Zen and the Struggle Between Communism and CISC
The "tongue-in-cheek" essay which appears below was published in EE Times in May 1991 (in a slightly
condensed form) with the above "RISC Police" cartoon. It also was
presented at the IEEE Asilomar Microcomputer Workshop in April 1991
where it was used as a test of Dave
"The Grandfather of RISC" Patterson's sense of humor...he proved not
to have any! In addition, after reading the article in EE Times, Roger
Ross [CEO of then Cypress' Ross Technologies],
contacted SPARC International and tried to get Metaflow Technologies censured by
the organization. (SPARC
International is Sun Microsystems'
[puppet] organization which promotes the so-called
"open" SPARC RISC architecture.) To our surprise, Gordon Moore,
then Chairman of the Board of Intel
Corporation, sent us a nice thank-you
letter for writing the EE Times article, and for our "addition of
a ray of sanity" to the RISC-CISC debate.
Everything we said then about RISC vs. CISC in the article has turned
out to be true! What is somewhat ironic is that at the time we wrote
the article, and for the next several years, Metaflow was busy designing
a "RISC" SPARC microprocessor for Hyundai Electronics (code named
A Wall Street Journal headline from that
time captured the general mood of the microprocessor design
community, which had all but written off CISC...and Intel 80x86 processors.
However, what most people failed to grasp at the time is that the
so-called RISC chips we were all designing (e.g., SPARC, MIPS,
RS-6000, HP-PA, etc.) were, at best, only marginally less complex than
the CISC chips against which we were competing (e.g., 80x86 and 68K).
In fact, out-of-order, speculative-execution microarchitectures
(something Metaflow pioneered) were destined to "level the playing
field" and could (and indeed would) erase any performance advantage the
"small-is-beautiful" RISC architectures might offer. At the time, we
knew in our hearts that RISC wasn't fundamentally better...which you
chose (RISC vs. CISC) was a simple marketing decision...not
a technical conclusion!
B.D. Lightner -
Thu Jul 25 23:37:38 PDT 1996
Zen and the Struggle Between Communism and CISC
Val Popescu and Bruce D. Lightner
Metaflow Technologies, Inc.
La Jolla, California
May 2, 1991
Copyright ©1991 by Metaflow Technologies, Inc. All rights reserved.
RISC (and Communism) Will Rule the World!
The following is a quotation from SPARC International's monthly newsletter.
SPARC International is a non-profit organization founded to promote
Sun Microsystem's RISC architecture.
RISC is just a name that has been chosen for an architecture that has
deliberately cut all ties to the past (i.e., software bases) and has
implemented the best instruction set possible in light of our more
advanced understanding of compiler theory, program behavior, and the
shifting performance boundaries between a computational engine and its
memory subsystem. Therefore, almost by definition RISC is the future
of computing. Over the next five years, a totally new computing
standard will emerge based upon RISC architectures. Reduced
instruction set computers will completely redefine the computer
industry's existing price/performance curve that is based on
complex instruction set computers and become the industrial computing
standard that will lead us into the 21st century. [From
"RISC: The Future Defined" by Roger Ross, SPARC-Line, SPARC
International Monthly Newsletter, September 1990.]
``RISCing'' outrage by certain readers, we substitute
``communism'' for ``RISC'', and re-phrase the above quotation
Communism is just a name that has been chosen for a social and
economic theory that has deliberately cut all ties to the past (i.e.,
outdated capitalistic society) and has implemented the best economic
system possible in light of our more advanced understanding of
political theory, human behavior, and the shifting class boundaries
between the state and its workers. Therefore, almost by definition
communism is the future of global economics. Over the next five years,
a totally new economic order will emerge based upon communist
doctrines. The communist economic system will completely redefine the
world's industries' existing distribution of wealth and power that is
based on capitalist economic theories and become the industrial social
and economic standard that will lead us into the 21st century.
People living in communist countries were exposed to daily dosages of
similar propaganda. Instinctively, they rejected this as hogwash, and
in the end communism itself was rejected (almost) unanimously. The
following is a joke told in eastern Europe before the overthrow of the
At school, young Boris was reprimanded by his teacher because his
uniform was wrinkled. Boris' excuse was: ``Comrade teacher,
my father turned on the radio this morning and heard the usual
story about how in five years our country would produce more tractors
than America. He turned off the radio, turned on the TV and saw our
party chief telling us about how in ten years the whole world would be
communist, and peace would reign forever. My father switched off the
TV and forbid my mother from plugging in the iron. He was afraid it
might start barking communist slogans too!''
Everyone Makes Mistakes
Government statisticians can determine exactly how many people wear
size 10 shoes. The central government's economic planners can tell the
shoe industry exactly how many size 10 shoes to manufacture. But, as
the communists discovered, the people cannot be depended upon to
necessarily buy all the size 10 shoes. Perhaps the communists needed a
``shoe police'' to arrest people for wearing the wrong shoes!
Although communism's problems seem evident today, it wasn't always
clear what was wrong with communist economic theory. Before the World
War II, a few brilliant minds in England (at Oxford and
Cambridge) were honestly embracing communism---in addition
to many ``have-nots'' seizing the opportunity to loot the ``haves''
elsewhere in Europe. In the end, communism failed not because it was
intellectually flawed---perhaps it was---but because it failed to
deliver prosperity and happiness to the masses. And it failed not
because it lacked clever leaders or tactical astuteness, but because it
deliberately ignored the realities of the marketplace.
It is hard to resist the temptation to draw parallels between the
history of communism and the present day RISC versus CISC debate. RISC
was masterminded at Berkeley and Stanford, two institutions of higher
learning similar in stature to Oxford and Cambridge. RISC
architects such as John Hennessy, David Patterson, and others are
certainly brilliant, convincing RISC advocates who have enlisted many
followers in their campaign to prove the righteousness of RISC.
Repeatedly they have intellectually overpowered their lesser
counterparts who represent the older, less glamorous (and historically
profitable) CISC school. The RISC promoters' cause was joined by
``have-not'' semiconductor companies that hoped to enter the
microprocessor business (LSI Logic, IDT, Cypress, etc.)---eager to
``loot'' a share of the business from the established microprocessor
vendors (Intel, Motorola, etc.).
At the same time, we have the technical tabloids, incessantly blasting
their readers with RISC propaganda pieces, with a style not unlike the
communist propaganda of the past few decades. Many of the tabloids
predict the total victory of RISC over CISC in a few years---at least
one RISC processor on every desk, peace and prosperity for all! We are
reminded of young Boris' story.
There are no obvious signs that RISC is going to bring prosperity and
happiness to all---nor does it have to! Companies that know how to run
their business are sure to prosper either way---RISC or not. What
ultimately will determine the outcome of all of this will be the paying
customer. The consumer will not buy the size 10 shoes specified by his
government unless there is a genuine need for them---or a law is passed
to mandate their consumption.
The microprocessor market seems to follow its own course, ignoring the
RISC-CISC rhetoric and zealots' predictions. Intel is steadily moving
up in the ranks of semiconductor companies by selling ``historically
dead'' and ``intellectually brain-damaged'' CISC microprocessors. At
the same time Sun is flooding the market with truckloads of RISC-based
SPARCstations. Meanwhile, RISC hasn't saved the likes of CDC
and Data General from an ever shrinking market share, nor did it save
Tektronix who got into---and back out of---the workstation business one
more time. Perhaps Tektronix and others need the help of the ``RISC
police'' to enforce cultural purity in the computer industry!
If this turns out to be a war between right and wrong, history
tells us that right will probably win. But which camp, RISC or CISC,
is on the right side? History has shown us that being young, smart and
energetic may not be enough to overcome the old and stagnant. Most
often what counts is economic might, not political righteousness. The
outcome of World War II was not decided by how brilliant the German
officers were, or by how technically clever their weapons makers were,
or by how hard their people worked, but ultimately by how many Liberty
ships and B-17 bombers the United States could produce each month---and
that was a simple function of available resources.
The huge installed base of 80x86-based PC's and the $15,000,000 DOS software
market is a hard reality. These customers provide a formidable revenue
stream to Intel---a sometimes reluctant promoter of CISC---unmatched by
any other microprocessor vendor.
The early RISC designs were simple (the legendary one semester project
at U.C. Berkeley). The early RISC processors performed well, when
compared to their contemporary CISC counterparts---given the right
benchmark programs. Using RISC design techniques, a company with
access to simple gate array technology could enter the microprocessor
business with minimal piggyback financing and a small (really small!)
design team supported by almost trivial design tools. One year later,
out pops a RISC computer---software left as an exercise to the buyer
(the C compiler being the second semester project at U.C. Berkeley).
For Intel to do the same thing, it takes a proverbial ``cast of
thousand'' engineers---and lots of blood, sweat, and tears. Intel's
``class project'' takes a full three years---and they don't even
produce a compiler (they reuse the old one)! In the process Intel
develops an in-house CAD/CAE system which allows the management of
million-transistor IC designs. This process is very expensive, but it
represents only a small part of Intel's gross margins on 80x86 parts.
Also, when it is time for the next generation 80x86 part, the CAD tools
and Intel's cast of thousand engineers are ready.
On the other hand, the ``small is beautiful'' RISC camp has just begun
to wean themselves from ASICs. Faster and cheaper RISC parts are now
needed. The world demands memory management units, bus controllers,
and cache controllers which run at the same speed as the simple RISC
integer units. Guess what? Once you include the necessary
support chips which allow the designer to assemble a complete
workstation, the total design solution looks equally
complex as the design of an 80486! The mythical one-year RISC design
cycle becomes three years---the same as competing CISC designs.
And finally, the new RISC designs seem to be getting just as complex as
the CISC designs they hope to displace. We hear rumors about
superscalar/superpipelined ``super-duper'' RISC designs from both
the MIPS and SPARC camps---which last week were preaching the benefits
of ``simple is beautiful'' microprocessor design. Next generation RISC
designs will be anything but simple. They will require design tools of
increased sophistication, and lots of engineers, and that costs
money---lots of money.
Transistors Aren't Free
To stay on the performance curve, would-be RISC microprocessor
companies need to make R&D investments which are comparable to Intel's
investment---in absolute dollars, not percent of total revenue.
The current volume of sales of RISC microprocessor chips cannot support
such a level of R&D investment. What is going to make up the
difference? In case of SPARC, Sun is paying the bill, funding SPARC
microprocessor development out of their profitable workstation
business---no doubt hoping that one day the SPARC microprocessor market
will generate enough revenue to fund its own ever increasing R&D
Will one of the RISC microprocessor vendors have access to enough R&D
dollars to produce a multi-million transistor microprocessor in the
same time frame as Intel? If so, and if RISC proves it has an
intrinsic performance advantage over CISC, the war is over---RISC
should have a commanding price/performance advantage over CISC in the
marketplace. However, if no RISC vendor can afford the skyrocketing
cost of designing multi-million transistor chips, then despite all the
advantages offered by clever RISC architects and designers, the B-17's
will have won the war.
Money Can Buy Happiness
The promoters of RISC have bombarded us with evidence that with
everything being equal (i.e., time, money, and silicon technology) a
RISC microprocessor design will always out-perform a CISC
microprocessor design. But today the magnitude of RISC-generated
dollars is not equal to that of CISC-generated dollars. If absolute
revenue dictates a company's budget for ``transistors designed'', the
fundamental performance differences which distinguish RISC and CISC
designs easily can be erased. Architectural solutions exist which
allow CISC designers to overcome most of the advantages RISC design
techniques may offer---given adequate transistor and design budgets.
The table below summarizes some of the design techniques which can
overcome CISC's well-known architectural deficiencies:
|CISC Architectural Deficiency||Architectural Compensation
|Small register set||Dynamic register renaming|
|Destructive register model||Generalized operand renaming|
|Complex, slow instruction decoding||Clever design and lots of transistors|
|Coupled memory references/ALU operations|
(compilers cannot schedule code)
|Parse CISC instructions into dynamically|
scheduled RISC-like parcels
And then there is the question of selling price. More transistors cost
more money, and since RISC designs require fewer transistors than CISC
designs for the same performance level, RISC should prevail. This is
a genuine competitive edge for RISC---but not all transistors are
created equal. A half-million transistor TMS34020 graphics chip from
Texas Instruments sells for just under $90, while a chip with
comparable functionality, yet half the transistor count, implemented as
a gate array by your favorite ASIC vendor will cost you four times as
much. The reason is both economy of scale (volume) and die size
(design budget). Either of these factors can easily wipe out RISC's
There Ain't No Such Thing As a Free Lunch
So far, the predictions of the RISC pioneers have not been fulfilled.
The time period between successive RISC generations is not
significantly different than the time period between successive 80x86
(CISC) generations, when measured at the socket level (i.e., when the
respective microprocessor ends up occupying a socket in a mass-market
computer). One cannot expect this to change much in the near future.
As for the promise of vastly improved performance from RISC design
techniques, the jury is still out. Today, competing RISC and CISC
designs show parity with respect to so-called ``integer'' benchmark
programs. The apparent advantage that RISC designs exhibit for
floating point intensive benchmarks may be more a reflection of CISC
(i.e., 80x86) market emphasis than fundamental architectural
advantage. If and when B-17's with floating point turbochargers are
needed, one can expect CISC vendors such as Intel to crank up the
So, what is the future of computing into the next century?
It's probably too early to tell. The future of the computing industry
certainly will be determined by companies like Sun, Compaq, HP, IBM,
and DEC, that know how to build and market computers that make real
customers happy---the RISC police not withstanding.