LVDS and terminators
jmcgrath at qntm.com
Wed Oct 4 18:32:31 PDT 1995
Reply to: RE>>LVDS and terminators
> You are not telling me that I should wait for 6 months to buy my 4GB SCSI
> drive, since it'll be free by then? Well, I guess not, or both you and I
> are out on the street and can't afford to buy one.
In 6 months using my (conservative) cost decline a 4 GB drive loses
at least $240 of its value. In 1 year 4 GB drives have seen a $600
price decline. In many ways this is the FUNDAMENTAL economic
fact of life for the disk drive industry. It is why we have periodic price
wars (once you start building inventory the lower AUP you will eventually
get for your drive makes you want to lower inventory real fast - so
you cut current AUP to move inventory, igniting a price war). If it
were not for these dynamics our (Quantum and WD) stock would be priced
3 or 4 times higher (price wars make for highly variable EPS, which in
turn is punished by the market with a lower P/E ratio for our stock).
> Being in the business of selling the chips, I'd love the idea of everyone
> having to buy more expensive chips, that can do everything.
> But despite high cost of inventory, people do produce multiple products.
> Why? Mostly, because the cheapest products (i.e., highest volume products)
> always want to be cheaper, and don't want to carry excess baggage for the
> more expensive, lower volume products.
What does this have to do with the discussion? I told you what Quantum
will be buying (we actually do our own chips, but we do buy terminators).
And whether your chip has a common design or not has nothing to do
with whether the IO cells are common and thus whether I can keep 1
inventory. Whatever the buffer bendwidth is, it will be the same for
all of my PC products and all of my server products (although they
may differ between them). My goal is not to have multiple PC
products and multiple server products - I am already using different
chips for PCs and servers, so I really don't care about any other
differences between classes of products - I want the same drive in a
given class of product (e.g. 1 7200 rpm 4 GB drive, not two, a SE
and a LVDS). That implies the same ASIC and terminator for both
SE and LVDS for the same class of drive, and thus a miricle driver.
Remember, LVDS is not new - the committee discussed the concept
5 years ago, Bill Spence spent a lot of time on it, but the inventory
problem made adoption impossible. In many ways the SE/LVDS
combination is more important than the LVDS itself is, since without
it people simply could not afford to adopt the new technology without
pricing it very high (to cover the inventory risk). Without the miricle
driver our interest would have been greatly reduced.
> And your point about inventory is well taken. But that doesn't seem to
> directly translate to something having to transform from SE to LVDS and
> back in an instant notice. It seems to me it's more like the installation
> stays one setup or another, but you don't have to repurchase all your drives
> when you upgrade. Otherwise, why the hell are you doing with expensive
> a relative term, however) twisted-pair flat cable with teflon insulator?
> Or is there someone who sells the cable that transforms from A-cable to
> Fast-40 cable depending on the state of diffsens?
I don't get it. You do have to handle hot plugging, which can change you from
to SE quickly. And we have no cable specifications for LVDS that are any
different from the ones we have for Fast-20 (and I would strongly
> Yes, we do sometimes engage in practice where same silicon dies are put
> in different packages with different part number, and with some feature
> differentiation, priced differetly ("the price has nothing to do with
> the cost" - biz 101). But that is of no concern of NATIONAL STANDARDS
> that we are talking about here....
Properly speaking a miricle driver is no concern of the standard - as long as
I can implement it and conform to the standard. We are taking the approach
that the best way to get there is to start with a miricle driver analysis
and then deduce the formal standards requirements.
>> Obviously we will NOT be buying terminators that cannot do both LVDS
>> and SE for our drives (big hint to terminator manufacturers everywhere),
>> since that would defeat the purpose of the SE/LVDS in the drive.
> You are touching upon the exact point I was trying to make.
> If you studied the terminator requirement for LVDS and compared that to
> single ended, you can easily imagine that your "switchable terminator"
> is not *A* terminator but two terminators in a box.
> So you were talking about the cost??? So you or your customer is ready to
> spend extra for the do-it-all terminators and do-it-all drives even though
> you know you'll be staying on a mere single ended fast SCSI, because you
> want to keep your CDROM drive want to reserve the right to go Fast-40 one
> of these days?
I don't know how many times I have to say it, but to Quantum terminators
on our drive without switching capability are brain dead. We could only sell
those into systems that provided their own termination (and so we disable
ours). This greatly complicates life, since the easiest way to dispose of
inventory is through distribution, adn there we have no idea of whether
the final system has termination or not. A drive with a LVDS terminator
sold into a system without termination which wants to use SE is a dead
> We need more defined migration path to strategically place the Fast-40.
> It indeed is the main theme of the upcoming Fast-40 marketing meeting
> that you are hosting.
I agree that people should discuss this there - that is the purpose of the
meeting. But to be honest, it is so clear from where I sit that I have a hard
time seeing any possible reason not to do a miricle driver. My goal is not
to get everyone in the industry to agree on a path, but to provide input into
the standards process. In my opinion, if a significant number of people
want something, we should define it and let the market decide. Since isolated
SE and LVDS is just a subset of the miricle driver (i.e. the miricle driver is
hard problem), and since we need the miricle driver, that is (to me) the
focal point for the standards work.
More information about the T10