Robert Wessel <[email protected]> writes:
> That number is certainly too low.  EC12s have 1-4 books, each with six
> processor chips (yes that comes to 144, only 120 can be used), and
> anyone not buying a configuration topped out for the number of books
> installed will have more chips.  BC12s have either one or two chips.
> On both models the revenues for the sub-capacity models is
> significantly less on a per-processor chip basis than your calculation
> would imply (nominally a EC12-401 is about $840K, while a 89-times
> faster single book -720 with the identical number of processor chips
> is $10.8M - the BCs have even worse ratios).
>
> Also, you're assuming list prices for full function CPUs (if you're
> buying IFLs, zAAPs, zIIPs, ICFs, etc., they're cheaper).
>
> So the real number is likely several times higher, probably around an
> order of magnitude.
>
> Another way to do the calculation is to assume that the 10,000
> machines in existence are replaced every three years.  Even if those
> were all maxed out EC12-7A1s, that's only 120,000 chips per year.  And
> while neither that replacement rate or the machine size is remotely
> realistic, it puts a hard upper cap on the number of dies to be
> produced for sale in machines.  Perhaps a more reasonable assumption
> would be a four year replacement cycle and an average of ten processor
> chips per box, which gets you 25,000 chips/year.
>
> In any event, that doesn't help IBM's costs any.  The total
> manufacturing costs for those chips is minuscule, and the NREs still
> have to be spread over the same revenue base.

re:
http://www.garlic.com/~lynn/2014j.html#85 Demonstrating Moore's law
http://www.garlic.com/~lynn/2014j.html#86 Demonstrating Moore's law
http://www.garlic.com/~lynn/2014j.html#87 Demonstrating Moore's law

IBM mainframe processor financials had been doing about $4B-$5B annum
for a decade or so (but that number has dropped recently).  A much
bigger revenue number for IBM mainframe group has been the software,
services, and storage it sells (total mainframe group revenue has avg. a
little over 6times its processor system revenue, and is extremely
profitable being as much as 40% of total company profit).

assuming that max configured systems work out to have the best price per
processor ... and max configured systems running around $30M ... ibm has
been selling the equivalent of 133-166 max. configured systems per annum
for a decade or so (but recently dropped to less than half that) aka if
there are a larger number of smaller systems, assumption is that the
price/processor would be higher, and fewer total processors and therefor
chips ... using max configured systems equivalents would tend to put an
upper bound on total number of processors/annum).

At the upper limit, the previous decade would have been the equivalent
of 1660 new max. configured systems (@166/annum) ... and 1st qtr 2014
mainframe processor numbers on annualized basis is 56 max. configured
ec12 systems. so instead of 120processors in (@6processors/chip,
20chips) ... say ec12 actually has 240processors (although only max 101
are directly configured) that would make it 40chips & @56systems/annum
that is 2240chips/annum ... remapped to 14nm technology and 450mm wafers
that is less than 2 wafers (@1383chips/wafer), rather than less than
one.

If the aggregate mainframe chips that are less than couple dozen wafers
(or as few as one) in production fab mostly used for other things
... then the fab costs are amortized across the other chips ... and the
mainframe chips are nearly incidental. However, if the fab had to
recover the total technology costs just from mainframe sales ... it
would be quite a bit more expensive.

that is major reason that mainframe has migrated to being built using
mostly industry standard technology, industry standard disks, industry
standard fibre channel, industry standard chip process, etc.

-- 
virtualization experience starting Jan1968, online at home since Mar1970

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to