<qi-bot>
[commit] Maarten ter Huurne: ASFont: have a single routine for computing text width that also handles newlines. http://qi-hw.com/p/gmenu2x/ff546cd
<wpwrak>
rjeffries: sure, you can always add OTA. just kinda inconvenient in our case :)
<wpwrak>
rjeffries: ... considering that the device plugs into USB for power already, etc. also, for OTA, you would then have to address authentication.
<kristianpaul>
hmm, dfu i dint knew that existed :_)
<wpwrak>
kristianpaul: yet another thingie inherited from openmoko. there, it was quite an interesting experience ... combined with the effort of turning u-boot into the multi-tasking operating system it always secretly wished to be ;)
<wpwrak>
kristianpaul: but dfu per is is nice. you just have to do it without u-boot :)
<kristianpaul>
nice
<kristianpaul>
btw any show up from tuxbrain about the SMT?
<wpwrak>
kristianpaul: thundering silence so far
<wpwrak>
kristianpaul: i kinda wonder how this will proceed. maybe he'll pop up one day, with 100+ atben and atusb fresh from SMT, wondering if they work
<wpwrak>
we can make bets on the probability of that outcome :)
<wpwrak>
wolfspraul: what do you think ? diligent smt fab, but first time customer, new process, new board. board quite simple. what are the chances of an open loop production not ending in disaster ?
<wolfspraul>
open loop production
<wolfspraul>
:-)
<kristianpaul>
i understood he will send a few boards first to SMT..
<wolfspraul>
I don't know more than anybody else. My guess is Tuxbrain will continue, steady, but it's not very fast.
<wpwrak>
kristianpaul: you mean like 100 ? or a lot less than 100 ?
<wpwrak>
wolfspraul: naw, my question is how you guess is the probability of getting something that works if you aren't at the fab to check at least the first pieces that come off the line
<kristianpaul>
what was amount of boards for initial production?
<wpwrak>
wolfspraul: so far, it sounds as if he's all set to smt, yet he has no testing process. not has he exercised one.
<kristianpaul>
ah yes
<kristianpaul>
i hope not 100 wpwrak ..
<kristianpaul>
but for SMT that seems to be a minimun quantity i guess
<wpwrak>
kristianpaul: well, his plan it to make ~100 units. then see how things go. that's alright
<kristianpaul>
yeah i hope..
<wpwrak>
kristianpaul: he has about 500 PCBs. so he can make more boards if there's demand. so far, so good.
<wolfspraul>
the probability for that to come out as a working product is low, I'd say
<wpwrak>
kristianpaul: the problem is that it would already be quite bad if those first ~100 board would all fail to work
<wolfspraul>
that's the difficulty in hardware (manufacturing), you may even call it 'unfair', that a 1 penny mistake somewhere can render several thousand USD production results worthless
<wolfspraul>
and those losses _may_ be irrecoverable
<wolfspraul>
very annoying
<wolfspraul>
that's the #1 difference to software
<kristianpaul>
wpwrak: indeed
<wpwrak>
kristianpaul: in openmoko, we did it like this: when there was an smt run planned, a few hw and sw people went there. the fab ran one panel through the line, then waited.
<wolfspraul>
in software if you get something wrong, worst case your computer crashes, you restart
<wolfspraul>
in hardware you lost some irrecoverable cash -> bad
<wolfspraul>
and there is no mercy, the mistake may be tiny, yet the losses big
<wpwrak>
kristianpaul: the openmoko team tried to bring up the board. if it didn't work, they analyzed it for what would be the problem
<wolfspraul>
having said all that, ok the board is simple, but Tuxbrain is inexperienced - the smt people will have to guess a lot
<wolfspraul>
bottom line: high probability of failure
<wpwrak>
kristianpaul: if it did work, they would green light the rest of the productio run
<wpwrak>
wolfspraul: about my estimate too :-(
<wolfspraul>
the nasty thing is that the problem always hits you from an unexpected angle
<kristianpaul>
:-/
<wolfspraul>
murphy's law I guess
<wpwrak>
kristianpaul: i don't remember those teams ever coming back complaining there had been nothing for them to do ... (of course, gta01/02/03 were more complex than atben/atusb, but still ...)
<wolfspraul>
it's amazing how many thousand (indeed) little things can go wrong
<wpwrak>
wolfspraul: yeah. the only protection is testing.
<wolfspraul>
so yes, Tuxbrain needs to learn how to flash & test the boards, he needs to be at the line on production day, he needs to be suspicious and on high coffeine level, and press the _STOP_ button at any time he smells anything fishy.
<wolfspraul>
better press is one time too often than one time too few
<wpwrak>
right now, there are a few holes in the testing process i've set up, but i have the main ingredients
<wpwrak>
exactly
<wpwrak>
ironically, the smt fab is something like half an hour from his headquarters, so it's not really a big deal for him to be there
<kristianpaul>
good :-)
<wolfspraul>
in the m1 rc2 run, we had one such scary moment
<wolfspraul>
first board came off the line, Adam started testing
<wolfspraul>
something was strange, problem with the video-in crystal
<wolfspraul>
now... he he. stress level increases :-)
<wpwrak>
hates crystals
<wolfspraul>
gotta make a very quick decision: tell the guys to stop the line or not :-)
<wolfspraul>
production manager stays in the door and looks and you and awaits your call
<wolfspraul>
great!
<wolfspraul>
shit decision!
<kristianpaul>
first board came and SMT stoped until adam finished testing or was all in parallel?
<wpwrak>
it's the first board. of course they can hold it for a bit ...
<kristianpaul>
ah ok
<wolfspraul>
pick & place is hammering away, argh...
<wolfspraul>
"Adam, can you test a bit faster?"
<wolfspraul>
"should we stop?"
<wolfspraul>
:-)
<wpwrak>
the solder paste can be out in the open for 0.5-1 hour without problems ...
<wolfspraul>
it became too scary so we pushed the red button at some point, for about 45 minutes
<wolfspraul>
that means everybody on the line will be standing still doing nothing
<wpwrak>
good :)
<wolfspraul>
also not great, because they cannot switch (too many of your parts and settings everywhere)
<wpwrak>
they probably spent two hours setting up the line. so what :)
<wolfspraul>
after 45 minutes, it turned out it was a local problem with that crystal, and we let the run resume
<wpwrak>
.tw, right ?
<wolfspraul>
yes
<wpwrak>
good. they're not rock bottom, gotta be hyper efficient or else
<wolfspraul>
oh it's all fine and professional. but someone has to make decisions.
<wpwrak>
i expect spain to be even more relaxed in that regard
<wolfspraul>
and if those decisions are wrong, a small mistake can cause a large financial damage.
<wolfspraul>
that's manufacturing...
<wpwrak>
indeed
<wolfspraul>
because once the stuff is assembled/made, it may (may) be impossible economically to fix it
<wolfspraul>
so it's cheaper to throw away and do another run
<wpwrak>
e.g., a BGA turned some 90 degrees :)
<kristianpaul>
no no :-|
<wpwrak>
well, already a QFN would be more than sufficient ;-)
<wolfspraul>
kristianpaul: what you don't like that idea?
<wolfspraul>
man you should hang out on production lines more. it's fun!
<wolfspraul>
pretty close to chaos management...
<kristianpaul>
it seems :-)
<wpwrak>
atudb has two of them. 1/16 chance to get both right on the first try ;-)
<kristianpaul>
wolfspraul: not that i dont liked it, but bit scaring indeed
<wolfspraul>
oh, everything will happen to you there
<kristianpaul>
hmm i remenber the the batery explode for sie run2?..
<wolfspraul>
the unfortunate thing for a software guy like me, maybe the depressing part also, is that there is little to learn from some of those mistakes
<wolfspraul>
it's not like we are building a perfect production machine, and one day it will be bug free
<wolfspraul>
it will never
<wolfspraul>
it's just a never-ending chaos management exercise
<wolfspraul>
maybe that's why the Chinese are so unbeatably good at it (including the ones in Taiwan of course)
<kristianpaul>
hmm, manually mounting..
<wpwrak>
i guess you just have to design your process such that it can accommodate this kind of problems
<kristianpaul>
thats something to watch out
<wpwrak>
just like working with unreliable people. you try to gauge them and increase your safety margin. if they do what's expected of them, great. if they don't, you already have a plan B.
<wpwrak>
just good old risk management :)
<wpwrak>
the nice thing about automated processes is that once you get them to work, they tend to continue working
<wpwrak>
you still have to check from time to time if something drifted out of the safe range,
<wpwrak>
or if someone slipped you a bad batch of components, but in general, it quality is predictable
<wpwrak>
with a manual process, you depend entirely on people catching errors. so if they've been out all night the day before, yield plummets. and so on.
<wolfspraul>
kristianpaul: keep in mind that we are discussing technical correctness now. but manufacturing is about economics mostly. every thing you manufacture only has a certain (small) value X, and the next one may pose new challenges, because it's all physical goods and physical processes.
<wolfspraul>
those physical processes only become repeatable/predictable to a certain degree
<wolfspraul>
so that combined with the very small amount of money you can make for each one is hard
<wpwrak>
funny. hackaday overtook slashdot today. well, the day is still young :)
<wpwrak>
(the UTC day, to be precise)
<wolfspraul>
I think you are continuously balancing technical correctness (perfection of technical process), and cost on the other side.
<wolfspraul>
it is really just amazing/unbelievable that you can make a whole smartphone for 30 USD
<wolfspraul>
hundreds of companies collaborate on that
<wpwrak>
indeed. electronics are almost scare in that regard :)
<wpwrak>
s/scare/scary/
<wolfspraul>
so far we (copyleft hw) haven't found a way to successfully hook ourselves into this power yet, with our beloved free software :-)
<wolfspraul>
but we'll get there...
<wpwrak>
what we need is volume ... once you have that, everything gets easier
<kristianpaul>
smartphones are getting volume this days :-)
<wpwrak>
and everyone is at each other's throat ;-) naw, don't do smartphones, don't to tablets. waste of time.
<kristianpaul>
\. thinking? ;-)
<wolfspraul>
I fully agree on tablets, phones - right now.
<wolfspraul>
btw.. about volume. I think volume is the mythical man month of hardware.
<wolfspraul>
volume = investment
<kristianpaul>
good point
<wolfspraul>
looking at volume is too superficial
<wolfspraul>
of course higher volume is 'nice'
<wolfspraul>
but where does it come from?
<wolfspraul>
high volume = large order
<wpwrak>
of course :)
<wolfspraul>
someone has to put in a firm large order, with large down-payment
<wolfspraul>
and when can you do that?
<wpwrak>
order or upfront investment
<wolfspraul>
when you have a strong forecast
<kristianpaul>
at first place i see volume as demand, but indeed you can create demand with good investement
<wolfspraul>
it's the same
<wolfspraul>
order = investment (binds capital)
<wolfspraul>
wait wait
<wolfspraul>
the volume is the consequence
<wpwrak>
if you get the kind of order that binds capital
<wolfspraul>
it's not the first thing
<wolfspraul>
the first thing is potential market - how many of this thing can you sell?
<wolfspraul>
if the market is 100 pieces, investing for a 'volume' production of 1 million is very stupid
<wolfspraul>
nothing has _infinite_ market size
<wolfspraul>
so who determines that number?
<wpwrak>
there are two types: order to build and order to ship. if you tell them there's a 2 month wait, they'll run
<wolfspraul>
the market size = order = volume number
<wolfspraul>
think about it
<wolfspraul>
we assume every product can essentially have an infinite market, but of course that's wrong
<wolfspraul>
so every product has a theoretical maximum
<wolfspraul>
limited by all sorts of things, of course also strength of brand, strength of distributors/channel, and so on
<wpwrak>
market size is difficult on the lower end. if your target market is tens of millions, you can be off by quite a bit and it's still tens of millions.
<wolfspraul>
now, once you determined that number, you invest against this number
<wolfspraul>
in anticipation that you can perfectly execute according to your market and demand forecast
<wpwrak>
in our case, it can be 100, 1k or 10k. and it's more or less random.
<wolfspraul>
my point is: talk of volume clouds your vision
<wolfspraul>
volume is a consequence
<wolfspraul>
higher volume means higher investment
<wolfspraul>
the higher investment needs to be recouped, so you have to sell that higher volume too
<wolfspraul>
otherwise you will loose even more money with 'higher volume'
<wolfspraul>
this is assuming that we are not talking about technical difficulties that mandate higher volume
<wpwrak>
the problem is that price expectations are usually based on high-volume products
<wolfspraul>
I'm not talking about that.
<wolfspraul>
wait
<wolfspraul>
someone is making a large investment
<wpwrak>
so if you make a USD 5000 iPhone killer, who will think it's a bargain ?
<wolfspraul>
because he is sure that he can sell a large number of this product
<wolfspraul>
correct. if your target market is 'iphone', then you can do all the numbers with a forecast of 10 million in the first year.
<wolfspraul>
but then you also need a product that can actually sell to some customer who chooses your product over the iphone he would otherwise buy.
<wolfspraul>
and you need a big bad strategy for that, including strong brand, strong network pull, strong software, strong app market, strong lots of things.
<wpwrak>
there's also contamination around the edges. e.g., would you buy a USD 100 wpan card, knowing that you can get some generic USB wlan for maybe USD 10-20 ?
<wolfspraul>
I tried to make a point about volume.
<wolfspraul>
it is not true "high volume = low unit price"
<wolfspraul>
that's a very selective view
<wpwrak>
the price is not only determined by your product. it's the market value of "comparable" products
<wolfspraul>
high volume requires large investment
<wolfspraul>
the large investment indeed drives the unit price down
<wolfspraul>
but...
<wolfspraul>
you made a large investment which you have to recoup
<wolfspraul>
understand the problem?
<kristianpaul>
i think so
<wolfspraul>
so if you end up selling only 1% of what you thought you could sell, your "high volume = low unit price" will have very very badly failed ;-)
<wpwrak>
if the products your is being compared with are high volume, you'll have an extremely hard time establishing an honest by high price
<kristianpaul>
is not about volume :-)
<wolfspraul>
and in hardware it's super bad to have a lot of stock that doesn't sell sitting around, because your competitors constantly make better and better hardware
<wpwrak>
wolfspraul: yes, but that's the investor's point of view. are your customers investors ?
<wolfspraul>
so high volume may also increase the unit price
<wolfspraul>
unless your sales force is very strong, and the whole forecast and machine actually works
<wolfspraul>
my customers will not benefit very long from a bankrupt company
<wpwrak>
;-)
<wolfspraul>
high volume = large investment
<wpwrak>
correct
<wolfspraul>
the idea "high volume = low unit price" is flawed
<wolfspraul>
too superficial
<wolfspraul>
then it's much more important to have a realistic forecast first
<wpwrak>
no, it's not. high volume allows you to buy cheaply
<wolfspraul>
yes but you have to pay = financing
<wpwrak>
the question is if you can sell your high volume
<wolfspraul>
yes
<wpwrak>
if you can't, you're screwed, i agree
<wolfspraul>
yes
<kristianpaul>
large investment = big bad strategy ?
<wolfspraul>
then the high volume = low unit price idea will have badly backfired
<wolfspraul>
no no
<wolfspraul>
forecast needs to be correct
<wolfspraul>
not too high, not too low
<kristianpaul>
:-/
<wolfspraul>
the volume and unit prices are just consequences of that then
<wpwrak>
but if you volume is low and your per unit price is high, you people may not understand that
<wolfspraul>
one example from Apple
<wpwrak>
s/you people/people/
<wolfspraul>
so they put Samsung chips into their first-gen iphones
<wolfspraul>
supposedly paid about 20 USD / chip to Samsung
<wolfspraul>
ok?
<wolfspraul>
then they made a decision to do that part themselves
<wolfspraul>
they bought patsemi for 200 million USD
<wolfspraul>
and a lot of other investments for several years
<wolfspraul>
the total estimate I read what Apple spent before they were able to sell their first A4-based product was 1 billion USD!
<wolfspraul>
now... compare that to the 20 USD they could also have paid to Samsung
<wolfspraul>
that means they had a forecast that said "we will sell 50 million or more units of this"!
<wolfspraul>
50 million!
<wolfspraul>
and they made that decision back in 2008 or so. if you can zoom your mind back to 2008, you realize what an extraordinary entrepreneurial risk it is to make an investment decision based on a forecast of 50 million units before it pays off!
<wolfspraul>
so yes, superficially high volume = low unit cost. but practically going there means a large up-front investment, which needs to be recouped.
<wolfspraul>
if Apple would have sold only 5 million of those phones, each of their home-made cpus would have cost them 200 USD :-)
<wolfspraul>
wpwrak: of course if your unit price is high you may not sell. that's why there is such strong consolidation around a few product categories.
<wpwrak>
yup
<wolfspraul>
your customer will not care why your unit price is high or low
<wolfspraul>
so if it is high (if you are serving a niche market), your product better have some very unique features that will not let your customer take a more standard category product and tweak that instead.
<wpwrak>
your price is basically bracketed between the production cost, which is driven by volume, and what the competition costs, which is - at least in part - driven by their volume
<wolfspraul>
production cost is driven by investment
<wolfspraul>
volume is a consequence of that
<wolfspraul>
once Apple decided to make the A4, they knew they were on a path that would cost them 1 billion USD
<wpwrak>
that's why i keep on asking about progress with bringing investors aboard ;-)
<wolfspraul>
they accepted that that could mean that the CPU in each iphone costs apple 200 USD
<wolfspraul>
if the iphone would not sell well
<wpwrak>
investment = liberty to think bigger
<wpwrak>
(unique features) yes, the more competition you can eliminate, the better. of course, you must pay attention that you're not painting yourself into a corner.
<kristianpaul>
1 billion to be recouped.. that will take a while i think
<wolfspraul>
my point was that apple knowingly accepted the risk that their CPU price would go from 20 USD / piece to 200 USD / piece, if the iphone with that chip would 'only' have sold 5 million units.
<kristianpaul>
yup
<wolfspraul>
at 50 million (!) they would be down to the same price they could have paid Samsung
<wolfspraul>
and at 100 million or so the "high volume = low price" thing would have actually worked well in this case
<wpwrak>
i think our problem at the moment is not so much a lack of market but an immature market. e.g., ben sells badly because it sells to a market that would either expect much more powerful products or at least more frequent updates. i'm sceptical about mm1, too. i like the technology, but i don't know if it can sell. maybe it'll need an mm2 or maybe an mm3 before it becomes a hit.
<wolfspraul>
sure it's risky. we need to market them well.
<wolfspraul>
and ideally find a stronger marketing partner, which I will try much harder once I have more m1 at hand, and more complete (with accessories and box and stuff).
<wpwrak>
i think there's a lot of stuff with a good long-term prognosis but a poor short-term prognosis. the "grow by selling" approach doesn't really work there.
<wpwrak>
in my opinion, nanonote needs a financing horizon of about one year. the project is already about half a year in debt, so it's just about half an extra year. i'd consider milkymist even more long-term. i think the product focus is good, but nothing should depend on its commercial success (in the short term)
<wpwrak>
anyway, time for a bit of a nap :)
<phr>
are there a good quantity of nanonote available?  i thought maybe you are about to run out of them  b/c of slashdot or something
<wolfspraul>
phr: no way. if I run out of them I make more :-)
<zhicheng>
xiangfu: Hi~
<zhicheng>
xiangfu: Sorry for yesterday, my network is not very stable.
<xiangfu>
zhicheng: hi
<dvdk>
hi
<dvdk>
any qiots attending LinuxTAG? Going to be at linuxtag tomorrow (friday)
<wolfspraul>
qiots is good
<wolfspraul>
how is linuxtag nowadays? last time I must have been there like 10 years ago, in Karlsruhe I think
<LunohoD>
it's a pity that linuxtag was moved to berlin
<dvdk>
linuxtag is very small
<dvdk>
but some interesting talks maybe
<dvdk>
sometimes one stumbles into OSS projects one didn't notice before.
<dvdk>
that's how i got involved with the nanonote
<kyak>
dvdk: heard about it on some conference?
<dvdk>
this is the list of projects represented this year