My desktop, that I pieced together in 2011, has more computing power (including the GPU) than the fastest supercomputer in 1999 (~3 TFLOPS for GPU+CPU perf).
If you have some cash burning a hole in your pocket, you can piece together a ~10 TFLOP machine today entirely from NewEgg without doing anything too fancy.
Are you sure about that? The TOP500 list uses the linpack benchmark to measure double-precision FLOPS. The numbers you cite for your machine sound more like those you would get with single-precision.
For comparision, the GTX Titan can do about 1.5 TFLOPS in double-precision and about 3.2 TFLOPS in single-precision doing GEMM.
That reminds me of the joke about the A7 in the iPhone being faster than the old PowerPC cores in the G5s, which wasn't the first time that a phone's processor beat an older laptop's, but the first time that the comparison was Apple's to Apple's. Terrible pun, but it happened in what, a 5 year span?
I play the same game, and I love it (no one ever believes me about it, either).
Ah, no, you're correct! I had originally misremembered the statement (going with G4 and laptop), and forgot to change the word laptop to desktop when I switched it to the G5. That was totally my bad.
That's much less clear if someone isn't aware of how negative exponents are defined. That's why I chose this notation, and why I put into words "cut in half 12 times."
> That's much less clear if someone isn't aware of how negative exponents are defined.
Yes, I've noticed that in other contexts. It's too bad because it makes the representation less clear, more complex than necessary. I first became aware of this when I tried to say that gravitational force declines as r^-2, to numerous protests.
Moore's law is that the density of discrete components (transistors, storage cells) increases. Cost and speed are knock-on effects. It just so happens that density does drive CPUs to be faster for the same production cost and storage to be either more capacious or physically smaller. But some aspects of computing performance don't benefit from density, like memory response time for example.
"But the fact that so many were so impressed by an assertion that an iPhone possesses the capabilities of $3,000 worth of 1991 electronics products – when the actual figure exceeds $3 million – reveals how fundamentally difficult it is to think in exponential terms."
Is nonsense. The point of a general-purpose computer is that it's general purpose. But people aren't buying an iPhone to get a general-purpose computer, they're buying one to get something that does things. The comparison to 1991 gadgets is in terms of thing-doing, not MIPS and storage and whatnot. The comparison of raw compute power is interesting, but it doesn't invalidate the other kind of comparison.
The comparison to 1991 gadgets is in terms of thing-doing, not MIPS and storage and whatnot. The comparison of raw compute power is interesting, but it doesn't invalidate the other kind of comparison.
That was my reaction as well. In 1800 you only needed 1 or two horses to pull a cart to get you from point A to point B. A typical car today might have 200 horsepower, but you don't actually compare that to having 200 actual horses in 1800.
It works both ways though, a 200hp car can get you across the county a lot faster than 200 horses. Just as an iPhone can do a single threaded task faster than 3 billion dollars in 1991 hardware.
But a 200hp car won't get you there 200 times faster than a horse. It'll get you there anywhere from 2 to 20+ times faster depending on the distance and the speed (city vs highway), but the comparison is pretty meaningless.
(in other news, Google seems to have definitions for various animals' speeds. Neat!)
It's a fundamentally different question. The first article asked "how much would it cost to buy devices in 1991 to provide similar functionality?" while this one asks "how much would it have cost to buy the same amount of raw capability?"
This would have been an interesting article except it loses credibility by pretending the two questions are the same, when they are not.
> while this one asks "how much would it have cost to buy the same amount of raw capability?"
Even that isn't really what it asks. If you wanted the raw storage capacity of an iPhone in 1991, you wouldn't buy 32GB of flash memory. You'd buy hard disks. 32GB of flash in 1991 would have been slow and poorly suited to the job.
Speaking of hard drives, a GB of storage was not $10K in 1991. It ranged from $4.25K to 2.79K during that year. http://www.jcmit.com/diskprice.htm You'd have to go back to 1988 to hit the $10K/GB price.
And of course, it depends on what you're storing. The Radio Shack ad has several things that would use storage -- a VHS camcorder, a microcassette recorder, a portable CD player. All of them used different storage mediums at different prices. If you want to price out the storage for the "1991 iPhone equivalent" the way the original analogy asks, "minutes of audio" and "minutes of video" is the unit, not gigabytes. The article gets away from those terms by posing a different question and pretending it's the same question.
This is ridiculous. You can't count modern clock speeds in 1991 dollars.
Also, "massively underestimates the true pace of innovation?" Seriously? It's already too fast! People have such hard-ons for change and shiny things now that they don't even remember why they wanted it in the first place!
I'm sure if you looked at the raw material cost the iPhone would have been a lot cheaper, right? I mean, if you took all of the knowledge on how to build the factories and design the tools and fabrication equipment, and you had people skilled enough to understand it and implement it, the iPhones in 1991 would probably cost a lot less.
What are you saying exactly? Apple doesn't take raw materials as input and produces iPhones as input. Apple takes components as input. Those components didn't exist.
If someone built a few of them using today's knowledge, it would be expensive. Very expensive.
Only at scale are they cheap. They are only at scale if everyone is buying them. The same process that would cause the scale in components might cause the change in commodity prices.
My point is simply that the USD is now worth less and the raw materials are now more scarce and expensive than in 1991. You are right in calling me out about the scale though, as I guess that even if you had all the knowledge to create all the components necessary, you still would have a much smaller market and therefore less offsetting of the initial cost. However, if say, you made 2 phones for everybody, maybe that would bring the market up enough to reach the same scale as today's market?
If you had all the knowledge to create the components, you would still need a decade or more of industrial advancement to reach the point where you could actually build the necessary components. Knowing how to build every piece of an iPhone wouldn't get you an iPhone in 1991, even if you had all the money in the world. You couldn't build the fundamental technologies fast enough.
Industrial advancement IS the knowledge of creating components. Hordes of people at giant companies are improving every minor aspect of the tools that create these components. If we had all of the knowledge these companies have built up at any point in history, it could be implemented. The tools are the HIGHEST cost in creating components because they cannot be changed dramatically, and it's difficult to put them through iterations of improvement because you need to keep using the expensive tool you made to make up the cost of what you paid for it.
Industrial advancement isn't just building on knowledge. It's building on infrastructure. You couldn't go back to the stone age and produce steel, because the infrastructure for creating steel simply didn't exist. There were no iron mines, no foundries, no skilled blacksmiths. You could speed up the process, but you'd still be dead before the first steel sword was forged.
Likewise, the infrastructure for building iPhone components didn't exist in 1991. The A7 is built on a 28nm process. The 486 in 1991 was shipping with (I believe) a 0.8 micrometer process. So the infrastructure of the time was three orders of magnitude away from being able to build this chip. And this says nothing of the other components which faced similar technological hurdles. You couldn't make all these hurdles go away. Building up major infrastructure takes time, and no amount of knowledge can reduce that time to zero.
As shown in this story by the simple example of a pencil, the ability to produce products of the type we have now at the expense we have now might as well require it to be 2014, not 1991.
Exactly. The word you're looking for is "undefined". Even with four million dollars, it's impossible. As other people have mentioned, it might have been possible to build something that worked in a similar fashion, but it most certainly wouldn't have had even a similar form factor.
It's a product of the technology available at its time.
Software was many orders of magnitude more efficient in 1991 than now, so if you're trying to make a functional equivalent (which I'm assuming because the example equivalences are 1000x the size of a phone) rather than a copy, it's important to remember that you wouldn't need either the same processing power or same storage capacity.
An equivalent excluding media, rather. A few copies of peanut butter jelly time would fill up a functionally equivalent iPhone's disk.
Interesting article, but guess what the point is, coming from the AEI. It's actually about gutting social security. If you take their view of incredible deflation in tech products, it cancels out the inflation in everything else. That means the inflation adjustment in social security is less, and in the long run, social security becomes insignificant. This type of messaging is true mastery of the long game.
Their inflationary calculations are very tendentious and ignore substitution effects. 32GB of flash memory in 1991 would have cost $1.44 million: Sure. So you wouldn't have used flash memory, you'd have used something cheaper, like audio tapes and film, and you would have had an inferior experience for it... but not $1.44 million worse of an experience. These figures are meaningless. That's why the CPI uses an actual market basket that's meant to represent what people actually buy. People whose payments are linked to the CPI have not in fact seen their payments deflated by such astonishing factors.
The BLS (which is responsible for the official CPI) has a recent overview on possibilities of error and related considerations: http://www.bls.gov/opub/btn/volume-1/consumer-price-index-da... Their conclusion: "Measurement of price change in a large economy is sufficiently complex that the accuracy of an estimate is difficult to gauge and is likely to be debated. The CPI cannot claim to be a completely precise measure of inflation and publishes the variance of its estimates. Several potential sources of bias have been identified in the CPI and addressed, though there continues to be debate over to what extent and in what direction bias may still exist and the ways in which BLS can continue to increase accuracy."
Sure, members of the public or small and medium sized organizations may have substituted, but if yours was the sort of organization that bought, say, a Cray Y-MP for $35 million, paying $3.56 million and putting this thing on a network would have been a bargain, which makes me think that the authors may not have taken the value of this being an integrated system into account. And even if they pegged the iPhone at exactly the same price as a Cray Y-MP they would still be low due to the iPhone's vastly lower power consumption.
Finding out that this article was written by the American Enterprise Institute made me feel dirty until I reflected on the fact that nothing in the article appears to be grossly incorrect, and discarding an interesting article for some possibly bad political implications seems like a bad way to live one's life.
Another thing that no-one has yet pointed out is that even this article's calculation could still probably be extremely low if the cost of developing iOS and the related apps are included.
Putting aside for the moment that this article doesn't even mention Social Security:
How much of a decrease to baseline CPI would constitute a "gutting"? It's not as if the current formula is engraved on stone tablets from the gods: Like any estimate, it could probably be improved over time, up and down.
Anyway, if AEI has a motive in writing pieces like this, I think it's more to emphasize how incredible technological advancement has improved our buying power and consequently our lives. I run into a surprising number of folks on forums like this who are convinced that they'd much rather live in the golden "Mad Men" years of the 1960's, when you (a man) could raise a (typically white) family on one blue collar income, which came with a house, car, a wife with time to raise the kids, and a golden retirement in Key West. It's easy to lose track of just how much everyday "magic" you'd be giving up, even if you're not some poor minority who'd be really screwed trying to earn a living back then.
iPhones dont fill your belly or provide heat in the winter.
Inflation isn't the same in every sector, so inflation adjustments should also be targeted to the problem they are trying to offset. This is why what you measure for inflation is so important. To address the differences in what different population segments purchase, the BLS has come out with different consumer-price-indexes that weigh different categories of goods differently: http://www.bls.gov/news.release/cpi.br12396.a06.htm
Electronics such as the cell phone have become integral parts of consumer's lives. Fact is you are at a severe disadvantage in modern society without them. Are they food and shelter necessary? No. But as a society we are hoping to move beyond just barely subsisting. Our poor (in America) are EXTREMELY well off compared to the bottom 10% of the world. There is a difference between absolute and relative poverty. Without food, water, and shelter you are absolutely poor. Without a cell phone and computer you are relatively poor (by American standards).
I challenge anyone to imagine what a disadvantaged life they would have without cheap, ubiquitous computing ability and connectivity.
Agreeing on an objective definition of "poor", if only for sake of a given discussion, would help immensely. Use of different terms for objectively different conditions would help too (i.e.: not using the same word for "below subsistence" and "still better off than >80% of world population").
This is such an odd calculation and application of Moore's law.
Without going into any analyses or details, travelling to the moon should be much cheaper and faster (yes I am aware of low-cost orbital space flights -- and even paying it with bitcoin! --, but it still isn't travelling to the moon).
48 comments
[ 3.8 ms ] story [ 120 ms ] threadWhat's the latest year that my backpack be a Top 500 Supercomputing center.
The results can be pretty entertaining sometimes.
If you have some cash burning a hole in your pocket, you can piece together a ~10 TFLOP machine today entirely from NewEgg without doing anything too fancy.
Technology is funny.
For comparision, the GTX Titan can do about 1.5 TFLOPS in double-precision and about 3.2 TFLOPS in single-precision doing GEMM.
I play the same game, and I love it (no one ever believes me about it, either).
3,560,000 * 2 ^ -12 ≈ 869.14
Based on: 1/2 ^ 12 = 2 ^ -12
Yes, I've noticed that in other contexts. It's too bad because it makes the representation less clear, more complex than necessary. I first became aware of this when I tried to say that gravitational force declines as r^-2, to numerous protests.
Oh, well -- a tempest in a teapot, as they say.
The effect of Moore's "law" is that processing power doubles approximately every 2 years, while the cost halves.
If 1 "unit" of processing currently costs $1; in 2 years time, 2 units will cost $1. Or that 1 unit costs 0.50.
"But the fact that so many were so impressed by an assertion that an iPhone possesses the capabilities of $3,000 worth of 1991 electronics products – when the actual figure exceeds $3 million – reveals how fundamentally difficult it is to think in exponential terms."
Is nonsense. The point of a general-purpose computer is that it's general purpose. But people aren't buying an iPhone to get a general-purpose computer, they're buying one to get something that does things. The comparison to 1991 gadgets is in terms of thing-doing, not MIPS and storage and whatnot. The comparison of raw compute power is interesting, but it doesn't invalidate the other kind of comparison.
That was my reaction as well. In 1800 you only needed 1 or two horses to pull a cart to get you from point A to point B. A typical car today might have 200 horsepower, but you don't actually compare that to having 200 actual horses in 1800.
(in other news, Google seems to have definitions for various animals' speeds. Neat!)
This would have been an interesting article except it loses credibility by pretending the two questions are the same, when they are not.
Even that isn't really what it asks. If you wanted the raw storage capacity of an iPhone in 1991, you wouldn't buy 32GB of flash memory. You'd buy hard disks. 32GB of flash in 1991 would have been slow and poorly suited to the job.
Speaking of hard drives, a GB of storage was not $10K in 1991. It ranged from $4.25K to 2.79K during that year. http://www.jcmit.com/diskprice.htm You'd have to go back to 1988 to hit the $10K/GB price.
Also, "massively underestimates the true pace of innovation?" Seriously? It's already too fast! People have such hard-ons for change and shiny things now that they don't even remember why they wanted it in the first place!
If someone built a few of them using today's knowledge, it would be expensive. Very expensive.
Only at scale are they cheap. They are only at scale if everyone is buying them. The same process that would cause the scale in components might cause the change in commodity prices.
So your point is hard to understand...
Likewise, the infrastructure for building iPhone components didn't exist in 1991. The A7 is built on a 28nm process. The 486 in 1991 was shipping with (I believe) a 0.8 micrometer process. So the infrastructure of the time was three orders of magnitude away from being able to build this chip. And this says nothing of the other components which faced similar technological hurdles. You couldn't make all these hurdles go away. Building up major infrastructure takes time, and no amount of knowledge can reduce that time to zero.
As shown in this story by the simple example of a pencil, the ability to produce products of the type we have now at the expense we have now might as well require it to be 2014, not 1991.
It's a product of the technology available at its time.
An equivalent excluding media, rather. A few copies of peanut butter jelly time would fill up a functionally equivalent iPhone's disk.
edit: MIDI Peanut Butter Jelly Time?
The BLS (which is responsible for the official CPI) has a recent overview on possibilities of error and related considerations: http://www.bls.gov/opub/btn/volume-1/consumer-price-index-da... Their conclusion: "Measurement of price change in a large economy is sufficiently complex that the accuracy of an estimate is difficult to gauge and is likely to be debated. The CPI cannot claim to be a completely precise measure of inflation and publishes the variance of its estimates. Several potential sources of bias have been identified in the CPI and addressed, though there continues to be debate over to what extent and in what direction bias may still exist and the ways in which BLS can continue to increase accuracy."
Finding out that this article was written by the American Enterprise Institute made me feel dirty until I reflected on the fact that nothing in the article appears to be grossly incorrect, and discarding an interesting article for some possibly bad political implications seems like a bad way to live one's life.
Another thing that no-one has yet pointed out is that even this article's calculation could still probably be extremely low if the cost of developing iOS and the related apps are included.
How much of a decrease to baseline CPI would constitute a "gutting"? It's not as if the current formula is engraved on stone tablets from the gods: Like any estimate, it could probably be improved over time, up and down.
Anyway, if AEI has a motive in writing pieces like this, I think it's more to emphasize how incredible technological advancement has improved our buying power and consequently our lives. I run into a surprising number of folks on forums like this who are convinced that they'd much rather live in the golden "Mad Men" years of the 1960's, when you (a man) could raise a (typically white) family on one blue collar income, which came with a house, car, a wife with time to raise the kids, and a golden retirement in Key West. It's easy to lose track of just how much everyday "magic" you'd be giving up, even if you're not some poor minority who'd be really screwed trying to earn a living back then.
Inflation isn't the same in every sector, so inflation adjustments should also be targeted to the problem they are trying to offset. This is why what you measure for inflation is so important. To address the differences in what different population segments purchase, the BLS has come out with different consumer-price-indexes that weigh different categories of goods differently: http://www.bls.gov/news.release/cpi.br12396.a06.htm
I challenge anyone to imagine what a disadvantaged life they would have without cheap, ubiquitous computing ability and connectivity.
Without going into any analyses or details, travelling to the moon should be much cheaper and faster (yes I am aware of low-cost orbital space flights -- and even paying it with bitcoin! --, but it still isn't travelling to the moon).
So $3 million is understating. They couldn't have built anything close to today's smartphone even with $3 billion with B.