Slashdot Mirror
Less Is Moore
Hugh Pickens writes "For years, the computer industry has made steady progress by following Moore's law, derived from an observation made in 1965 by Gordon Moore that the amount of computing power available at a particular price doubles every 18 months. The Economist reports however that in the midst of a recession, many companies would now prefer that computers get cheaper rather than more powerful, or by applying the flip side of Moore's law, do the same for less. A good example of this is virtualisation: using software to divide up a single server computer so that it can do the work of several, and is cheaper to run. Another example of 'good enough' computing is supplying 'software as a service,' via the Web, as done by Salesforce.com, NetSuite and Google, sacrificing the bells and whistles that are offered by conventional software that hardly anyone uses anyway. Even Microsoft is jumping on the bandwagon: the next version of Windows is intended to do the same as the last version, Vista, but to run faster and use fewer resources. If so, it will be the first version of Windows that makes computers run faster than the previous version. That could be bad news for computer-makers, since users will be less inclined to upgrade — only proving that Moore's law has not been repealed, but that more people are taking the dividend it provides in cash, rather than processor cycles."
Less: 120884 bytes
More: 27752 bytes
Wow, that's right!
Summation 2
Let's be honest here. What does the average office PC run? A word processor, a spreadsheet, an SAP frontend, maybe a few more tools. And then we're basically done. This isn't really rocket science for a contemporary computer, it's neither heavy on the CPU nor on the GPU. Once the computer is faster than the human, i.e. as soon as the human doesn't have to wait for the computer to respond to his input, when the input is "instantly" processed and the user does not get to see a "please wait, processing" indicator (be it a hourglass or whatever), "fast enough" is achived.
And once you get there, you don't want faster machines. More power would essentially go to waste. We have achived this moment about 4-5 years ago. Actually, we're already one computer generation past "fast enough" for most office applications.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Some of you may remember the 1980s and early 1990s, where PCs started out costing $5,000 and declined slowly to around $2,500 for name brand models.
Around 1995, CPUs exceeded the GUI requirements of all the apps then popular (this is pre-modern gaming, of course). Around 1996 and into 1997 the prices of PCs fell off a cliff, down to $1,000.
Those who fail to remember history...
This could simply be down to the tanking economy: people look at what they're spending, and quickly realise that:
1) the upgrade treadmill over the last twenty years has produced insanely powerful and dirt-cheap hardware. When was the last time you had trouble running Linux on your hardware? I'm old enough to remember!
2) and that you don't need teraflops of CPU/GPU power just to draw greasepaper-style borders around your Microsoft Word windows. Perhaps the entire industry has woken up and seen how unbelievably wasteful modern computing is, and have decided to take the dividend of Moore's Law in cash instead.
3) recessions are good for purging wasteful and suboptimal behaviour generally.
Maybe people will realize what an obscene waste of money and computing power and operating system like Windows Vista, which requires a gig of RAM to run, really is.
Moore's law does not say "that the amount of computing power available at a particular price doubles every 18 months." Moore's law says that the number of transistors that can be placed inexpensively on an integrated circuit increase exponentially, doubling approximately every two years.
Um, no it wasn't. "Moore's law" is a term that was coined after Thomas Moore gave a presentation showing that the company was managing to double transistor density each month. This observation created an interesting problem for the company. What should they do with all those extra transistors?
One option was that they could keep getting higher yields on existing chips, eventually driving the cost per unit to mere fractions of a penny. The other option was that Intel could do something useful with all that extra circuitry and maintain higher prices.
Considering that contemporary CPUs of the time were barely more powerful than the interrupt controller sitting next to them, using that silicon for sophisticated 32bit processors with on-die floating point units and SIMD instructions seemed like a no-brainer for the company. Thus as each successive generation of technology has made CPUs smaller, Intel has used the extra space to add more features and more optimizations.
At this point, things are getting a bit ridiculous. CPU manufacturers have so much extra space on which to work that they can fit 2-4 CPU cores on a single die and STILL produce a smaller chip than the last generation.
Javascript + Nintendo DSi = DSiCade
Even Microsoft is jumping on the bandwagon: the next version of Windows is intended to do the same as the last version, Vista, but to run faster and use fewer resources. If so, it will be the first version of Windows that makes computers run faster than the previous version.
Without Vista, MS wouldn't be able to claim that 7 was faster than their previous version of Windows.
And now I'm going to do something shocking and unprecedented. I'm going to look-up the actual quote, instead of guessing what Moore's "Law" means.
"April 1965:
"The complexity for minimum component costs has increased at a rate of roughly a factor of two per year ..." Notice he claimed *complexity* not power doubled, and that it happened EVERY year. His original statement has not held true.
"I disapprove of what you say, but I will defend to the death your right to say it." - historian Evelyn Beatrice Hall
Yes, and as so many have pointed out, their history of doing so is now backfiring on them in a big way. And it's not just with Vista, it's with Office as well.
Case in point - several months ago my department bought upgrade licenses to Office 2008. I was perfectly happy with Office 2004, but I installed Office 2008 because I knew that if I didn't, I wouldn't be able to read whatever new formats that Office 2008 supported. It had happened with every other Office upgrade cycle in my experience - you either upgraded or you'd be unable to exchange documents with your peers.
But something funny happened this time - I have yet to receive a .docx, .xlsx, or .pptx file from anyone. I have quite consciously chosen to save every document in .doc, .xls, or .ppt "compatibility" format. Everybody I talk to says they're doing exactly the same thing. Everyone now knows the game that Microsoft plays, and no one is willing to play it anymore. I could have stayed with Office 2004 and never noticed the difference. So what motivation will I have to upgrade to the next version of Office?
If it weren't for Microsoft's OEM licensing deals, Vista would have a tiny fraction of its current market share. XP is "good enough". But Microsoft doesn't push Office onto new machines the way it does Windows, the older Office formats are also "good enough", and you have open source alternatives like OpenOffice if Microsoft tries to deliberately break Office compatibility on the next version. I fully expect Microsoft's Office revenues to take a steep dive in the next few years. The Vista debacle is only the beginning.
One of the things I learned many years ago, is that computer and computing speed isn't a function of how fast something runs. Rather it is a matter of whether or not you actually run something.
If computer speeds are twice as fast, and it currently takes you ten seconds to accomplish Task A, and a new computer will allow you to accomplish that same task in 5 seconds .... getting a new computer is not that big of a deal.
However, if you run Task B, which takes 1.5 hours to complete, and a new computer will run that same task in say 4 minutes (Real world example from my past, log processing), the difference isn't necessarily the 86 minute difference, but rather if and how often you actually run that task.
It is endlessly amusing to see "real world benchmarks" that run in 3 minutes for most processors, separated by less than 2 x. Or frames per sec. Or ...... whatever.
When shit takes too long to get done, you tend NOT to do it. If the difference is a few seconds, that is nice and all, and a few seconds may be of interest to "extreme" hobbyists.
But Real World differences are not marginally decreasing from 300 to 279 seconds. Sorry, but those extra few seconds aren't going to prevent you from running that Task.
The true measure is not how fast something gets done, but whether or not you actually do the task, because the time involve is prohibitive.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
While I can see the desire for cheaper rather than more powerful, I do wonder how much of the power/price tradeoff curve actually makes sense. Traditionally, the very high end of the curve makes very limited sense, since it is the nightmare world of low yields, early adopter taxes, and super critical enterprise stuff. In the middle, the power/price curve tends to be roughly linear; before gradually becoming less favorable at the bottom, because of fixed costs.
As long as a processor, say, has to be tested, packaged, marked, shipped, etc.(which costs very similar amounts,whether the die in question is a cheap cutdown model or a high end monster) there is going to be a point below which cutting performance doesn't actually cut price by any useful amount. Something like the hard drive is the same way. Any drive has a sealed case, controller board, motor, voice coil unit, and at least one platter. Below whatever the capacity is of that basic drive, there are no real cost savings to be had(incidentally, that is one of the interesting things about flash storage. HDDs might be 10 cents a gig in larger capacities; but that doesn't mean that you can get a 4gig drive for 40 cents, I had a quick look, and you can't get anything new for under about $35. With flash, you might be paying 100 cents a gig; but you pretty much can get any multiple you want).
Cost, overall, is gradually being whittled down; but, once all the low hanging super high margin products are picked off, there is going to be a point past which it simply isn't possible to exchange price for performance at any reasonable rate. Used and obsolete gear offers a partial solution(since it can be, and is, sold below the cost of production in many cases) but that only works if your needs are small enough to be fulfilled from the used market.
In recent years not only has CPU performance been increased, but the efficiency in terms of power consumption per unit of work has greatly improved.
Even if the majority of users begin realize they have no practical use for top end CPUs with gobs processing power, everyone still benefits from higher efficiency CPUs. It reduces electric bills, simplifies cooling systems, allows for smaller form factors, etc. I think in the future the power efficiency will become more important as people start to care less about having the ultimate killer machine in terms of processing power. People are already performing actions on their mobile devices(iPhone, Blackberry, etc) which were possible only on a desktop in past years. The strict power requirements of these devices with tiny batteries will continue to demand improvements in CPU technology.
I'm waiting for the day when it is common to see completely passively cooled desktop computers, with solid state hard disks, no moving parts, sipping just a few watts of power without emitting a single sound.
The article you pointed out is pure nonsense. It claims that bloat isn't important due to the fact that memory cost dropped. Not only that, it tries to base that claim on this idiotic metric of dollar per megabyte and how the fact that software like microsoft's excel bloat from a 15MB install in the 5.0 days to a 146MB install in the 2000 days is somehow a good thing because in the 5.0 days it took "$36 worth of hard drive space" while "Excel 2000 takes up about $1.03 in hard drive space". No need to justify a 100% footprint. We are saving money by installing more crap to do the exact same thing.
In fact, the idiot that wrote that article even had the audacity to state:
Up is down, left is right, bloat is actually good for you.
But people still complain. Although it appears that we should be grateful for all that bloat, we are somehow being ungrateful by believing that all that bloat is unnecessary. But fear not, the idiot that wrote the article has a nice accusation for all those bloat haters out there:
Yes, that's it. We don't hate orders of magnitude increase in bloat simply to be able to perform exactly what has been easily done with a fraction of resources. We don't hate to be forced to spend money on hardware to be left with a less than adequate solution when compared with the previous generation. We simply hate windows. Good call.
The article is bad and you should feel bad for posting a link to it.
Slashdot, fix your code or at least hire someone who is competent at it to do it for you.
This is dangerous territory for Microsoft to be in. Levelling off of computer power means that buyers are getting off the upgrade treadmill -- they're not buying new computers every couple of years. Preloads on new computers are where Microsoft makes the bulk of their Windows sales.
To make matters worse, without constant upgrades, Microsoft and ISV's can't count on new API's becoming widespread anytime soon, so they have to write applications for the lowest common denominator. This prevents Microsoft from forcing its latest agenda onto everyone -- and even worse, it could potentially provide the WINE team enough time to reach feature parity with Windows XP. (Spare me the lecture, have you tried WINE lately? It's surprisingly good these days.)
All in all, Microsoft is being forced to stand still in a place where they can't afford to. Commoditization sucks when you're a monopolist, doesn't it?
Tired of FB/Google censorship? Visit UNCENSORED!
Well, actually it's just proof that history repeats itself. Because this thing has happened before. More than once.
See, in the beginning, computers were big things served by holy priests in the inner sanctum, and a large company had maybe one or two. And they kept getting more and more powerful and sophisticated.
But then it branched. At some point someone figured that instead of making the next computer which can do a whole megaflop, they can do a minicomputer. And there turned out to be a market for that. There were plenty of people who preferred a _cheap_ small computer, than doubling the power of their old mainframe.
You know how Unix got started on a computer with 4k RAM, which actually was intended to be just a co-processor for a bigger computer? Yeah, that's that kind of thing at work. Soon everyone wanted such a cheap computer with a "toy" OS (compared to the sophisticated OSs on mainframes) instead of big and powerful iron. You could have several of those for the price of a big powerful computer.
Then the same thing happened with the micro. There were plenty of people (e.g., DEC) who laughed at the underpowered toy PCs, and assured everyone that they'll never replace the mini. Where is DEC now? Right. Turned out that a hell of a lot of people had more need of several cheap PCs ("cheap" back then meaning "only 3 to 5 thousands dollars") instead of an uber-expensive and much more powerful mini (costing tens to hundreds of thousands.)
Heck, in a sense even multitasking appeared as sorta vaguely the same phenomenon. Instead of more and more power dedicated to one task, people wanted just a "slice" of that computer for several tasks.
Heck, when IBM struck it big in the computer market, waay back in the 50's, how did they do it? By selling cheaper computers than Remington Rand. A lot of people had more use for a "cheap" and seriously underpowered wardrobe-sized computer than for a state of the art machine costing millions.
Heck, we've even seen this split before, as portable computers split into normal laptops and PDAs. At one point it became possible to make a smaller and seriously less powerful PDA, but which is just powerful enough to do certain jobs almost as well as a laptop does. And now it seems to me that the laptop line has split again, giving birth to the likes of the Eee.
So really it's nothing new. It's what happens when a kind of machine gets powerful enough to warrant a split between group A who needs the next generation that's 2x as powerful, and group B which says, "wtf, it's powerful enough for what I need. Can I get it at half price in the next generation?" Is it any surprise that it would happen again, this time to the PC? Thought so.
A polar bear is a cartesian bear after a coordinate transform.
My own opinion is that it sets the limit for where they can stop and relax their efforts, internally.
They can never stop and relax. They're chasing an exponential growth curve.
Given the amount of secrecy in this industry, I'm not certain how you can back this statement up with any fact. My own assumption is that 'they' have developed technology far more capable than what they currently claim to be working on at any given time. I personally believe that what they claim is on the drawing board is actually in prototype, what they claim to be in dev is actually ready for production, and their 'latest and greatest' is already old tech.
I've worked at several processor companies with top-of-the-line fab tech, including Mr. Moore's. While my NDAs probably mean I can't tell you anything specific with regard to scheduling, I can tell you without fear of revealing any secrets that you're way off base. They are not sand-bagging with more advanced tech waiting in the wings.
The only sense in which you are correct is that the 'latest and greatest' thing you can buy is old tech relative to things then under development. That's because there's typically a year give or take (usually give) between receiving the first silicon from the fab in the transistor node the product was designed for and all the validation, bug fixes, and spins on the product before it's ready to be sold. That means the fab tech has to be done and mostly stable by the time you start this process, so go roughly six months back before that where they're making test chips in the new fab to make sure it's working. And development of that fab tech before it's ready to run its first test chip wafer is two or more years before that, with R&D going on for years before that.
So yeah, when you could buy a 65nm CPU in the store, there may have been a 45nm CPU or just a test chip coming out of a fab somewhere, and a 32nm lithography machine being developed somewhere else, and a lab somewhere working out how 22nm lithography would work. But that's not 'sandbagging' because all of those things were years of serious non-stop development away from becoming products! Keeping on the exponential growth curve means that there has to be a constant pipeline of developments, and this pipeline is quite long.
And believe me, if they could increase the rate at which those future techs become available for making product, they would. "Sandbagging" means wasting competitive advantage, and wasting money. The machinery in the fabs for each node cost billions of dollars, and they depreciate rapidly. If they had some new tech working flawlessly, but weren't using it in products and just waiting in the wings, they'd be flushing hundreds of millions down the toilet. Time to market is one of the most important things they look at.
Honestly, if you look at actual press releases and actual product launches, it's much more likely that what they claim is a prototype is really on the drawing board, and what they claim to be ready for production is really still in development -- see the AMD Barcelona for the most recent example. You think they had the Phenom II just waiting in the wings while they got beat up in the press and the market over the launch of Phenom?
Now this isn't to say that they wouldn't sandbag if it were possible, and to some minor extent they have. When K7 had a big leg up over P3 in frequency headroom, or Core 2 vs aging K8s, sure they held back a little to get more margin on a cheaper part. But we're talking a speed grade delayed by a month or two. Barely noticeable noise on the curve. Actually tracking that curve requires non-stop expenditures and execution of R&D, and any significant slip-up could send a company flat on its face. To slow development on purpose? Ridiculous.
In my mind, this is the only way to sustain this curve - by limiting the release of new technology onto the market until Moore's says that it is time for it.
Think about it in terms of l
The enemies of Democracy are