[ExI] Hard Drives, What Comes Next

Eugen Leitl eugen at leitl.org
Wed Oct 30 09:46:51 UTC 2013

On Tue, Oct 29, 2013 at 05:14:45PM -0600, Kelly Anderson wrote:

> Agreed. Though you have to compare apples to apples, and sometimes that's
> hard. For example, we can talk about the exponential growth of hard drive
> space, but that negates the value of getting at those bits faster, which is
> also important.

My secret hope is that the gap is long enough so that solid
state can catch up (SSDs have already killed 10-15 krpm SAS
by way of size, price and reliability, I don't even bother
with SAS SSDs but buy consumer SATA for production) to 
spinning rust, and put a dagger into its back. But, that is unlikely.

However, I expect that spintronics on large surfaces
(think inkjet printed on plastic or metal, and then
rolled up on a cylinder) can make spinning rust static,
putting the head into each bit. Expect good things from
the same kind of technology that brought you large,
flexible OLEDs. Serial layer deposition is an additive
technology which doesn't degrade underlying layers, so
here you have a chance to go to m^2 substrates, with
many more layers than cm^2 Si allows you -- at the price
of much larger, much slower switches -- but if these
are fully static (spintronic) that may be well worth
it, particularly since you're extremely flexible, and
prototype pipelines are so much shorter. 
> > Either you believe that there is a magic mechanism by which
> > a technology arises just in time to be passed on the torch from
> > the failing without stumbling, or you don't.
> >
> It's not magic. It is the result of work.

Yes, work that takes decades, typically. So far we've been
pretty lucky, but now the air is getting really thin.
> > So in 2027 the world will have 16 TW cumulated PV
> > capacity, that's 48 TWp (currently it has 0.1 TWp,
> > added within 20 years, this is lost in the error
> > margins).
> >
> > You've got 7 doublings yet to go. I'll spell Ray's implications
> > out, all errors are mine:
> >
> > year added capacity TWp
> >
> > 2010
> >      +0.06 TWp (actual data)
> > 2012
> >
> >
> >
> > 2013
> >      +0.19 TWp (so far 0.037 TWp added in 2013, my guess this will be
> > 0.077 TWp instead of 0.19 TWp, so -0.113 TWp gap)
> >
> So let me make sure I understand what you're saying. You're saying that Ray
> predicted that there would be 0.19 TWp installed by year end 2013, and

Ray did not say that number specifically. The number emerged simply
>>> 0.1 + 0.0625 + 0.125 + 0.5 + 1.0 + 2.0 + 4.0 + 8.0

or in terms of TWp
>>> 0.1 + 0.0625*3 + 0.125*3 + 0.5*3 + 1.0*3 + 2.0*3 + 4.0*3 + 8.0*3

The 0.1 TWp is cumulative deployment over the entire 20-40 years.
As you see, it doesn't at all figure. The 15.7 TW figure is also
not that relevant, because in the next 2 years you'd double that
(to 31.8 TW).

> there is actually 0.077 TWp installed.
> According to Wikipedia, by end of 2012, the 100 GW installed capacity

Yes, this is cumulated installed Wp capacity so far, accounted for.

> milestone was achieved. If my math is correct, that makes 0.1 TWp, and
> there is still a year between that number and the end of 2013. So where
> does the 0.077 come from?

I've found a decent graph for a number of years
specifically http://csmres.co.uk/cs.public.upd/article-images/IMS.jpg
and a newer one 
so you can fill in
2010 10 GWp
2011 28 GWp
2012 32 GWp
2013 35 GWp

and further projections show linear growth.
Assuming they're correct, they say that 0.32 TWp will be
deployed 2010-2017. This is obviously a drop in the
bucket, and a giant deviation from Ray's very bold projections.
> The thing that I remember Ray saying is that the price of solar panels per
> watt was on a curve. Are we at all on target for that?

The price is the most irrelevant part of PV. Solar inverters,
floatglass, aluminium, steel, electricians, grid upgrades,
storage capacity, synfuel production capacity are all entirely
on different curves, and you need all of these to match his
prediction. There is also the considerable energy investment
due to long energy payback times (the EROEI over lifetime is ok, 
but we're looking at 30-40 years of lifetime here -- which is
why we need to turn all the residual coal into PV and wind
> Ray gets things wrong, but so far, he has gotten more right than he has

I've once taken the time and looked at what exactly he predicted.
A lot of it is trivial (it is not particularly difficult to predict
something which already exists), a lot wrong, and the rest any
moderately well informed pundit would get right.

> gotten wrong. I never said it was a perfect science, but I know of no
> better approach for trying to predict the future. Do you?

There are few areas of interest to me, mostly life extension,
whole body/brain emulation, real AI and MNT. I don't see any
reliable way to predict it, other than I know that plotting
lines on semilog paper is a provably wrong way to do it.
You're not quite plotting what you think you're plotting,
and these lines tend to start curving down sooner than you
would have liked.
> The thing that Ray sometimes misses is human psychology issues and
> infrastructure issues. I've long said he was wrong about speech
> recognition. He isn't a god and I'm not his acolyte.
> > 2015
> >      +0.38 TWp (mind the gap?)
> > 2017
> >      +1.5 TWp  (mind the gap?)
> > 2019
> >      +3.0 TWp  (mind the gap?)
> >
> http://www.renewindians.com/2013/02/global-installed-solar-capacity-to-Reach-330GW-by-2020.html
> According to Global Data  The Global Solar PV installed capacity will leap
> frog from the current 96GW to 330GW by 2020 and will showcase a CAGR of

0.330 TWp by 2020, vs. 3.0 TWp just added between 2019 and 2021 alone.

> 16.5%.
> In another article, it states:
> “No one would have predicted even 10 years ago that we would see more than
> 100 GW of solar photovoltaic capacity in the world by 2012,” said EPIA
> President Winfried Hoffmann. “The photovoltaic industry clearly faces
> challenges but the results of 2012 show there is a strong global market for
> our technology. Even in tough economic times and despite growing regulatory
> uncertainty, we have nearly managed to repeat the record year of 2011.”
> So by your rendering of Ray's Numbers, by 2020 Ray predicts that we will
> have just under 5 TWp installed, and the solar industry is thinking more
> like 330 GW will be installed by then.
> I'd be willing to bet here on the record that we will have more than 330 GW
> installed by 2020.

You're probably correct, but it will be considerably lower than 3.0 TWp, and
the gap will only grow larger with each passing doubling period. That's because
while deployment *is* cumulative, the deployment rate is sigmoidal. It looks
like an exponential initially, where you double very little and hence can
afford it, to linear mid-term, and then saturates as you run into limits.
> > 2021
> >      +6.0 TWp  (mind the gap?)
> > 2023
> >      +12 TWp   (mind the gap?)
> > 2025
> >      +24 TWp   (mind the gap?)
> > 2027
> >
> > Do these numbers really look good to you? You'll notice we're already kinda
> > off-track, here.
> >
> They don't look especially good, no.  But I was discussing hard drive
> prices and then you changed the subject before we could talk numbers.

I don't have a good feel of problem space in areal density, so talking 
about limits there makes little sense, especially since two unrelated technologies
need to be on track to hit the limits which are still comfortably far
away. The uncertainty is sufficiently high so you can just throw dice. 

The situation is different in PV and cutting-edge semiconductor 
photolithography. So there the deviations between the exponential
model and a linear or saturating growth model are easier to see
and thus it's easier to validate or falsify your growth model.
> > How do you like your serving of crow? Roasted, boiled, smoked?
> >
> Roasted please. 2012 was a bad year for solar. They didn't sell as much as
> they did in 2011. Not making excuses, but I don't think anyone would have
> predicted that. Ray is famous for saying that economic downturns won't

I did predict that, actually. Doubling an expensive resource is not
sustainable, as very soon you run out of money and production/deployment
capacities. This is why end of clock scaling and end of financial and
Moore size scaling in semiconductors doesn't surprise me. If you have
a good model of near future limits, you can plan to route around it,
which however requires long-term planning and a lot of cash spent in
advance, which is not favorable to one's career in industry or politics.

> effect his data over the long term, but in this case I'm guessing that is
> precisely what has happened.

More information about the extropy-chat mailing list