Um, I doubt this, very very much.

The Air has a 50 W/h battery. That's 50W for a full hour. So if they are saying that it would last 2.5 seconds, that means the putative 1990s computer would have to consume 60 x 24 x 50 = 72,000 Watts!

Someone at the Atlantic needs to figure out basic math.

I read it to mean they were factoring in the computing power as well.

i.e. they estimated that it would take 72,000 Watts-worth of 1990s computers to match the processing power of a Macbook Air.

Though that does add further caveats to an already less-than-ideal comparison (as an MBA is not going to last seven hours at 100% CPU load).

Sounds like they ought to mention how long that is in football fields.

If you placed all the iPhones in the world end to end on a line, some of them would be in the ocean.

There, I wrote about iPhones. Now give me my ad revenue.

How many iPhones in the world? If I guessed about 100 million that would be about 8000 miles, which wouldn't necessarily be in the ocean if you started the line in Europe and cut across Russia/Eurasia.

I'd chuck mine in the ocean just to make it true.

Or even better, compare it to a butterfly flapping its wings (http://en.wikipedia.org/wiki/Butterfly_effect) as this kind of comparison is sensitive to so many different factors you can't take any conclusions too seriously.

Still, not a bad yardstick to see how far we've come with respect to computing power per watt - maybe this is why the rate of change is increasing more than exponentially - you have multiple exponential rates converging in a positive feedback system...

>maybe this is why the rate of change is increasing more than exponentially

Doesn't seem so. Classic Moore : 7h (25200s) / 2.5s ~= 2^14 that places us in - 1.5 years for a power of 2 - 1990.

>more than exponentially - you have multiple exponential rates converging in a positive feedback system...

not really. The system as whole improves with the same exponential rate precisely because all and each of its components improves with the same exponential rate. Exponential rate of CPU and exponential rate of RAM produce the same one exponential rate of the improvement of the whole system comprised of the CPU and the RAM. Exponential rate of the CPU only, for example, would result in much less than the exponential rate for the whole system.

I wonder how many library congresses a modern iPad holds.

Or in Libraries of Congress.

Yes, it must be. And from 1990 to 2011 number of transistors per cm2 has doubled 14 or so times.

If they are going to compare it with computational power, they should also factor in battery development since 1990.

What is missing is how many 603ev chips would be needed to approximate the Sandy Bridge Core i5's performance. And of course, how many feet thick that would make the 2.5-second-battery 1997 MacBook Air.

Did you read the article? "Imagine you've got a shiny computer that is identical to a Macbook Air, except that it has the energy efficiency of a machine from 20 years ago."

It would take an awful lot of 1990's machines to equal the computing power of a macbook air.

I'm being pedantic here -- units are Serious Business -- but "W/h" is "Watts per hour". You want "Watt hours", which you could write "W-h".

Or as Wh. Like we write Nm for Newtonmeter.

If you read the article, it's clear they mean an hypothetical 1990s computer with the computing power of the modern macbook air, and the power efficiency typical of computers of that era.

Completely nonsensical, I'll grant you.

I would guess that "computer that is identical to a Macbook Air" is key here: it means the current Air's power. That would probably be a mainframe in those days?

I'm not sure that such a thing existed.

It's not a fair comparison. A 90s mainframe would probably smoke an Air if they were competing on IOPS with a thousand connected users!

I wouldn't be too sure. The low-end 13" MacBook Air has a 1.7Ghz dual-core i5 processor, 4GB RAM, a 128GB SSD, and a Thunderbolt port providing two bidirectional 10Gb/s channels.

From what I can find, an IBM S/390 from the early 1990s could have up to 6 processors and 6 vector coprocessors, up to 9GB of internal storage, and up to 256 fiber-optic links running at 10MB/s (for an total bandwidth of a Thunderbolt port plus a USB 2.0 port).

I don't really know how fast the CPU and RAM were in the IBM mainframe, but I'd suspect the Air's clock speeds are high enough to make up for only having two processors, and the Air's SSD would make it much faster for data sets that don't fit in RAM.

And as for the reliability advantage a mainframe is supposed to have: you could buy several MBAs per month to act as hot-spares for the price of renting the mainframe, and the MBA has a UPS built-in.

It's easy to forget that 1Gbit/second an insane amount of text / numberic data and 1990 was a LONG time ago. Just concider that the extimated monthly transfter acroos internet backbones in December 1990 was 1TB. That 1GBit lan cable was 100's of times faster than all of the internet backbones in 1990.

So while, Mainframe software is vary efficent the hardware still sucked compared to modern systems. Just think a 1GBit/second Fible Channel did not show up until 1997 http://en.wikipedia.org/wiki/Fibre_Channel.

PS: The #5 super computer in June 1993 had 4 cores Processor NEC 400 MHz (6.4 GFlops). http://www.top500.org/system/377 A 999$ mackbook air uses a 1.6GHz dual-core Intel Core i5 processor that would crush it in large part due to that 3 MB L3 cache but also due to being able to do far more in of those cycles.

An interesting aspect is how much more slowely memory sizes have caught up. Today's laptops can crush a 90s supercomputer in GFlops, but are just now catching up to a mid-1990s Cray in main memory (4GB). I remember being fairly befuddled when I started college in 2000, and my CS dept's UltraSPARC servers were slower than my Pentium III in CPU, but had memory measured in a unit that I had only heard of being used for hard drive space.

A fair comparison though would have to be a modern Cray-class supercomputery thing to that Cray though. That's going to have more than 4GB of RAM; modern Cray equivalents open the bid at petabytes of RAM.

Oh, I agree modern supercomputers beat 1995-era ones on all axes; it's mostly the different rates at which consumer-class computers catch up that are interesting. A modern MacBook Pro wipes the floor, CPU-wise, with a 1995 Cray, but only just matches it RAM-wise. So comparisons like "today's [consumer thing] is as good as [year's] [enterprise thing]" depend heavily on whether you're comparing them on CPU or RAM.

There are few, if any, current systems with petabytes of RAM; the very largest have tens to hundreds of terabytes.

I think rather that a 1990s computer would take the full battery (or 180 kJ) to accomplish the same tasks that the macbook air could in 2.5 seconds of operation (~50 J?). So if you changed the task efficiency of the Macbook air to be that of an old computer, but kept the same processing power, the Macbook Air would be consuming 72,000 Watts.

I think that's the reason for the distinction between "a 1990 computer" and a "computer with 1990 energy efficiency." It's explained quite clearly in the article itself.

A counterpoint: the Air still does not match the battery life of a TRS-80 Model 100 from 1983: 20 hours on 4 AA batteries.

A Model 100 was pretty much the opposite of the iPad: great for creative activities (writing, taking notes, programming) but poor for consumption activities.

If you strictly limit the creative activities there are to ones that involve typing. I'll bet the iPad is way better at making music, drawing, photo and film editing, etc, etc.

Writing is important – but not everything.

exactly. I've posted this before, but: "if a person can't produce something on an ipad that's a reflection of the user, not the device."

Can I produce a program for iPads on an iPad?

do web apps for iPad count? (see textastic)

so if someone could make a 68030 as efficient as an i5, I could have a PowerBook 170 that would run for a week on a charge? Yes, please.

( I'll need someone to knock up a Twitter reading client in MPW too, pls)

Heck, why not go even further back?

http://atomsandelectrons.com/blog/2010/04/apple-t/

Modern MCUs are relatively power efficient, and certainly are available in 68030s performance region. But their architecture might not be very suitable for general purpose computing.

This is just another way of stating Moore's law. Chip transistor count doubles every year and a half; chip power use doesn't.

Oh Macbook Air. For some reason my tired brain kept thinking about what an Air Battery would look like.

Something like this: http://www.youtube.com/watch?v=f2a3uTU0fgc

Wow. It exists! Thanks.

If only battery technology had kept up with computing technology.

We'd definitely all be driving electric cars today if that happened.

The best (commercial) batteries today are only a few times better than the ones used in electric cars 100 years ago.

I've read the rate of improvement for batteries is around 8% a year. That's still exponential, just not as good as transistor count. But nothing else improves that fast.

Makes sense to me. If they calculated that in 1990 it took x watts to generate y processing power by now x has gone wayyyy down if y remains the same. They don't have the best documentation but it doesn't seem unreasonable.

Here is an older, portable Macintosh with more (claimed) battery life than most of the modern Macs.

http://www.youtube.com/watch?v=gGwVTq_xcZk