On the future of computing hardware

04d August 7, 2016 -- (asphalt tech)

One, systems tend to get smaller.

Mathematicians with an inclination for studying so-called "systems" may without much effort look at various examples of such "systems" and devise quantitative measures of complexity. For example the mammal's organism is highly complex on various levels of abstraction, say, when looking at its average genetic makeup; the same can be said about social networks1, global weather2 or the motion of stars.

These complexities are however against the intuition we have in engineering. That is because human-made systems are not only designed to be functional, but to also be as deterministic as possible; and such determinism cannot be achieved without a proper understanding of the theoretical model used in designs and especially its limitations. "Simplicity is the ultimate sophistication", and "everything should be made as simple as possible, but no simpler", but also "simplicity is prerequisite for reliability", to quote only a few aphorisms.

We can observe empirically that systems' complexity, or rather some measure of their "size", evolves cyclically. Dinosaurs evolved to be huge creatures, only to be replaced in time by miniaturized versions of themselves3. The first simple computing systems had the size of a room, while large-scale integration has led to computers which fit on the tip of one's finger. This reduction in size comes with reduction in certain features' size and/or complexity, which is why for example humans don't have tails, nor the amount of body hair of some of their ancestors, nor sharp canines.

Looking at the evolution of various industries in the last few decades4, it is clear that they are currently in the latter phase of the growth-reduction cycle. It is only natural that the so-called hardware, and more specifically numerical computers made out of silicon5 will follow.

Two, on shaky foundations; transcending ideological confusions.

The fact that nowadays' hardware industry stems from the needs of marketing, as opposed to the needs of the market, is well-known. This charade started with personal computers, then it was followed by mobile phones, tablets, and now so-called "smart devices" from "the Internet of things", which operate in cycles of what is known in economics as "planned obsolescence"6.

Purely from the perspective of computing hardware, this leads to the proliferation of functionality that is useless, or worse, harmful to the average consumer. For example the new generation of Intel Skylake processors come with Intel SGX, Intel MPX and possibly others; all this at the expense of reliability7, and at the same time without advancing the state of the art in any fundamental sense. All they do is offer some clients new ways to shoot themselves in the foot, and Intel are by no means a singular case8.

Degrading processor quality aside, there are quite a few pragmatic factors lying behind the existence of the planned obsolescence-aggressive marketing vicious circle. For one, it's easier to scale silicon production up than to scale it down, as it's more effective to fabricate a large wafer containing smaller and smaller processors than to produce in smaller numbers9. Meanwhile, some of the steps in the manufacturing cycles (still) require significant human workforce (for the time being), which makes eastern Asia the prime choice for production10. Thus the per-unit cost of producing a processor is higher for smaller batches (say, a hundred at most), while e.g. smartphone assembly in high numbers will necessarily rely on poorsters in China.

This context however underlies a problem of a more ideological nature, that has been recently, yet unfruitfully, discussed in the free software world: there is a general lack of choice in general-purpose hardware architectures, and given the situation above, it's not likely that this will improve. We are most likely heading into a duopoly between ARM and x86, while the fate of MIPS is not quite certain11, IBM's Power is somewhat expensive and FPGAs likewise, barring the very low-entry ones. The open RISC-V architecture offers some hope, given Google's interest in them, but I wouldn't get my hopes up just yet.

This is not the first time I am writing about this. I have touched on the subject on the old blog a while ago, and I have also discussed some of the (still current) issues in a previous Tar Pit post. People don't seem to have gotten their heads out of the sand since then, so I will reiterate.

Three, I want to build my own hardware.

To be honest, I don't care too much about Intel, Qualcomm, Apple and their interests. If Romanians were building their own computers12 back in the days before "personal computer" was a thing, then I don't see why I couldn't do this in the 2010s. Whether it's made feasible by 3D printing, some different technology or maybe some hybrid approach13, this is a high-priority goal for the development of a sane post-industrial world and in order to pick up the useful remains of the decaying Western civilization. However one would put it, small-scale hardware production is the next evolutionary step in the existence of numerical computers.

It is readily observable that14 the computer industry is heading towards a mono-culture, not only in hardware, but also in operating systems15 and in "systems" in general. This will -- not might, not probably, but definitely -- have the effect of turning the "systems" world into a world of (often false) beliefs, much akin to Asimov's Church of Science, where people will not even conceive the possibility of existence of other "systems".

I am of course not crazy enough to attempt to stop this. The industry can burn to the ground as far as I'm concerned, and this it will, and it'll be of their own making. What I want is to gather the means to survive through this coming post-industrial wasteland.

  1. By which I mean specifically not the jokes the average Westerner calls "social networks". Facebook, Twitter, Reddit et al. are only networks in the sense that they reduce the level of interaction to at best that of monkeys throwing typewriters around; at best. The average level is rather Pavlovian in nature.

  2. Weather, not climate. Mkay?

  3. Which taste like chicken.

  4. Popescu's "The end of democracy" is required reading on this matter.

  5. Unless some other better, faster, stronger material takes its place. Which I kind of doubt, given that science is also in a period of stagnation, and that is putting it very mildly.

  6. Which, although probably perceived by the naïve as capitalist, is rather reminiscent of the old communist planned economy model. And not unlike communist economy, it often leads to higher prices and lower quality products. To bear in mind next time you're buying that new Samsung or Apple smartphone.

  7. Dan Luu's "We saw some really bad Intel CPU bugs in 2015, and we should expect to see more in the future" is required reading on this particular matter.

  8. ARM are somewhat more conservative, but they offer SoC producers enough freedom to shoot their clients in the foot. Without any doubt, the average Qualcomm phone is most probably running Secure World software that the end user will never care about, and that the curious mind will never have the chance to reverse engineer -- without considerable financial resources, anyway. That's a good thing, you say? Well, it's your opinion, you're entitled to it, please stick it up your ass.

    But what am I saying? By all means, please do buy whatever shiny shit Apple or Samsung are selling you. As long as it's not my money...

  9. There are quite a few technical reasons for this too, and some of these escape me. For one, opening a semiconductor plant isn't exactly cheap, and the ratio of defects to units produced is significant, yet in principle easy to estimate statistically. The cost of making a circular wafer is also not small, which makes economic feasibility a tricky thing. Meanwhile, bear in mind that Moore's law is on its way to becoming dead and buried, given that CMOS-based technologies are reaching their physical limits.

  10. To be perfectly clear, the world's largest silicon producer is at the time of writing not Intel, but Taiwan Semiconductor Manufacturing Company, Ltd.

  11. Imagination Technologies, the intellectual property holder, have not been faring too well in 2016.

  12. Remember the story of the ICE Felix HC that was the toy in my early computing days, before I had the slightest idea of what an algorithm actually was. Now ponder the fact that there is no qualitative difference between that piece of junk and today's latest, greatest, whateverest computer. Yes, you can do the exact same things on that old heap of junk, and by "exact" I certainly do not mean watching porn, by which I mean that this is why your children will prefer make-believe sex instead of fucking real women, which uncoincidentally is why Arabs are the superior ethnicity and those God-awful political corectnesses will stop being a thing in less than a generation. But I digress.

    As a funny historical footnote, the same Romanians attempted at making a Lisp machine back in the '80s, which makes me hopeful that the same thing should be achievable almost three to four decades later. I hate repeating myself, but it's quite literally either this or the dark ages.

  13. The approach itself is relevant only as far as it solves the most problematic economical aspects, i.e. logistics (needs to be made using readily available and/or easily procurable materials) and low production costs of a small number of units (tens to a few hundreds at most). Processing speed and size are secondary aspects, a Z80 (or maybe something equivalent to a 80386) should really be enough for most general-purpose-ey uses.

  14. Much to the baffling ignorance of otherwise intelligent people. Unfortunately for them, nature abhors singularities; as one of the older and wiser men in the Romanian Computer Science community used to remind us, there is, simply put, a cost for abso-fucking-lutely everything in life.

  15. In case you were wondering, despite being the most usable kernel to date, Linux is definitely the ultimate abomination. Its greatest feature is that it's not too difficult to strip of all the crap.

    If you don't know what I'm talking about, take a look at CVE-2016-0728 for example.