Historical memory prices
From
EricP@ThatWouldBeTelling@thevillage.com to
comp.arch on Wed Sep 16 13:58:02 2026
From Newsgroup: comp.arch
In case anyone is interested, I came across this article
on the state of the memory market circa 1973.
The Intel 8080 and Motorola 6800 were launched in 1974
and 8086 launched in 1978.
In 1975 DEC offered the LSI-11 and PDP-11/44 with either
core or semiconductor (using 4k*1b chips) memory modules.
It talks about costs in cents-per-bit, and "The total
random-access-memory market is sizable, with estimates putting
it at $300 million, excluding IBM. In 1973, between 30 billion
and 40 billion bits are expected to be sold".
Electronics/February 1,1973
Core memories are still on top
------------------------------
The booming memory market is still overwhelmingly dominated
by cores; semiconductor units will not draw even before 1976
by Paul Franson. Los Angeles bureau manager
A few facts seem lost in all of the noise about semiconductor memories replacing cores in computers. For one, cores still hold 90% of the
sockets in traditional mainframes outside the IBM captive market.
Also, both cores and semiconductors are enjoying unparalleled growth,
which, though limited by capacity, is expected to continue through at
least 1976. After that, semiconductors should overtake and pass
coresrCoa bit later than many semiconductor people had predicted.
One reason the semiconductor takeover has been slower to come
than the chronically optimistic semiconductor houses had predicted is
that the present standard semiconductor memory, the 1,024-bit
p-channel metal-oxide-semiconductor memory, doesn't appear to be the
part for mass mainframe storage.
Users are looking for the 4,096-bit n-channel product, just now
appearing in sample quantity, to solve their demands for low cost and
high speed. While fast semiconductor storage is finding widening use
in mainframes, core is being used more and more for massive bulk storage
that is itself encroaching on traditional, but slower, electromechanical memories. Semiconductors are also uncovering applications in which
cores have never been used: smart terminals, calculators, and some
peripherals are examples.
The total random-access-memory market is sizable, with estimates putting
it at $300 million, excluding IBM. In 1973, between 30 billion and
40 billion bits are expected to be sold, with cores now accounting for
90% of shipments. By 1976, however, the ratio should change dramatically:
in a market for 90 billion to 100 billion bits, semiconductors and cores
will split 50-50. Semiconductor usage is doubling each year from its
present low base, with core rising at 15% to 20% annually. The growth
rates will slow, but even core is expected to grow in number of bits
through at least 1976.
Present large-quantity prices-per-bit at the systems level are
typically 0.7 to 1 cent for core, and 1 to 1.7 cents for semiconductors.
Costs are higher if power supplies and enclosures are included,
but they're dropping, in any case.
Robert Lowry, president of Technology Marketing Inc., Costa Mesa,
Calif., designers of custom computer systems, says the large-volume core
user can purchase a complete 600-nanosecond system of 8,192 bits by
18 words for about 1 cent per bit, or buy core stacks and assemble the electronics for under 0.5 cent per bit. Buying raw cores at 0.1 cent and stringing them brings the cost down 0.1 cent, and the company that
makes its own cores can end up paying as little as 0.35 to 0.38 cent.
These figures, which also are quoted by others, explain why such
large core users as Data General Corp. of Southboro, Mass.,
and Digital Equipment Corp. of Maynard, Mass., have recently set up
their own plants to manufacture cores. "Data General should be able to
pay off its half-million-dollar investment in ayear by making a billion
cores," says Lowry. A Data General spokesman agrees: "Don't look at us
as being committed to cores. We are committed for two years, and the
facility will pay for itself by then." Similarly, Graham Tyson, president
of core maker Data Products Corp. says. "We paid off our last capital expenditure for cores in 28 weeks."
Ampex's Computer Products division in Marina del Rey. Calif., and
Electronic Memories and Magnetics (Em&m). Hawthorne, Calif.. are the
two largest core producers, each shipping about 200 million bits per
week, with Ampex probably ahead. After them come Data Products,
Mountain View, Calif.. with 125 million a week: Fabri-Tek.
Minneapolis. Minn.. with 70-80 million a week; and Lockheed Electronics,
not necessarily in that order.
Semiconductor memory costs are much harder to pin down. Figures
that are confirmed by users, however. come from Intel Corp. in
Santa Clara. Calif., which developed the 1103. the most popular
1,024-bit mos memory. Mike Markkula. Intel's U.S. marketing
manager for components, says "Minicomputer guys are paying about 0.4
cent per bit for 1,024-bit RAMS. On top of this, the other systems costs
are from 0.7 to 1 cent per bit." Memories being ordered (but not
built) today, he says. are somewhat cheaper, with the semiconductor
part at about 0.3 cent per bit.
Markkula says that 350,000 1103s
are being shipped per month, and with other 1,024-bit parts
(Advanced Memory Systems 6002, Mostek 4006, and National
Semiconductor 5260), 10 billion bits will be shipped in 1973.
He also feels that the 1.024-bit RAM will reach a
peak in 1977, with the 4,096-bit RAM passing it in 1976.
The great 1103 race. Intel and Canada's Microsystems International
Ltd. (MIL) are jockeying for first place in 1103 business,
with MIL probably ahead. Behind them in 1.024-bit mos memories
come Advanced Memory Systems, Mostek, Texas Instruments,
and a host of others.
The most popular semiconductor memory after the 1,024-bit mos unit
is the 256-bit bipolar RAM, which now has 20% of the market,
according to Jerry Moffitt, mos marketing manager at Texas Instruments
in Houston. compared to 70% for the 1,024-bit MOS RAMS.
Fairchild Semiconductor is biggest here, with Intersil, Intel,
Motorola, National and TI also shipping.
Most experts don't see the 1,024-bit MOS RAMs reaching the
0.1-cent-per-bit chip cost level required to cause a really strong
movement away from cores. That honor seems destined to fall to the
4,096-bit n-channel RAM now being introduced by various companies,
which project systems costs of 0.5 cent by 1976. Intel and Mostek
first showed parts last year, and MIL has introduced a product.
But the big RAMS are not expected to have a major impact until 1974,
says Phillip Harding. director of marketing for the Electronic
Memories division of EM&M.
Who uses what. Semiconductor memories are now used only in
some IBM equipment and a few minicomputers, but most new major
mainframes are being designed around semiconductor storage.
At Honeywell Information Systems in Waltham. Mass., for example, a
spokesman says that. all new product development is committed to semiconductors. However. HIS, like NCR but unlike all the other major
computer makers, has no core capability.
At Digital Equipment Corp., the largest minicomputer company,
vice-president Andrew Knowles says that DEC is shipping more machines
with semiconductor memories than ever, but core shipments are growing
even faster. He attributes the growth to the fast-growing demand
for small computers, which still use cores, and to the higher cost
of semiconductors.
Among other companies, Control Data Corp. is still holding out for
cores. However, according to Curtis W. Fritze, vice-president for
corporate planning, if forced to choose, CDC "tends toward the
more sophisticated bipolar memories. Eventually they will provide
an order of magnitude cost advantage over cores," he says.
And Univac hasn't decided whether to go with semiconductors or
the plated-wire memories it has been using.
At Varian Data Machines, Irvine, Calif., Warren Sullivan, director of
product management, says only 20% of the orders for the Varian 73 are
for the more expensive but faster (330 nanoseconds versus 600 ns)
semiconductor memory, but he expects that to change to 50% before
long. Sullivan says that users are buying more memory in general.
Improvements expected in semiconductor RAMS include, besides the
higher density and lower cost expected from the 4,096 n-mos chips, easier-to-use, lower-power chips. MIL expects to have an eight-chip
32,768-bit memory module in a single 1.5-by-2-inch dual in-line
package by the fourth quarter.
Tape and stacks. Meanwhile. among the major improvements in
cores in recent years are the tape method of making cores used by
Lockheed and Data Products Corp. of Mountain View, Calif., and the
substitution of planar "stacks" for the earlier three-dimensional ones.
Core makers are continually stuffing more cores on one board, and
sharing drive and sense circuit among more cores.
Common are 8,I92-bit mats, with work being done in 16,384 bits.
R.D. Miller of Data Products points out that, in the core business,
twice as many bits does not mean an automatic doubling in price.
Instead, doubling the bits usually costs only 20% to 25% more.
Smaller cores are also in the works, as are improved materials,
like the Ampex TIN (temperature independent) core that eliminates temperature-compensating circuitry. Ampex also has reduced kiln time
from 8 hours to 45 minutes, and improved yields from 20% to 80% over
past years. All core makers are taking advantage of dropping
semiconductor prices; only part of the system cost is in the cores.
And more exotic developments yet may lie in store. Thom Harleman,
manager of marketing at Ampex Computer Products division, points out,
"It's been known for some time that it's possible to store more than
one bit in each core, but the uniformity of present cores allows us
to do things that weren't practical before."
Still, there seems to be a limit to how far core prices can drop.
Richard Egan, marketing vice-president of Cambridge Memories in
Concord, Mass., referring to today's larger mats with their longer
sense lines, states, "I think the string is running out on core memories.
The handwriting is on the wall."
--- Synchronet 3.22a-Linux NewsLink 1.2