The Sliding Lower Bound

thoughtful emoticon

What is the least powerful computing device you can imagine?

An abacus probably comes to mind, an arithmometer, maybe even a math grenade… Well, ok, that can be pretty low. Let’s put it differently: What is the least powerful computing device you can purchase? The answer is somewhat surprising if you’ve never thought about it in detail.

The lowest powered digital computers that come to my mind that actually merit the title of “digital computer” and which are currently still in production are Atmel tinyAVR series1, the weakest of which is a pretty involved RISC architecture CPU with 512 bytes flash program memory, 64 bytes rewritable EEPROM, 30 bytes RAM, and a 12MHz clock speed. Which beats the pants off the first stored program Turing complete electronic digital computer, whichever of the contenders for the title you pick.2

The fact, that you can replace a device that once took the better part of a building with an integrated circuit you need to glue to something else so you don’t lose the thing, evokes thoughts of progress and advancement of civilization, but there’s one nagging thing. The current market price for an Attiny4 is something like $0.38 – which is a bit much if you want to put one, say, into disposable packaging or a birthday card. Surely you could go cheaper at the expense of the computing power?

And therein lies the problem: No, not really. It’s not impossible to make weaker computing devices, but it’s not profitable to produce them, it’s not profitable to retail them, and in general the laws of economics say that there’s a minimum cost to design and produce a chip, and it puts a lower bound on the possible cost of the chip itself. With that lowest possible cost, competition ensures highest possible computing power will be squeezed into it. For certain reasons, atomic components of such devices do exist and are still being produced, but they are actually more expensive, rather than less.3

You might not think it is a problem, but it sort of is, and the first time I encountered it was many years ago, when USB flash drives first came into prominence, while floppy disks were dying off. In a situation where you would use a floppy disk, hand it to someone else and forget it, you could use a flash drive, right? Only, the cheapest ones were a bit too expensive for such practice, especially considering how few files they actually need to store. While drives with higher storage capacities were progressively more expensive, there simply weren’t any drives with lower storage capacities available anymore. Why? Because it’s not profitable to produce and retail them.

There were multiple attempts to make them cheaper still, from Lebedev’s idea to Intellipaper, to GIGS.2.GO and somehow you still can’t buy them in retail – while they obviously can be made, and the use case still exists,4 it looks like they aren’t profitable enough to bother.

While the storage space in flash memory is mostly determined by actual silicon surface area, and is therefore more linearly reflected in costs than computing power, the trend is still there:

Economics prevents electronic devices from being exactly as simple as they need to be. This applies even to the simplest things, and with devices that are explicitly general purpose computers it’s even more obvious.

This has several noteworthy consequences.

  • Pretty much anything is eventually going to be a hackable general purpose computer, down to the very simplest objects that do nothing other than blink a led.5
  • Almost all of the effort spent on producing more computing devices actually ends up spent making machines that do little more than consume energy to produce heat.6
  • There is basically no reason whatsoever to reuse old hardware, since its power requirements typically make it more costly than using a more modern computer to do the job of multiple older computers.7
  • There is very little incentive to go out of your way to write efficient software, since general purpose computers it will run on exceed its needs by orders of magnitude.

I’m not sure if any of these are entirely good or bad things, they’re just fact. But they are certainly something that needs to be kept in mind more often.

  1. You can probably find something weaker, but it’s irrelevant for the point I’m making here. ↩︎

  2. But if it doesn’t, just take the next Attiny, which is only a few cents more expensive. ↩︎

  3. In fact the time when it will be cheaper to use a microcontroller than the once ubiquitous 555 timer is not far off. ↩︎

  4. When dealing with a government bureaucracy, this is often the only practical method. ↩︎

  5. Case in point: Since a while ago, Arduino boards no longer include an FTDI232 chip for handling USB connectivity. USB is either a feature of the main IC itself or, as in the case with the Uno, handled by a separate general purpose microcontroller. Why? No difference in price, but more flexibility. ↩︎

  6. For your desktop machine which stands there doing nothing useful, there’s SETI@home and similar projects. But up until the Internet of Things gets off the ground, there will be no hope to put the smaller devices to work. Though come to think of it, a global neural network style AI running on the spare computing cycles of everyone’s smartphones could be just what Kurzweil ordered↩︎

  7. And if you don’t feel like rewriting software, you just wait one more year and then run it in virtual machines. ↩︎