[Top] [All Lists]

Interesting overview of the market we're talking about - part 2

To: riscy@sword.eng.pyramid.com
Subject: Interesting overview of the market we're talking about - part 2
From: caret@pyramid.com (Neil Russell)
Date: Thu, 22 Jul 93 15:05:02 PDT
Reply-to: riscy@pyramid.com
Sender: owner-riscy@pyramid.com


The x86 interpreters (as comes standard with NT) have a serious
performance hit.  Binary translators have much less of a performance hit.
The basic idea is to dis-assemble a binary, translate it to the new
assembly language, optimize that, then assemble that.  Once completed the
software has been ported to the new architecture without needing the
source code.


DEC has 3 versions of binary translators.  One for translating from VAX to
Alpha, one for MIPS to Alpha, and one for x86 to Alpha.  It seems these
work on the vast majority of applications.  The performance hit is
minimal.  If you compile the SpecInt92 software for MIPS and then
translate it to Alpha it only runs 10% slower than if you had compiled it
for Alpha.  DEC demonstrated an x86 to Alpha binary translator at Spring
Comdex.  The Alpha has a very good chance of running x86 code faster than
the Pentium (probably even if optimized for the Pentium).  The new 300 Mhz
Alpha (December time frame) may give double or triple the Pentium
performance on 486 code (assuming translator is tuned by December).  At
least the first 2 translators DEC did were designed to be used by random
end-users, the x86 one probably is too.  DEC demo-ed their x86 to Alpha
translator at the spring Comdex (around 5/24/93) (info from post by Dileep
Bhandarkar bhandarkar@msbcs.enet.dec.com).  Even if it is not ready to
ship with their first release of NT (estimated to be a month after
Microsoft ships NT), DEC will be talking about it.  Also note that it does
not need to work on all programs - even if half the time it says "sorry,
use the emulator" it is still a win over no translator.  



Margins in the PC market are not very good (ask DELL and Apple).  If PC
vendors can make systems using a $70 MIPS chip instead of a $950 Pentium
you can bet they will want to (if not DELL at least vendors like Compaq
who still realize that computers are a fast moving field).  If PC vendors
are having trouble getting Pentium chips (and they seem to be) there is
all the more reason for them to make RISC systems.

Intel has also pissed off a number of PC vendors both by not being able to
ship and by threatening them over using Cyrix or AMD parts. 

Compaq had promised to develop a MIPS based PC.  There was an "ACE"
agreement between Microsoft, MIPS, and Compaq to develop a new Windows NT
system.  Compaq fired their president and changed plans.  They said they
were told the Pentium was going to be out soon and be as fast as the MIPS
chips.  They have hinted that they may be reconsidering the idea of RISC
based systems.


The June PC World reviewed 25 low cost 486 PCs and picked the top 5.  Two
of those were based on Cyrix chips - including the very top machine.  In
today's low cost PCs (like $1000 to $1,500) the difference of $100 in CPU
is significant.



At a CPU price of $70, MIPS/NEC must be planning on selling millions of
chips.  The development costs (design and fab setup) on this chip was most
probably over over $100 mil.  Intel is said to have spent about $500 mil
on the Pentium - not sure how reliable that was - but I do know Intel is
spending around $2 bil to set up 2 fabs.

The very reason Intel got where it did is that it used to second source
its chips, making their chips attractive to IBM.  Today MIPS,
DEC/Mitsubishi, and Motorola/IBM each have this attraction but Intel
itself does not (and IBM has started making their own chips) (though
AMD/Cyrix sort of act like second sources for Intel).

Other companies such as DEC, Motorola or HP may have $70 chips coming too.


In the past Unix has claimed to make software portable between different
machine times.  The claim was that you "just need to recompile".  However,
since each vendor had their own Unix implementation, this never really
worked.  Applications had to be converted to run on the different Unix
types.  The total Unix market is about 2% the number of machines as the PC
market. However this was sliced up to Sun, HP, DEC, IBM, SGI, etc,  each
needing a software port.  This meant a lot of work and support for each
small slice of the market.  Even so, there is a fair amount of software

Windows NT will come from Microsoft no matter what machine type it is to
run on.  Software will really "just recompile" for different machine
types.  The NT market is more or less expected to pass the size of the
Unix market within 12 months.  Porting an existing Windows application to
NT is said to be easy (trivial if it is a Win32 application).  Given how
easy it will be to support other machine types once an application works
on NT, software vendors will do so.   At the very least Microsoft will be
providing its software for x86/MIPS/DEC based machines, which is already
lots of software.

Users will probably be able to get "updates" for their existing x86
programs that come with MIPS/DEC binaries as well.  So if they have a $300
program and the upgrade cost, say $100, they can move it to a MIPS machine
for $100 (i.e. they don't need to pay full new price).  For most of their
applications binary translators will probably work, and for the rest the
emulator of windows NT is probably fast enough, but for the few
applications where performance really matters they can pay for an update.

Once the software takes off, there will be no question for a new user who
does not already own lots of x86 code.  If a user can buy the software he
wants for a machine that is about half the cost he will go that route.
Given that a the MIPS machine should be at least $1,000 cheaper (support 
is also cheaper since 40 Mhz external and the $70 MIPS CPU vs $950 Pentium) 
a user can justify the costs of a few software updates (they might even 
have been planning on getting updates anyway).


The first 486 systems, as shown in September 1989 Byte page 95, had prices
from $18,000 to $40,000.  Many of the computers advertised in that issue
were 286 systems.  The 386 systems had prices of $2,500 to $13,000.  So
the low end 486 was 7 times as expensive as the low end 386.  Given the
large price difference it makes sense that most people were not in a hurry
to switch to 486.

Today ALR (800) 257-1230 has a Pentium system for $2,500.  This is not really
that much above the price of even the cheap 468DX2/66 systems, about twice 
the cost.  Today almost nobody advertises 386 desktop system.  Given how much 
smaller the price differences are this time, a much large percentage of the 
people buying computers are going to want to buy ones with the new CPU.

Back when the 486 came out the prices were such that supply and demand met 
just fine.  However, today I think we have the makings of a massive 
Pentium shortage.


Apple is ahead of schedule and so will be shipping PowerPCs in time for
X-mass sales (i.e. not wait for superbowl in Jan to be exactly 10 years
after Mac announcement as has been plan).

IBM will be shipping PowerPCs in Sept.



Consumer and business confidence is down compared to end of last year so
people are probably ordering less PCs.  The backlog of orders has probably
been satisfied.  Thus we could see a "glut" of x86 CPUs.

Unit sales of i486 drop off a bit as people wait for Pentiums/NT/RISC.

People planning on upgrading to P24T (Pentium) just get cheap 486/33
systems (where Intel gets $200/CPU and not $500).  This is called "less
favorable product mix".

It may turn out to be true that most "Pentium-upgradable" systems are not
- see Electronic Buyers News, July 5th.  In that case people may not buy
so many 486 systems and just wait for Pentium systems.  This may not be a
big problem as a small fan mounted on the chip may solve it (if you can
get power for the fan).

Production of i486 down as some fab(s) move from 486 to Pentium.

IBM/Cyrix/AMD take another 10% of Intel's current 486 sales.

Prices of i486s drop.

MPP sales drop sharply (though only few % of Intel's sales it hurts

Higher capital expenditure as they rush to expand production capacity for
Pentiums without sacrificing current 486 production.


It takes around 3 years to get a fab line built and running.  They are
expensive (current ones are around $1 bil, though in the past they were
much cheaper).  Intel has only been really rich for the last 9 months.
They have not had time to take their really great profits and put them
into fabs.  As a result, their fabs are OK, but not really top notch.
Intel is currently using 0.8 micron, and switching to 0.6 next year.  IBM
(who had money 3 years ago and is only poor for the last 9 months or so)
already has 0.35 micron fabs.  NEC also has better fabs.  Better fabs mean
faster cheaper chips and higher volumes.


PowerPC 603 is supposed to be faster than the 601 and real cheap, like
$50.  Not sure exactly how long till it comes out.  MIPS has a $50 chip on
the way too.  DEC and HP are also coming out with low cost CPUs.  These
are probably all as powerful or more powerful than the Pentium.  With
binary translators it is not clear why anyone would pay more than $50 for
a CPU.  The current $950 price on a Pentium is not going to last, nor is
the $500 on 486.



(a look at some opposing opinions)


Maybe Intel has a top-secret GaAs 400 Mhz 486 that they are about to
announce.  MIPS has made a GaAs version of their architecture (not as a
product) and the 486 has few enough transistors that a GaAs version might
work.  However, the 400 Mhz part would be hard, and without a reasonable
sized on-chip cache (very doubtful they could swing that) the real
performance would not be too impressive.  Also, it would not be cheap.


Intel may get legal action to stop PC makers from shipping Microsoft
Windows or OS2 with systems using AMD/Cyrix/etc chips.  Intel is trying to
use the International Trade Commission to stop Twinhead (Taiwanese
company) from selling PCs with Windows pre-installed since Windows uses
the VM layout covered in some Intel patent.  This is bogus as Intel should
be suing Microsoft.  Microsoft sent a letter to the ITC saying they
thought that it was inappropriate for the ITC to take action on this as it
was not really a trade issue but a domestic patent dispute.  In the worst
case they ship the systems without Windows installed and Microsoft gives a
good price to people who show they have recently purchased a system from

RISC really works better with cheap DRAM.  Code can take nearly twice the
memory as in a CISC (you need more DRAM to rum Windows NT on a RISC
system, like 16 MB vs 12 MB).  By claiming "dumping", Intel got the
government to take action that has kept DRAM prices from dropping much
over the last 7 years (went up then is now a bit below - but nothing like
the factor of 2 every 1.5 years that there used to be).  However, as more
and more DRAMs are produced outside Japan and as more PCs are produced
outside the US, the US/Japanese government controls on DRAM will stop
working.  Note that PC makers outside the USA have a big competitive
advantage over USA based ones at this point, thanks to Intel.  Since DRAM
in the USA is really way overpriced at this point, it is more apt to take
a major drop than to go up or stay at these levels.  Still, at todays high
prices the extra 4 MB for the RISC system is about $100, which is not much
compared to the CPU price differences.

FLASH is an important emerging technology and Intel is leading in this
market.  Though still small, it could become a very large market very
soon.  One problem with this that battery backed up DRAM could be more
cost effective if DRAM drops like I think it will.  Also, Intel had
serious production problems and this has made it easier for others to
enter the market.  Also, Intel is in patent disputes with Rohm over FLASH.
Intel seems to be violating Rohm's patents and so may not have rights to
make FLASH chips. 

Intel could get legal action claiming that NEC's $70 R4200 CPUs is
"dumping" so that the government stops NEC from competing inside the USA.
However, there is a big market outside the US and MIPS can get other
companies inside the US to make the chip.

The danger of legal action is that the more Intel does the more chance
others have of getting an anti-trust "restraint of trade" suit against
Intel to stick.  Cyrix has had such a suit in progress since Dec 1990.
Not sure, but AMD is probably working this angle too.

Maybe the P6 is much better than the Pentium and can be out within a year.
The year part is the trick.  Also, it is probably twice as many
transistors, so yields will be even more of a problem.


<Prev in Thread] Current Thread [Next in Thread>
  • Interesting overview of the market we're talking about - part 2, Neil Russell <=