OLPC upgrades

[Mitch Bradley]

Guess what? The people at OLPC, who aren't stupid, already considered 
every point in the message cited below, a long time ago. So why aren't 
we doing them? ...* *On Wed, Jan 28, 2009 at 9:57 PM, Carlos Nazareno 
<object404 at gmail.com>wrote:

> > where entire servers would melt down because of the sheer number of
> > geeks a link to the front page of slashdot.org would bring.
> >
> > foru simple words can save the entirety of OLPC and the AMD, and in
> > fact make both filthy rich, rich enough to execute our insane kumbaya
> > vision.
> >
> > military grade hacker toy
>   

I've been in the computer industry for over 25 years and I have seen 
very, very few hacker-oriented products succeed at any significant scale.

The problem with selling to hackers is that they all want something 
different, and they want shiny and new, so you can't get enough volume 
of one product to get to economy of scale.

And with a hacker product, you *never* get a large single order or a 
predictable continuing business stream, so planning future production is 
virtually impossible.  See below about the importance of that.

> >
> > say those four magic words, sell the XO via geek online stores, and
> > 1CC will be so slashdotted to high heck with orders that the waiting
> > list will take years to fill out. 
... which is pretty much the classic way to go out of business.  If you 
can't fulfill orders in a timely fashion, the customers will get mad and 
cancel the orders and badmouth you to death.  To fill orders in a timely 
fashion, you have to build up stock in advance, which means that you 
have to buy parts in advance (lead times can be months), which means you 
either have to pay in advance or convince the suppliers to bet on you, 
which is very hard because the suppliers that are still in business have 
learned the hard way that extending credit to unproven companies is a 
very good way to stop being in business.

If you guess high and order too much, you can lose money even if you 
have sold a lot of product.  If you guess low, you get into the unfilled 
orders / unhappy customers problem, so you have to do something heroic 
to get the production up quickly, which usually means you end up paying 
more for parts and services.

In many cases, there can be a 9 month delay between when you have to 
commit the money for the parts and the time you actually get paid.  
Multiply that times a lot of units, and you are looking at a lot of 
money at risk for a very uncertain future return.

> At similar price points, the XO-1
> > puts the Nintendo DS, Tapwave Zodiac, GP2X, Sony PSP, Chumby and iPods
> > + iPhones to shame.
>   
Huh?  Nintendo and PSP are game machines backed by long lists of titles, 
with well-established companies that have put lots of effort into 
creating products that appeal to a specific market, and a long history 
of many other companies that got it wrong and died.  iPods are, or were, 
a fashion statement.  You're not likely to see anybody jogging with an 
XO dangling from their neck or strapped to their arm.  iPhone is the 
current fashion statement, and it is associated with a valuable service 
called a telephone network.  You can't whip an XO out of your pocket and 
call Billy Bob. 

Comparing XO to these products is meaningless, except insofar as the 
iPhone / PDA products are starting to become capable enough to subsume a 
lot of computer functions.  The XO-1 isn't a credible substitute for a 
phone at all.
> >
> > Support? real men don't need no steeenkin manuals or directions! just
> > ship em in a plain cardboard box with a power adaptor and all will be
> > good to go!
>   

You might be surprised at the amount of time I have spent answering 
questions from hackers.  In the sort of open development and 
communications environment that we have, hackers know how to contact 
you.  In some cases, the interaction with them has been quite valuable, 
but the "don't need no support" notion is just not what happens.

> >
> > Start selling current batches of 1GB Nand 256MB RAM. These are
> > collector's items and will increase in value.
> > Move out all existing stocks, and then fix fab to ship next version
> > with 512MB RAM + 4-8GB NAND. Moore's law and Windows Vista have
> > brought down the price of memory to dirt-cheap levels.
> >


As Tiago pointed out, the Geode is designed for DDR1 memory, which is 
not only off the leading price / technology curve, but is in the realm 
of "you have to make special arrangements to secure a reliable supply".  
There may be ways to adapt Geode to use DDR2 memory, but it's not a slam 
dunk.  You can't just do a simple board relayout and be good to go.  At 
the very least you would have to design and debug and production-qualify 
some sort of outboard interface chip, and redo all the memory integrity 
tests.  Do you have any idea how critical memory interface signal 
integrity is these days?

On the NAND front, the new generation of high-capacity NAND has moved to 
MLC (multi-level cell) designs, which have much smaller noise margins, 
so they need wider ECC than the XO's interface chip can support 
(otherwise the error rate becomes unacceptable).  Also, the new chips 
use 4K pages, which the XO's interface chip can't support.  So 
increasing the NAND size beyond the 2 GB level requires either multiple 
chips (higher cost and where to you find the board space), or 
farther-reaching changes than a simple board layout.

Moore's Law doesn't play out in a continuous fashion - it is "quantized" 
by generations of process technology.  At every such generation, 
interface details change to the point where you often have to switch big 
chunks of the design in a coordinated fashion.  Sometimes the change is 
in power supply voltages, other times in signaling methods (different 
types of drivers/receivers, DRAM vs. RDRAM vs. SDRAM vs. DDR vs. DDR2), 
other times in interconnect techniques (IDE -> Ultra IDE -> SATA, 
Multibus -> PCI -> AGP -> PCI-e, ISA -> LPC, USB, ...).

Any change at all requires an often-substantial amount of human effort 
to fix drivers, firmware, and documentation, to debug and fix unexpected 
interactions, etc.  Not to mention the hardest thing of all, which is to 
make all the new stuff work correctly during power management events.  
Turning power on and off sounds like it should be easy, but I can tell 
you from long experience that it is Really Tricky.

> > Chipmakers produced so much RAM and opened so many factories in
> > anticipation of Vista's insanely stupid bloatedness (that was the
> > first time I ever saw a software company make bloatedness and
> > inefficiency as a prime selling feature) that when it miserably
> > failed, they now have so much RAM on their hands they don't know what
> > to do with it.
>   

No, they produced a lot of RAM because if they wanted to be in the RAM 
business at all, they had to invest billions in the next generation of 
process technology, and to recoup that investment they had to sell a lot 
of RAM.

And Vista uses a lot of RAM because the next generation of computers has 
a lot of RAM.  Vicious cycle. 

By the way, Vista works reasonably well if you remove most of the 
demoware crap that the system vendors tend to load onto the machine. 
Early reports indicate that Windows 7 is more efficient than Windows XP 
(which by and large works reasonably well on the previous generation of 
hardware).  So the rant above is hard to take seriously, besides being 
totally irrelevant because the new RAM is incompatible with the old 
processor family.

It's also worth pointing out that the new low-power x86 processors, Atom 
being the poster child, are still stuck with power-hungry support chips 
- memory and display controllers.  That might change "soon",  but for 
now it's still the case.

> >
> > Chipmakers are in deep trouble right now and RAM prices have never
> > been so low. 

Everybody is in deep trouble right now, because the worldwide economy is 
in a "bust" cycle.  Which means that the chipmakers and the investors 
and the system manufacturers and pretty much everybody is being extra 
cautious about how they invest their money.

> Anyone looking to upgrade RAM, now is the time. OLPC?
> > Have Quanta start on the XO-1.1 with the increased storage + memory.
>   
That would be nice, if it were possible to do so without a deep 
redesign.  But, as outlined above, there's not a lot of low-hanging 
fruit.  Getting to the next level will require nontrivial changes, at 
which point you really want to cast a wider net to see how much more you 
could get by changing some assumptions.  We're working on it.

> > P.P.S. Beggars can't be choosers. Asking for donations but putting
> > conditions on it is just twisted ethics. Just sell 'em at $200 and
> > economies of scale + moore's law will take care of things.
>   

Really?  Have you done the math?