, Gigops         RAM           Hard Drive    DSP's, Gigops
1990 0.002 - 0.016 1 - 4 MB 80- 200 MB NA - 0.050
1995 0.155 - 0.304 4 - 16 MB 0.5 - 2 GB NA - 1
2000 1 - 2 64 - 256 MB 4 - 20 GB 1 - 9
2005 10 - 20 512 - 2,048MB 50 - 250 GB 10 - 100
2010 100 - 200 4 - 16 GB 0.5 - 2.5 TB 100 - 1,000

    The above predictions don't depend upon technological breakthroughs but upon work which is presently underway or in the planning phase among semiconductor vendors. These prognostications appear realizable even if major improvements in semiconductor densities don't materialize beyond the critical 0.2 µ design-rule level. However, there is extensive planning underway for manufacturing down to the 0.1 µ level and there is some reason to think that semiconductor progress will continue well into the future. If so, a fanciful set of estimates for the following 15 years from 2010 through 2025 might be as follows:

  Year  Speed, Gigops        RAM           Hard Drive    DSP's, Gigops
2015 0.25 - 0.5 32 - 128 GB 2.5- 12.5 TB 250 - 2,500
2020 1 - 2 128 - 512 GB 25 - 125 TB 1,000 - 10,000
2025 3 - 12 0.256 - 1 TB 0.25 - 1 PB 3,000 - 30,000
2030 10 - 40 0.512 - 2 TB 1 - 4 PB 10,000 - 100,000

    Can there be such a thing as too much speed? What can the average beer-swizzling couch potato do with a computer that can predict the weather for the next ten years while simultaneously running all the factories on the Eastern seaboard? Write letters? Do his income tax?
    The answer is probably that, done right, real-time 3-D graphics and virtual reality rendering require enormous computational resources. Laser holographic displays might become feasible if we could muster a multi-terops processor. Model-based encoding will require massive computational resources, particularly as we progress to higher and higher- fidelity displays. Tomorrow's computer will become a near-sentient entity, and this will require enormous computational speeds. For example, various experts have estimated computational rates for the human brain that range from 10 terops to 10,000 terops. In short, we are a long way from maxing out our computers. (In 1965, we would have wondered how in the world you could use 300 7094's worth of computing power in your home!)

Appendix A: Historical Data

  • 1965: IBM 7094, $3,000,000. 32 K (110 kilobytes) of 1.1 µsec. cycle time, interleaved and overlapped 36-bit core (72-bit memory accesses from alternate banks of memory). 2.75 µsec. fl. multiply? 4.7-7 µsecond fl. divide. No disks. Only 7-track, 1-inch wide, 800-bit-per-inch tape drives. (800 6-bit characters per inch).
  • 200 KFLOPS?

    0.5 SPECint92s? 900,000 operations/second?

    1966: Univac 1108: $11,000,000. 750 nsec. cycle time. 3 processors sharing 1.1 MB of plated-wire memory. Fastrand disks and drums. 2.2 MB of "432" drums. 1.25 GB of total (slow) disk storage.

    50 MFLOPS? Perhaps 250 times the floating point speed@ 1/4500th the cost.

    155 SPECint92s. Perhaps 200- 300 times the speed@ 1/4500th the cost

    About 1,000,000 times the price/performance ratio.

    1995 Supercomputer:

    100,000 SPECint92s?

    50,000 MFLOPS(?), or about a 200,000:1 improvement in performance at the same price.