The Future of Computing

[fishsupreme]

Bill Gates says:

Ten years out, in terms of actual hardware costs you can almost think of hardware as being free -- I'm not saying it will be absolutely free -- but in terms of the power of the servers, the power of the network will not be a limiting factor," Gates said, referring to networked computers and advances in the speed of the Internet.

The world's largest software maker is betting that advances in hardware and computing will make it possible for computers to interact with people via speech and that computers which can recognize handwriting will become as ubiquitous as Microsoft's Windows operating system, which runs on more than 90 percent of the world's personal computers.

"Many of the holy grails of computing that have been worked on over the last 30 years will be solved within this 10-year period, with speech being in every device and having a device that's like a tablet that you just carry around," Gates said at the Gartner Symposium ITxpo, held by information technology researcher Gartner Group.

Via understands it. Bill Gates actually gets it. Intel, with their new move to Pentium M and relabeling processors with model numbers instead of clock speeds, is starting to get it.

We've reached the point where "More power!" is not what most computing users want. Sure, we (the kind of geeks who read the AnandTech forums, where I found the link ot this article) still want it... but your average computer buyer does not. We've reached the end of more, faster, better -- and are now at the beginning of smaller, cheaper, and good enough.

In 10 years, super-powered enthusiast systems will be around... but their price will have gone *up* from where they are now, not down. The vast majority of people will be using systems not that much more powerful than what we use now... that they bought for under $100 and that fit in a desk drawer.

Intel's latest CPU (Pentium IV Prescott) is a disaster. It puts out enough heat to fry an egg on, and isn't much faster than what they already had. The higher they clock their CPUs, the harder it gets to ramp up more -- they're seeing diminishing returns. They can either keep throwing themselves into that wall, or recognize that most people don't want to go faster anyway; they want a computer that costs less, takes up less space, and is cheaper. They'd like it to be easier to use, too, but that's more a software than a hardware issue.

I think that future high-end CPUs will not be faster so much as more parallel -- one physical CPU will be multicored, and contain 4 or 8 or 16 logical CPUs inside it, all capable of processing simultaneously. This will be even better for software written to make use of it, but will not benefit traditional software at all... but current CPUs are already plenty fast for traditional software.

Comments

[pyran.livejournal.com]

For the record, you can't actually fry eggs on a Prescott. :-)

[eagle243.livejournal.com]

Interesting theory. I'll check back in 10 years. ;)

[prester-scott.livejournal.com]

they're seeing diminishing returns

The end of Moore's Law, perhaps?

[phanatic.livejournal.com]

Naw, just reaching the limits of currently-implemented technology. Still plenty of tricks available.

[pyran.livejournal.com]

Really, they could probably use some of those tricks already, but said tricks are pretty expensive. Cooling technologies exist to keep the chips from overheating, but they're a complicated pain in the ass.

[phanatic.livejournal.com]

All fabs are "pretty expensive."

[fishsupreme]

I think we could keep up Moore's Law for a very long time. We are nowhere near the theoretical limits of processing power. Intel and AMD could keep giving us doubling processing power every 18 months for as long as we're willing to pay for it.

And there's my key point. It's not that I think we can't make it go faster -- but it gets more and more expensive every time. Intel and IBM are practically the only companies who can afford to make a modern microprocessor fabrication plant; the cutting-edge technology is amazingly expensive. I think what we're seeing is not the end of Moore's Law but rather the end of the general public's need for it.

Without the general PC-buying public needing more power, it's going to be harder and harder to justify the expense of giving us more power.

[phanatic.livejournal.com]

It's not that I think we can't make it go faster -- but it gets more and more expensive every time.

If that's the case, the you are saying that we can't keep up Moore's Law. Moore's paper didn't just deal with increasing transistor densities, but with decreasing transistor costs. In that paper, he writes "The complexity for minimum component costs has increased at a rate of roughly a factor of two per year." His curves weren't just saying that you get more transistors, but that you get more transistors for your buck. If the only way to get more performance is to throw exponentially more money at the problem, then you're no longer following along with Moore's Law.

[fishsupreme]

True. But I think we're getting to the point where more power is like getting a bigger order of fries at McDonald's for your money... the fact that you can get twice as many fries for only $0.20 more means little if you aren't going to eat twice as many fries even if you have them.

It's not that we won't be able to keep making more powerful chips. It's that right now people spend $1500 buying 2+ GHz P4-based systems with far more power than they need because that's what's sold. I think a lot of those people are going to stop doing this when they can get a system that's powerful enough for all their needs for $300... then $200... then $100. And since those people vastly outnumber those who actually need the more powerful systems, the money for developing those more powerful systems for an ever-decreasing user pool dries up.

I also find it kind of amusing that Gates is talking about hardware becoming a negligible portion of PC cost while disregarding the equally real danger that software will go the same way. I'm a firm believer in commercial (rather than open-development) software as the best way to create a high-quality product for general public use... but the same arguments I'm making for the money for high-end hardware drying up (i.e. people won't buy premium hardware when really cheap "good enough" hardware is available) also applies to software. Microsoft Office 2010 may be the best office suite ever written, but who will pay for it if a "good enough" free alternative is available? You can bet it won't be the people buying $99 PCs.

enough power

[kmo.livejournal.com]

I think a lot of those people are going to stop doing this when they can get a system that's powerful enough for all their needs for $300... then $200... then $100. And since those people vastly outnumber those who actually need the more powerful systems, the money for developing those more powerful systems for an ever-decreasing user pool dries up.

I was under the impression that, for a while at least, a lot of PC users were motivated to upgrade their hardware when their systems could no longer run the latest crop of PC games. Is that still the case?Nowadays $300 will buy you an XBox, which seems to pack more than enough computing power under the hood to fuel just about any gamer's jones. Does that mean that the lastest, greatest PC games no longer drive gamers to shell out the big bucks for new hardware?

Re: enough power

[fishsupreme]

I'm including gamers (such as myself) in the "geeks" category, not the "regular people" category. Games still require the latest and greatest hardware. Unfortunately, as much as I like PC games, console games vastly outsell PC games.

And actually, nowadays $150 will buy you an XBox. On the other hand, PC gamers and console gamers seem to have different tastes. For instance:

Console gamers purchased:
25.1% Action games
19.5% Sports games
16.6% Racing games
7.6% Children's/educational games

PC gamers purchased:
27.4% Strategy games
15.9% Children's/educational games
11.5% First-person shooters
9.6% Family/puzzle/board games

Note that other than children's games, each platform's best-sellers aren't even in the top four for the other platform. And I think the main reason is input device... keyboard and mouse lends itself really well to some things (e.g. word/puzzle games, first-person shooters) while gamepads lend themselves really well to other things (action and sports games).

I think that the dominance of console games over PC games will continue to increase, though I don't see PC games going away altogether any time soon.

Re: enough power

[phanatic.livejournal.com]

console games vastly outsell PC games.


But PC games are what drive the technology that finds its way into console systems like the XBox.

Look at what's in the XBox: A Coppermine P3 with a piddling 128k L2 cache running at under 800 MHz; 64 megs of RAM with a 133Mhz FSB; an 8 gig, very slow drive; and a GPU that was hot beans relative to other console systems when it came out, but is now vastly antiquated.

Sure, that's cheap; you couldn't even buy a PC with such feeble specs today. Systems like that aren't what drives development.

[phanatic.livejournal.com]

Intel's latest CPU (Pentium IV Prescott) is a disaster.

Ah, but AMD's latest CPU, is a rock-solid balls-fast success.

[fishsupreme]

And expensive as all hell.

AMD wants to be a premium CPU vendor. Unfortunately for them, Intel is still the "name brand" CPU and they are not -- AMD's sales are not going to encroach too badly on Intel's market share if they don't undercut Intel on price.

As a nerd, I would buy an Athlon64 (or even the outrageously priced Athlon FX series) were I building a new computer now, but I can't see any remotely compelling reason for the average computer user to do so.

[phanatic.livejournal.com]

And expensive as all hell.

If you're the type who's going to spend $269 on a P4 Prescott 2.8Ghz, I don't think $270 for a 3200 Athlon64 is "expensive as all hell," considering price/performance.

[flummox.livejournal.com]

Intel's latest CPU (Pentium IV Prescott) is a disaster. It puts out enough heat to fry an egg on, and isn't much faster than what they already had.

I find it interesting that this same sentiment keeps getting expressed for initial product releases. The first pentium couldn't beat a 486 (and could also fry eggs). The pentium pro sucked eggs on 16 bit applications. Willamette could only just beat a Coppermine chip and couldn't compete with the Athlon.

I think the jury is still out on whether or not Prescott will be a disaster, per se. All I know is that we're selling all the ones that we're making today.

As for multi-core, intel would be mighty foolish not to develop in this direction. It's all a question of when.

[fishsupreme]

Oh, there were certainly issues like that in the past. And it's gone both ways. On one hand, there are cases like the Pentium Pro, which was slammed on its initial release, but I think turned out to actually be one of Intel's best releases -- the Pentium Pro was a perfectly usable system from the time of its release up until about when Office 2003 came out last year (I was actually still running the previous version of office on a Pentium Pro at that time). That's amazing longevity for a CPU.

On the other hand, there were cases like the 1.13GHz Pentium III Coppermine, which was clearly a case of an architecture being driven beyond its capacity. Currently, what I've been seeing in the microprocessor market seems more like that release than any other; Prescott so far looks like a CPU that's being driven beyond its architectural limits. It is new, so we won't know for sure for a while, and no doubt Intel is working on fixing the (mostly thermal and power-related) issues with Prescott; however, my speculation is that Prescott will not be remembered as one of Intel's successes. I'm not so sure about Tejas either, though obviously no information on that is generally available right now. I think the real future, though, will come from Dothan and similar cores. We've reached a point where power and thermal issues cannot be an afterthought, and I think we'll be reaching a point where they're actually more important than performance for the general PC market.

I'm not at all surprised that Prescott is selling well today. The people who think about thermal and power issues are, frankly, nerds; I'm under no delusion that I am at all representative of the general public. I think that alternatives to high-powered computers will become more and more common in the coming decade, but that this is a change that's just beginning.