Hardware Progress: $397 (fwd)

Robert G. Brown rgb at phy.duke.edu
Wed Mar 27 14:20:54 PST 2002


On Wed, 27 Mar 2002, Christoph Wasshuber wrote:

> I would think a cost reduction by 2-5x maximum
> in 2006.

More practically, microprocessor manufacturers and other tech component
manufacturers don't generally focus on reducing consumer costs, they
focus on increasing speed (and other overall performance measures) at
constant consumer cost and relatively high margins.  They can do this
because tech profit margins depend to a significant extent on demand in
a fairly inelastic market with very limited competition.

[Parenthetically:  A good example of this is hard disk.  I'm fairly
certain that 10 GB hard disks could be sold for $25 retail, much as
floppy drives are today, if vendors were forced by competition to focus
on decreasing price instead of maintaining margin by increasing disk
size.  If, however, there were $25 10 GB hard disks, perhaps 2/3 of
system buyers would probably choose to buy them where now they are
"forced" to buy $100+ 40 GB drives, soon to be (I'm sure 60 GB $100
drives). For some years manufacturers have worked fairly hard to ensure
that the cheapest drives they make to sell retail for around $100.  Even
so competition gradually pushes down the price of the cheapest
mass-market retail hard disk from thousands of dollars in the 80's (in
1980's dollars!) to maybe $80 today.

Compare disk to memory, where there are far more manufacturers and
(aside from massive price fluctuations caused by supply side dimples)
absolute price of the cheapest components has come down remarkably.
Think of the price of 512 MB SDRAM DIMMS a couple of years ago, and the
cost of 64 or 128 MB DIMMS today.]

So don't count on seeing much of a price reduction in CPU (top to
bottom), count on seeing a performance increase at constant cost with a
more or less constant minimum buy-in price (for the older technology)
and a consistent, much higher price for bleeding-edge technology. If a
price drop does occur, it will more likely be related to competition and
manufacturing scale than to improvements in VLSI scale.

As for as what that increase (at constant cost) will be -- I believe
that the average performance doubling time is roughly 16 months
(somewhere between 12 and 18) although the actual doubling time is very
irregular with quantum jumps (new processor families) and then smooth
increases as they ramp up the clock and clean up the design.  Over four
years I'd believe anything from 2^5 = 32x down to 2^3 = 8x for CPU speed
alone, depending on whether the four year boundary begins and/or ends
before and/or after a quantum jump (which can yield as much as factor of
2 relative to clock).  I'd be very surprised if CPU speed "only"
increases by a factor of 5-6 over four years -- that would be a doubling
time of around 18 months (at the high side of the range) and we KNOW
that there will be The Great 64-Bit Crossover in the next year or two
(see e.g. the AMD "processor roadmap") which will probably be worth most
of a factor of two alone.

   rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu






More information about the Beowulf mailing list