Intel revises the future

By Stephen Ballantyne

Friday 5th March 2004

The fine tuning of its 64-bit strategy to better cope with immediate challenges from competitors is probably no more than a short term adjustment for Intel.

Because Intel is one of the giants of the IT industry, it's easy to suppose that like the giants of tradition it is a slow, lumbering beast more concerned with whose bones it's going to pick next than with the longer view.

Actually, this doesn't seem to be the case. Intel is in the basic hardware business and it has to deliver in ways that other companies don't. Not long ago Intel appeared to be one leg of the mighty Microsoft-Intel-Compaq triumvirate; Intel made the chips, Compaq put them in attractive boxes and Microsoft provided the software to make them go.

Amazingly, that old order has taken quite a battering, and deservedly so. Compaq is now part of Hewlett-Packard, where its marketing department can continue to make promises just like its competitors; at least until Windows XP came along, Microsoft's products (and those of the rest of the software industry) frequently disappointed users by promising more than they could deliver. Only Intel has had to make stuff that really does what it says.

Most customers accept ergonomically unsatisfactory computers or underperforming software as part of the territory but they won't cut Intel that much slack. The time a few years ago when Intel shipped Pentium chips that under some rare circumstances would get their arithmetic wrong in about the sixth decimal place showed how demanding Intel's market is.

Therefore the company really can't afford to relax. It has to think not only about what is needed now but also what will be demanded next year, and the year after, and as far into the future as possible. If Intel doesn't, its competitors will that is the commercial engine that drives Moore's Law, still quoted as holy writ in the company's keynote sessions.

Providing the longer view is the job of Intel's chief technical officer, Pat Gelsinger. An egregiously brainy sort of guy, Mr Gelsinger is now, as he was at last year's IDF, obviously the ideas man for Intel.

This year at Intel Developers Forum in San Francisco he concentrated on the implications of what we already know was happening processors are becoming faster while storage is becoming cheaper. Much faster and much cheaper.

Up to now, speculation about the implications of this growth have tended toward slightly nutty fantasy. If Moore's law continues to operate for another 10 years, and Mr Gelsinger appears confident it will, by the end of this decade computer chips with processing power similar to that of small mammal brains should be commonplace.

"What's beyond the era that we have now and what will be the applications, the computing models and the architectures that we are going to invent collectively to take us to that next era?" Mr Gelsinger asked in his keynote speech, before going on to offer some suggestions. His alternative uses for all that processing power and inexpensive storage have the virtue of building on the techniques we already understand.

According to Mr Gelsinger, the entire internet contains about 530 petabytes (see sidebar). That's a lot of data but with storage currently retailing at about $1659 a terabyte, it's still less than a billion dollars worth of storage space, and it'll be a lot less next year.

Nobody seems to come up with a rule of thumb to describe how fast storage prices are falling but the drop is pretty rapid. So rapid that Mr Gelsinger speculates about the possibility of individuals storing exabytes of personal data and using very rapid processors to mine that information. He reckons a detailed audio record of every conversation an individual might conduct in a lifetime would amount to about three exabytes, which presumably you'll be able to carry around with you in a portable device.

Recording the soundtrack of an individual life is all very well but it isn't much use unless you can extract useful information from it. Mr Gelsinger sees major opportunities for software development in recognition, mining and synthesis (RMS) applications. These applications typically involve relatively simple routines that have to be repeatedly applied as quickly as possible, a task that should mop up the processing power Moore's Law predicts will be available almost as fast as it can be invented.

This is what Mr Gelsinger calls the "era of tera" and he thinks we should be well into it by the end of this decade. And despite the lack of a clear programming model for thought, he says we should expect some major changes in the way we interact with computers, particularly thanks to improved speech recognition technology and the use of phased microphone arrays able to intelligently distinguish between speakers and recognise what they are saying.

"Over the next 10 years," he says, "we're going to cross over some very important thresholds that will make computers very human-like in their ability to interact."

The precise details of what the era of tera will be like depend on what software developers are able to come up with and that's not Intel's job. But Mr Gelsinger is reasonably confident RMS is the field most likely to produce results.

"We're not sure what will be the 'killer app' of tomorrow Google for picture content? Conversation retrieval? Reality simulation indistinguishable from life? but we're engaging with the developers who are working on such products, the better to anticipate their needs with our hardware."

BIG, BIGGER, INCOMPREHENSIBLY HUGE

Keeping count of the numbers involved in storing large amounts of data is a game in itself:

* Bit: The smallest unit of information, with a value representing either 1 or 0, true or false

* Byte: 23 (eight) bits

* Kilobyte: 1024 bytes, or 210 bytes in scientific notation (very approximately 103 bytes)

* Megabyte: 220 bytes, from the Greek for large (about a million, or 106, bytes)

* Gigabyte: 230 bytes, from the Greek for giant (about a billion, or 109, bytes)

* Terabyte: 240 bytes, from the Greek for monster (about 1012 bytes)

* Petabyte: 250 bytes (about 1015 bytes)

* Exabyte: 260 bytes (about 1018 bytes)

* Zettabyte: 270 bytes (about 1021 bytes)

* Yottabyte: 280 bytes (about 1024 bytes)

Comments from our readers

No comments yet