The Inexorable March of Technology

The original formulation of Moore's Law stated that the density of integrated circuits (the number of transistors on a chip of a given cost) doubled every eighteen months. The computer industry has used this as a benchmark for how its research should progress; the result is that, with every 18 months that pass, the performance you can acquire for a given price doubles.

This is an intensely unintuitive concept - almost as bad as compound interest, but operating over a much shorter time-span - so since the middle of 1994 I've been regularly noting just how much various components cost. I've accumulated a megabyte or so of Excel-format data; what follows is probably only a tiny fraction of the conclusions you can draw from it, but to present the whole thing would be unutterably bulky.

Microprocessors

Three graphs, showing how many MHz you can get for £100 and £200 (mail-order, excluding VAT and delivery), as well as the fastest speed-grades announced, at quarterly intervals for the last several years.

Hard disc space

I have become somewhat notorious in some areas for the quote "disc space is cheap". These two graphs (price per gigabyte for best-value disc, size of disc available for £100) should give some backing to my claim. Note that the price-per-gigabyte graph is plotted on a logarithmic scale because otherwise the more-than-a-factor-two drop from £3.51/G to £1.57/G between October 2000 and October 2001 is an imperceptible change in the gradient of a line almost indistinguishable from the X axis.

I admit that the price of the cheapest hard disc available has not dropped anything like as far: in August 2000 £63.50 bought you 15GB, in September 2002 £38 buys you 10GB. I believe this is because the mechanical parts of a disc drive have a certain unavoidable level of complexity, whilst increasing capacity has often been a matter of improving the coatings on the disc platters and the design of the read head, both of which happen on the microscopic scale and cost a vast amount in research but then rather little in unit cost. And, until discs as large as anybody could reasonably want were routinely available, there wasn't much incentive to reduce cost and size together.

Whole systems

Let's see what the same sort of sum would get you in seven consecutive years:

You'll notice that the components mentioned above don't amount to a whole system; however, additional parts like cases, power supplies, floppy drives, network cards and monitors have been at rather more constant prices over those years. The speed figure is the geometric mean of the integer and FP speeds taken from Spec95 benchmark submissions for machines for which Spec95 has been run, or Spec2000 submissions scaled so that the 315/223 score for a P3/700 matches its 33.8/24.3 Spec95 score, rescaled for P100 = 1.0

Over six years, there has been an improvement of nearly a factor thirty in clock speed, and of between thirty and forty times in actual performance since the Pentium 4 can do more in one clock-tick than the original Pentium; compared to its 1995 counterpart, the 2002 system has 64 times as much memory, 120 times as much disc space, and staggeringly better graphics, and moreover costs £94 less (and more like £300 less in inflation-adjusted pounds).

But this hasn't meant that a user can now compose thirty essays, design thirty graphical layouts or write thirty programs in the time once taken for one. New capabilities came in over the first four or so years: the 1995 machine couldn't play DVDs, could barely play MP3 audio files (had MP3 been around then - it became popular in 1998), would take hours to convert a music CD to MP3 format, and would be painful to use for editing photos. The 1999 machine could do any of those tasks without difficulty.

But there are very few useful tasks for which the capabilities of the 2002 and 1999 machines differ. The 2002 machine can convert a whole CD of music to MP3 in two minutes whilst the 1999 one might take fifteen; it can play a DVD using 15% of its processor rather than 90%, but generally users aren't doing anything else on the computer while watching a DVD. Admittedly, it can compress a DVD of video in ninety minutes rather than eight hours - but a two-hour task would usually be left overnight in any case. A 120G disc merely accumulates more cruft than a 20G one because there's no longer a perceived need to delete anything, however irrelevant or redundant; few programs take advantage of 512MB of memory. The only differences come when playing games - the 2002 machine will play Doom 3 respectably when it's released, the 1999 one will not.

Recently, the life-span of computers has not been much over five years: in October 2002, my department replaced its last machines of the 1996 specification, and the older machines are starting to become actively rare. You can't give away a 486 computer, most have already been thrown away, and I have a 486DX/66 processor and heat sink sitting on the side of my desk as a memento mori. It's possible that this will change as people realise the limited capability differences between new and older machines, but I fear "computers last five years" has already become part of business planning.

Remember, more computing power was thrown away last week than existed in the world in 1984.