post #1 of 1
5/24/09 at 5:41pm
Thread Starter 
SF writer Charles Stross gave the keynote speech at the LOGIN 2009 conference.

http://www.antipope.org/charlie/blog...in_t.html#more

Really interesting piece.

I also think he's basically right. We'll all get to be geezer gamers!

Exceprt (piece is longer than this, this is just the beginning):
Quote:
I've just given one of the keynote speeches at the LOGIN 2009 conference here in Seattle. Here's more or less that I said ... Imagine you're sitting among a well-fed audience of MMO developers and gaming startup managers (no, nobody video'd the talk):

Good morning. I'm Charlie Stross; I write science fiction, and for some reason people think that this means I can predict the future. If only I could: the English national lottery had a record roll-over last week, and if I could predict the future I guess I'd have flown here on my new bizjet rather than economy on Air France.

So that's just a gentle reminder to take what I'm going to say with a pinch of salt.

For the past few years I've been trying to write science fiction about the near future, and in particular about the future of information technology. I've got a degree in computer science from 1990, which makes me a bit like an aerospace engineer from the class of '37, but I'm not going to let that stop me.

The near future is a particularly dangerous time to write about, if you're an SF writer: if you get it wrong, people will mock you mercilessly when you get there. Prophecy is a lot easier when you're dealing with spans of time long enough that you'll be comfortably dead before people start saying "hey, wait a minute ..."

So: what do we know about the next thirty years?

Quite a lot, as it turns out — at least, in terms of the future of gaming. Matters like the outcome of next year's superbowl, or the upcoming election in Germany, are opaque: they're highly sensitive to a slew of inputs that we can't easily quantify. But gaming is highly dependent on three things: technological progress, social change, and you.

Let's look at the near-future of the building blocks of computing hardware first.

On a purely technological level, we've got a pretty clear road-map of the next five years. You know all about road maps; the development cycle of a new MMO is something like 5 years, and it may spend another half decade as a cash cow thereafter, The next five years is a nice comfortable time scale to look at, so I'm going to mostly ignore it.

In the next five years we can expect semiconductor development to proceed much as it has in the previous five years: there's at least one more generation of miniaturization to go in chip fabrication, and that's going to feed our expectations of diminishing power consumption and increasing performance for a few years. There may well be signs of a next-generation console war. And so on. This isn't news.

One factor that's going to come into play is the increasing cost of semiconductor fab lines. As the resolution of a lithography process gets finer, the cost of setting up a fab line increases — and it's not a linear relationship. A 22nm line is going to cost a lot more than a 33nm line, or a 45nm one. It's the dark shadow of Moore's Law: the cost per transistor on a chip may be falling exponentially, but the fabs that spit them out are growing pricier by a similar ratio.

Something like this happened, historically, in the development of the aerospace industry. Over the past thirty years, we've grown used to thinking of the civil aerospace industry as a mature and predictable field, dominated by two huge multinationals and protected by prohibitive costs of entry. But it wasn't always so.

Back in the nineteen-teens, it cost very little to get in on the game and start building aircraft; when a timber magnate called Bill went plane-crazy he and one of his buddies took a furniture shop, bought a couple of off-the-shelf engines, and built some birds to sell to the US navy. But today, it takes the company he founded close to a decade and ten billion dollars to roll out an incremental improvement to an existing product — to go from the Boeing 747-100 to the 747-400.

It turns out that the power-to-weight ratio of a modern high-bypass turbofan engine is vastly higher than that of an early four-stroke piston engine, modern construction materials are an order of magnitude stronger, and we're just a hell of a lot better at aerodynamics and design and knowing how to put the components together to make a working airliner.

However, the civil airliner business hit a odd brick wall in the late 1960s. The barrier was a combination of increasing costs due to mushrooming complexity, and the fact that aerodynamic drag goes up nonlinearly once you try to go supersonic. Concorde and the Tupolev 144 — both supersonic airliners — turned out to be dead ends, uneconomical and too expensive to turn into mass consumer vehicles. And today, our airliners are actually slower than they were thirty years ago.

In the medium term (by which I mean 5-15 years) we're going to reach the end of the exponential curve of increasing processing power that Gordon Moore noticed back in the late 1960s. Atoms are on the order of one nanometre in size; it's hard to see how we can miniaturize our integrated circuits below the 10nm scale. And at that point, there's going to be a big shake-up in the semiconductor business. In particular, Intel, AMD and the usual players won't be able to compete on the basis of increasing circuit density any more; just as the megahertz wars ended around 2005 due to heat dissipation, the megaflop wars will end some time between 2015 and 2020 due to the limits of miniaturization.

There's still going to be room for progress in other directions. It's possible to stack circuits vertically by depositing more layers on each die; but this brings in new challenges — heat dissipation and interconnection between layers, if nothing worse. There's room for linear scaling here, but not for the exponential improvements we've come to expect. Stacking a hundred layered chips atop each other isn't going to buy us the kind of improvement we got between the 8080 and the i7 core — not even close.

This is going to force some interesting economies of scale. Over the past couple of decades we've seen an initially wide-open playing field for processors diminish as bit players were squeezed out: we had SPARC and PA-RISC and IBM's Power architecture and SGIs MIPS and ARM and the 68000 series and, and, and. But today we're nearly down to two architectures in the consumer space: Intel on the PCs and Macs — which are basically just a PC with a different user interface, these days — and ARM on handhelds. Actually, ARM is about 95% of everything, consumer and embedded both — as long as you remember that the vast majority of consumer-owned computers are phones or embedded gizmos. The other architectures hang on in niches in the server and embedded space but get no love or attention outside them.

I expect to see a similar trend towards convergence of GPUs, too. It's expensive to develop them and graphics processors aren't made of sparkly unicorn turds; it's semiconductors all the way down, and constrained by the same as other components — memory, cpu, whatever. So I expect we'll see a market in the next decade where we're down to a couple of processor architectures and a handful of GPU families — and everything is extremely boring. New components will be either the result of heroic efforts towards optimization, or built-in obsolescence, or both.

I don't want to predict what we end up with in 2020 in terms of raw processing power; I'm chicken, and besides, I'm not a semiconductor designer. But while I'd be surprised if we didn't get an order of magnitude more performance out of our CPUs between now and then — maybe two — and an order of magnitude lower power consumption — I don't expect to see the performance improvements of the 1990s or early 2000s ever again. The steep part of the sigmoid growth curve is already behind us.

Now that I've depressed you, let's look away from the hardware for a minute.
Reply