Article 708 of intel.general: Path: ornews.intel.com!carthago!chedley From: chedley@carthago.intel.com (Chedley Aouriri) Newsgroups: intel.general Subject: PC Microprocessor market Date: 28 Nov 1993 22:55:08 GMT Organization: Intel-Corp,_Hillsboro,_Oregon Lines: 2163 Distribution: world Message-ID: <2dba8c$43d@ornews.intel.com> Reply-To: chedley@carthago.intel.com (Chedley Aouriri) NNTP-Posting-Host: carthago.intel.com Here is pretty long but interesting article posted on a public network (Usenet). It could be good reading material for Marketing people at Intel. You may forward it to whom it may interest. -- ..CHEDLEY.. chedley@carthago.intel.com ------------------------------------------------------------------------ Standard Disclaim: The above statements and opinions are strictly mine, and do not represent any company or organization's position. ...................................................................... Article: 55141 of misc.invest Newsgroups: misc.invest,comp.arch >From: vac+@cs.cmu.edu (Vincent Cate) Subject: PC CPU Market (extra long) Organization: School of Computer Science, Carnegie Mellon Date: Sat, 27 Nov 93 01:59:56 PDT Xref: ornews.intel.com misc.invest:55141 comp.arch:45185 I have done some homework on the PC microprocessor market and in particular Intel's part in this market. I have collected a lot of notes and pointers to information. Think of this collection as "share-info" in the vein of "share-ware". If you find this collection interesting or useful please show your appreciation and send a donation to me at: Vincent Cate 6943 Penn Ave Pittsburgh PA, 15208 Anyone is welcome to copy or reposted this information anyplace as long as it is copied in its entirety (including this section). Copyright 1993 (c) Vincent Cate I am a "poor grad student" and so can not really invest as much as I would like with the information I have, but I have invested. Just to be safe, I have checked with the SEC and they say that posts of this form are ok as long as I do not own more than 5% of the company, am not an insider or releasing information from insiders, and that I am not making up fraudulent "info", etc. I think I am ok. If you make a ton of money using this info (or are prevented from loosing a ton of money) please remember me. :-) Also, there is a reasonable chance that at some time in the future I will collect some more information and send it first to people that have donated in the past. I should be graduating in the near future (PhD in CS). I am interested in analyzing high-tech investments. If you would like to hire a consultant for that kind of work, please contact me. I would really like to consult for venture capital firms or even work full time for one. If you can put me in contact with a VC who has work, I would appreciate it. DISCLAIMER: The following text is provided as-is with no warranty of any kind. Some of the information came from the net and is not verified. Some notes were jotted down after returning from a library and may not be correct. There is some outright speculation. This information might be considered un-balanced in some ways. This collection of notes was put together over a period of time and some of the info could be out of date. There are sure to be typos and other types of errors. The reader should check any information before using it. If you find an error, I would like to know about it. If it is major, also post to the net. vac@cs.cmu.edu ****************** FIRST MY PREDICTIONS/SPECULATION ************** IBM, Cyrix, TI, NexGen and AMD will all have chips equal to or better than Intel's 486DX2/66 within the next 3 months. By December these 5 combined may be shipping between 1/4 and 1/3 of the 486s (i.e. 1/2 to 1/3 as many as Intel). Because so many companies have the rights to Intel's patents, Intel will not be able to legally stop the chip clone makers. Cyrix, TI, SGS-Thomas, IBM, NexGen, UMC, and AMD will be ramping up as fast as they can. This hurts prices on Intel's 486 chips years before Intel is ready drop prices and move customers on to the Pentium. The existence of really fast RISC machines and the competitiveness of the PC market will keep Pentium system prices very low. The fact that Pentium systems do not cost much more than 486 systems will cause many people to order them instead of a 486. However, Intel can not produce nearly as many Pentiums. In 1994 MIPS/NEC/IDT and Motorola/IBM each sell millions of chips. The MIPS group will sell more than twice as many chips as Intel sells Pentiums (mostly as x86 upgrades). The Motorola group will just pass Pentium production (mostly due to Apple). DEC, HP, and SUN will all ship in numbers that are a sizable fractions of the number of Pentium shipped (DEC may exceed Pentiums). By the end of 1994 it will become clear that the Pentium will never get close to the kind of market dominance that the 486 has. People will complain that Intel is "Having production problems with the Pentiums and not able to supply enough" since the market will be wanting something faster than a 486. Intel will have less than 20% of the high end PC market (Pentium class and up) by the end of 1994. Acer and other companies making MIPS systems will have good sales (partly because people can not get Pentium boxes). By mid 1994 Compaq, Dell, and a few other PC makers will announce RISC based PCs. The R4200 portables will take off like gangbusters and dominate the portable market by the end of 1994. The unit sales of x86 portables and sales $ probably will still exceed that of the R4200 portables, but the total profits on R4200 portables has a good chance of exceeding the total for x86 portables. And in any case, the R4200 will be growing much faster. The Pentium, MIPS, DEC, HP, Motorola, IBM, Cyrix, AMD and NexGen chips, will cause demand for Intel's 486 chips to drop with prices following. Note that in high tech just having some competing product announced can cut into sales of current products (announced by competitors or the same company - for example, Osborne Computers killed itself by announcing a future product too soon). Intel's bread and butter starts to evaporate. Following DECs example, other companies will come out with x86 to RISC binary translators. Over the next year I expect translators to go from x86 to each of the 5 major RISC architectures using NT (Alpha, MIPS, PowerPC, HP-PA, Sparc). A year from now each of these companies (except maybe Sun :-)) will have systems that, after translation, run 486 code faster than Intel systems. Once the RISC systems have their foot in the PC market software comes out compiled for them. At this point it is all over but the shouting. Intel is nearly a one-product company (seems that around 80% of sales are x86 or support chips and that this is 90+% of profits) so if 486 prices drop off they are in big trouble very quickly. They are more like Apple or DELL than they are like IBM or DEC (who are much more diversified) so as the prices drop on 486s Intel's profits will be hit fast. (Note those 4 companies had problems with profits even though their respective market shares were not hurt too much). One year from now, while still selling about as many chips, Intel will be losing money. In the long run Intel will survive (as will IBM etc) but in the near term (next couple years) things could get real rough. It seems doubtful that they will ever see the recent high profit margins again. **************************************************************************** HISTORY: IBM decided to use Intel CPUs because they believed that between Intel and the other companies that were second sourcing Intel's chips (like AMD) there would be plenty of supply and the prices would be low. Just in case, IBM got the rights to make Intel compatible CPUs if it ever wanted to. Up through the 286 Intel had second sources for their chips. The chips were cheap and plentiful. But the 286 was the last chip where Intel did this. Intel decided to go back on its second source policy (something AMD claims breaks at least one contract) and bet the company that customers were locked in enough that they would have to keep buying Intel CPUs. For 5 years only Intel made 386 chips. However in the 2 years after that Intel lost almost all of the 386 market. For 4 years only Intel made 486 chips. There have been various supply problems (like an order in November 1992 for 486 66Mhz chips might have taken more than 3 months to ship). Also prices have been rather high. Now IBM, AMD, TI, and Cyrix all make 486 chips. It looks like it will be less than a year after the Pentium came out before another company has x86 performance in that range. GENERAL PROBLEM: Intel has had such high profit margins lately that many companies are attempting to get in on the action in one way or another. AMD, TI, HP, SGS-Thomas, and other companies all have the right to use all of Intel's patents. This means Intel has almost no way to protect its intellectual property against chip clone makers. It can not keep others from making x86 compatible chips. AMD only got into trouble for copying the microcode which has a copyright on it (though because of the AMD/Intel second source agreement it is still not clear if AMD will be in big trouble or just have a lot of court fees). For the past 7 years Intel has always been at least 4 years ahead of its competition as far as having the highest performance chip that runs Microsoft applications (via DOS/Windows). Intel does not have any lead as far as Windows NT software (both MIPS and DEC systems are faster) and not much of a lead in the x86 compatible CPUs. IBM, Cyrix, NexGen, and AMD should all have chips as fast as the current Pentium in less than 1 year from when the Pentium came out. Worse yet, Cyrix and/or NexGen may actually have the lead in x86 performance during parts of 1994. This means Intel can not get really high prices for its fastest chips. MIPS/NEC has an R4200 that is about the power of the Pentium at about 1/10th the cost and 1/10th the electricity use (fantastic for portables). Binary translators will make it possible to convert software from x86 to other CPU types with little performance hit (as opposed to the current x86 interpreters). Several translators exist (DEC / Echo Logic) and and I expect others will also have translators within 6 months. Systems houses such as IBM, DEC, HP, SGI, SUN are all making their own chips these days. They all realize that, with todays technology, the best price/performance computers use microprocessors. A microprocessor is no longer a thing that only a few places like Intel make. If Intel's chip prices go down, even if unit volumes stay just as high, their profits will really be hurt (in that case a dollar less in sales price is a dollar less in profits). AGING COMPUTER ARCHITECTURES: A computer architecture that starts out as being a very reasonable can end up 15 years later not being so reasonable. The DEC VAX and IBM mainframes are some examples of this. Shorter term ones are Sun's Sparc (register windows turned out to be a really bad idea). Intel's x86 architecture has a couple of real problems that are starting to hurt. There are only 8 32 bit registers. Now 8 registers was fine for the 8080 more than 15 years ago, but now 32 general registers is more reasonable. Back then it was necessary to have very few to get everything on one chip, but not today. The R4000, for example, has 64 64 bit registers (32 general and 32 floating point). If you have many places to store things it makes it much easier to juggle several things at once (multiple functional units and pipelining really mean juggling data). The x86 floating point architecture is stack oriented which means that it is very hard to do several things at once (top of stack is a terrible bottleneck). There is good reason to believe that x86 is due to go the way of the Z80, VAX, and IBM mainframe over the next couple years (i.e. not dead but not very profitable). SOFTWARE IS MUCH MORE PORTABLE TODAY Some time ago all PCs had to be "clones" of IBM machines. Many DOS applications were written expecting all sorts of hardware to be in certain places and work in certain ways. However, IBM lost the lead in hardware and now software can not depend on such tight compatibility in the hardware. So most PC applications these days conform to the Windows API. Such code is far easier for an emulator or translator to deal with. In the past only Intel's chips ran MS-DOS or Windows, or the thousands of applications that run on these. However, recently there are number of ways that people can run Windows/DOS applications on non-Intel systems. PORTING WINDOWS SOURCE TO UNIX Now that programmers have been writing to a well defined API, people have been able to make environments to compile for the windows API on non- Intel systems. This makes it really easy for developers to port their application to non-Intel systems. MainSoft Corp 415-966-0605 Has a Unix environment that provides Microsoft Windows API. Works at low level (xlib) and is claimed to be very fast. Sun seems to be working on using their WABI tech to provide a Windows API on Unix. This is a very reasonable move on their part. Bristol Technology Inc (203) 438-6969 has something for porting Windows source code to Motif. It seems to take some minor code changes. See April 5 Infoworld, p 16 or March 29 PC Week page 49. Also FTP to bristol.com or email to jean@bristol.com Borland's ObjectWindows for AppWare is supposed to let you write programs that work on Unix, DOS, Windows, and NT. BINARY INTERPRETERS/EMULATORS/DYNAMIC-TRANSLATORS/SYNTHETIC-CPUS Windows NT comes with something that lets any chip interpret Intel instructions. This interpreting hurts performance but a MIPS or DEC chip can interpret x86 code at 486 speeds (it really translates code the first time through loops). NT already works on DEC's alpha and the SGI/MIPS chips and an early version of NT was demoed on PowerPC at November 93 COMDEX (my guess is it is that it is nearly ready for developers and will March or after for end users). Porting efforts have been announced for the Sparc. HP also seems to be planning on supporting NT on their chips. NT's x86 dynamic translation is really just Insignia's stuff. Sun (800) USE-SUN-X has a system called "WABI" that lets Windows programs run on Unix systems with X. It is included free with Solaris. They will be working with Novell so it runs on machines using Novell/USL's Unix as well as Sun's. Also HP and IBM will be selling WABI for their machines by the end of 1993. A very nice thing about WABI is that since it uses X to do the display functions, you can take an existing 16-bit-windows program and run the core part on one machine (say a fast server) and the display on an X-terminal or other cheap machine on the network. This makes servers and X-terminals work even for old applications. Sun will support Win32 early in 1994. Praxsys Technologies Inc - Cambridge Mass - In Sept 1992 they sold to Sun the basis for WABI. IXI (44-223-236555 - Cambridge England) has "Win-tif" which seems to let you use PCs as X-terminals with a Windows look and feel. So you can just run programs on the Unix server from a slow cheap PC you already have. They have a deal with SCO (800) SCO-UNIX, so SCO should get this out fast. Also, SCO will be coming out with Sun's WABI spring 1994. Some Berkeley folks have created "WINE WABI" which will be a sort of public domain WABI. It sounds like it may be mid 94 before it is available though. Not clear. Might show up via FTP from agate.berkeley.edu. Insignia Solutions (415) 694-7600 something called "SoftPC" that lets RISC chips emulate 486s. This has been out for Unix machines for some time now. The problem with it is it was emulating everything even when inside DOS or Windows code - which is slow. Insignia's stuff has been modified and included with NT - the modification was to "thunk" 16-bit windows calls into 32 bit calls for NT. So now it is not emulating while inside Windows. Insignia has a licensing agreement with Microsoft that lets them use the Windows source code. This (which is much like WABI) they call "SoftWindows". First shipping on NeXT in Nov 1993. HP-UX and Solaris should be shortly after that. This will be very fast as the Windows code is then native. They will support Unix and Mac (PowerPC). Alpha version running on 66 Mhz PowerPC was said to be as fast as a 25 Mhz 486 with an estimate of 33 Mhz 486 speed by release. A number of RISC chips are about as fast or faster than the 66 Mhz PowerPC 601 and less expensive than a 33 Mhz 486 (for example, PowerPC 603, R4600, R4200) so even in emulation they can beat a 486 on price/performance. Claim is that Apple and Insignia both expect the PowerPC 604 to be able to run 16 bit x86 windows programs as fast as the current Pentium (I think 604 is second half of 1994). Apple has software to emulate a Mac on a PowerPC system. Mac applications spend a very high percentage of their time in the system, so as long as that is fast emulation is ok. Still, Echo Logic's binary translator (see below) is probably the way to go Andataco (was Xcelerated Systems Inc ?) - "Liken" - Emulates Mac/68000 in a Unix X window. Need to load a copy of Mac-OS onto the system. This seems to violate Apple's license agreements. Not clear how the legality is all going to work out. I have also seen that they replace the toolbox. Said to be very good and rather fast. Exists at least for Sun and IBM AIX (RS/6000 or PowerPC). Will work on Alpha in Q2 1994. Quorum (415) 323-3111 has both an emulator and a translator. Both are for Mac applications. It is worth noting that major PC applications are also on Mac, so being able to run Mac software may be enough. Acorn Computers has an 80186 emulator for their ARM-based computers. It simulates everything, including OS and graphics (CGA, EGA+ or VGA), so it is quite slow. Said to run almost all MS/DOS applications that do not require 386 upwards. MIMIX simulator. FTPable from ftp.cs.vu.nl in minix/simulator is a public domain PC simulator. It has been tested on Suns but is only C code (no assembly at all) so it should port fairly easily. Said to not be too fast. IBM researchers prototyped a fast emulator for a subset of the System/370 hosted on a IBM RT. They used dynamic translation. IBM/ Cathy May, "MIMIC: A Fast System/370 Simulator", SIGPLAN 1987, Symposium on Interpreters and Interpretive Techniques, June 1987. Purple Mountain Computer Products makes an Atari ST emulator called "GEMulator". This runs on top of DOS and emulates a 68000 to run Atari programs. Said to take a 33MHz 486 to get 8MHz 68000 speeds. Larus' SPIM is the emulator used in the most recent Hennessey & Patterson computer architecture book. It emulates the MIPS R[23]000 instruction set on several OSs and has a spiffy X interface as well as a more pedestrian command line one. Runs at about 1/25 real time. "Trash" is another MIPS emulator. Runs at about 1/10th the speed. Translates to an intermediate form. It uses the threaded code interpreter techniques described in: "Some Efficient Architecture Simulation Techniques" Robert C. Bedichek, Department of Computer Science, FR-35 University of Washington, Seattle, Washington 98195 robertb@cs.washington.edu pp.53-63, Proceedings of Winter 1990 USENIX, Washington, D.C., January 22-26, 1990. David Keppel of University of Washington is doing work on "On The Fly" code generation. Franklin Software 408 269-8051 makes emulators for the 8051 series of microcontrollers that run on x86. The simulator comes with their compiler package. At 408 269-8051. I remember long ago of hearing of software that let a 68000 run CPM/8080 programs. It was said to run about as fast as an 8080. One common trick is to have the systems software recompiled for the new architecture (especially routines for graphics) so that calls to the operating system are running at native speed. In windowing software it is common to spend about half the time in the operating system, so getting the systems software in native mode really helps. Most of these "emulators" are really "dynamic-translators" that translate a basic block the first time through it. This is much faster than interpreting each instruction since subsequent times through the loop you are running native code. However, it is really hard to do optimization of a program when you are translating and running one block at a time. It may be that after running awhile they can optimize things a bit. For example, if one loop calls a small procedure N times then it might optimize those 2 blocks together. This sort of on the fly optimization would be really interesting. Guess it should be doable, just take some time to develop this. It would be be a natural extension for a system that does dynamic translation to save the translated code. If this were optimized it would be about as fast as the binary translators mentioned elsewhere. However, if this was done using trace scheduling techniques it could be even faster than a source code port. Marketing types could say, "The translator watches how you use the program and then optimizes it just for your usage pattern." As the dynamic-translators get better, their performance could approach, or even exceed, that of a source code port or a one-time-only binary translator (i.e. DEC's translator can be run again after it has done some tracing). z80 emulator on DOS (and several others) FTP pc.usl.edu:msdos/cpu.emulators DOS EMULATORS RUNNING ON x86 UNIX The hard part of emulation is providing the environment. There are a number of systems that do these while still running on x86 machines. "DOS as a Mach 3.0 Application", FTPable from mach.cs.cmu.edu, doc/published/dospaper.ps. This worked on Mach or OSF. Dosemu is a public domain x86 emulator that runs on Linux. FTP to tsx-11.mit.edu:pub/linux/ALPHA/dosemu for a copy or more info. Has emulated VGA, and also EMS and XMS memory. Claim is that most programs running in real mode 386 (and of course x86 in general) run pretty smoothly. Speed is bit slow but not bad. Locus Computing has something called "Locus Merge" or "Merge DOS" or "Merge 386" that runs x86 Windows and DOS programs on Unix machines. SCO has sold this as "SCO Merge". Not clear at the moment if this works on non-x86 Unix machines. Still, the hard part is emulating the environment. In any case it is probably not be as good as Sun's WABI because SCO is working to bring out WABI. PROFITABLE MONOPOLY PRICED PRODUCT LINES: Current estimates are that Intel is shipping about 2.5 million 486 chips per month and that that is where it is getting something like 60% of its sales and 80% of its profits. The Pentium production is probably around 50 thousand a month for the rest of this year. So they are making about 1/50th as many Pentiums. The Pentium is about $900. The top end 486 was about $500 but has just dropped to about $400. The average 486 sold now is probably about $150. The 486 costs about $30 to make. The Pentium costs were estimated at about $400. Profit on a Pentium is about $500/chip. If prices on 486s drop 30% then Intel needs to increase Pentium production and sales by 2,500,000*0.30*150/500 = 225,000 additional Pentiums/month. On a humorous note, it has been said that making 486s is as profitable as printing money. The logic was that it might cost $30 to make and for almost 20 times that much, and that printing a $1 bil was said to cost $0.05 or 1/20th of $1. Anyway, it is clearly very profitable and Intel's profit margins have been "the highest in the industry". Andy Grove (Intel CEO) even said that he thought they had gotten a little high. PEOPLE SWITCHING FROM 486 TO PENTIUM COSTS INTEL: The 486 is far more profitable for Intel to produce than the Pentium. The Pentium is about 3.5 times the size of the 486. This means that you get much lower yields. See Aug 2 issue of Microprocessor report. They estimate 181 486s on a wafer with 54% good (other sources indicate much higher yields). They also say 40 Pentiums with 9% yield. This means 98 good 486s per wafer vs 3.6 good Pentiums. This is 27 times as many good chips. They estimate 486 finished costs at $23 vs Pentium at $483. Even if profits were 5 times as much for each Pentium (say $500 vs $100) Intel would have about 27/5 = 5.4 times as much profit from a fab line making 486 chips as one making Pentiums. From someone on the net quoting Economist article from around Nov 15th: > "only 47 Pentiums can be carved from a standard six-inch wafer, from which >187 '486s could be made." And for yield, "a single wafer might produce 84 >of the [working] 486s; the same wafer would yield on average only 5.5 Pentiums. >... At $385 a chip, a wafer of the best 486s offers an income of $32,340; >at $985 each, a wafer of Pentiums might make only $5,418." Intel marketing has kept the price of the Pentium kind of low (compared to where supply and demand would have it) in order not to look too bad when compared to RISC chips. However, this means that the Pentium cuts into demand for 486 chips. Also, since the price is sort of artificially low there could be a shortage (i.e. a long wait to get Pentium chips). So the RISC competition at the high end is already causing Intel problems. COMPATIBLE CHIPS: While in the past Intel had 4 or 5 years till there were compatible chips of equal performance it seems their lead may now be 5 to 10 months. What is worse, they have a good chance of loosing the lead early next year (my guess is to Cyrix, though NexGen and IBM are possibilities). See "Multivendor 386/486 Market Burgeoning" in January 25 1993 issue of Microprocessor Report. Also see Byte December 1992, page 116 "Lab Tests: Does Brand Matter?". Cyrix, IBM, and AMD are now making 486 chips. AMD is producing 486 chips using a 0.7 micron process and though currently their fastest chips are 50 Mhz, they should not have trouble reaching Intel's top of the line 486 at 66 Mhz. IBM has a license agreement with Intel that only lets them sell 486 chips if they are on a board. IBM is now making chips and selling them on boards. The August's PC Magazine has some information about IBM's 486SLC2-66 Mhz motherboard. Destiny Computers 510-783-2727 has an IBM SLC2/66 Mhz motherboard for $499. Alaris/Asia Source Inc also has this board (maybe cheaper) 510-226-8000, 713-782-8899, 214-238-8899, 404-246-9000. Just a 66 Mhz chip from Intel is $500. IBM (800) IBM-2-YOU sells systems using these chips. Windows Sources 12/93 seems to say that the 100 Mhz are out now (I have not seen the article). IBM has now come out with an SLC3/75. The above places probably also sell this chip. The IBM SLC3/75 Mhz is a tad faster than Intel's fastest 486. IBM is getting 75 Mhz with 3.3 volts. IBM does have really good fab lines. The fastest Intel has at 3.3 volts is 40 Mhz. (See the Microprocessor Report that came out the first week of Aug.) The 99 Mhz should be out within 3 months, and be well above Intel's 486. The rumor is IBM will have 120 Mhz by January. Cyrix has 486 chips that about match Intel's 50 Mhz chips. In Q1 1993 Cyrix is expected to have samples of their M1 chips, which Cyrix should be twice as powerful as the Pentium for 486 integer code (see May 31 EE Times page 10). This new chip will not be as optimized for floating point as the Pentium. For integer code optimized for the Pentium this chip should still be 50% faster. NexGen plans to have something as powerful as the Pentium and compatible with x86 soon (see July 19 EE Times, page 2). They are said to be sampling the chip at a major motherboard makers in Taiwan. The chip will be 0.5 micron and HP and Fugitsu will be making it (July 13 1992 page 4) (Note that Intel uses 0.8 micron and will have some 0.6 next year.) As I update this NexGen is still said to be sampling and it is now November. It is hard to say how long it will take to go to full production, and we don't know how fast the samples are or how good the performance will be. They are at (408) 435-0202 but won't say much. United Microelectronics Corp (UMC), Taiwan, is now making 486 chips. Taiwan buys something like 2/3rd of all 486 chips. UMC has been making lots of support chips for the motherboard industry, and now with their own 486 they can make all of the types of chips needed. It is not clear yet how fast they will be able to ramp up production volumes. There are several others and there is a good chance that there are some that have not announced. In the past (see Byte and Jan 25 Micro report) as compatible chips got up to some performance level Intel had to drastically drop the prices of any of its chips below that level. One interesting graph in the Byte article page 118 shows how the price of Intel's 486/25Mhz dropped from about $300 to $150 in the quarter that Cyrix started shipping its 486/33. NEW CPU ARCHITECTURES: For years the higher performance of RISC chips was irrelevant since they could not run the PC software. This is no longer true. Also, since RISC chips were only used in workstations and parallel processors there was no reason to design very low cost chips. This is no longer true either. MIPS and friends are going after Intel's market with a vengeance. There are at least 7 companies planning on making Windows NT PCs using MIPS chips. The existing chips are as fast to maybe 50% faster than the Pentium. They are shipping today (no 6 month waits). The R4400 which is about 50% faster sells for $690. It seems people are and will be waiting for a long time to get Pentium chips. There are also some amazing MIPS chips coming (see Microprocessor Report May 31 1993). The first is the R4200 (or "ICE") from NEC which is estimated to be only $70 by mid 1994. It started sampling at the end of Oct 1993 at a cost of only $80. This is really low power (1.5 watts vs Pentiums 16) and just under the performance of the Pentium. You can get info on this by calling 1-800 I-GO-MIPS and with a touch tone phone asking the machine to FAX you documents 205 and 206 (as of 7/19/93 theses numbers worked but if they are wrong just get the index and then the docs on R4200). Since MIPS runs NT these can run all of the Intel software. At $70 this is twice as fast as Intel's 486 which costs like $500 (Intel's bread and butter). These chips are smaller than the 486 and so very cheap to make (like 1/10th to 1/20th the cost of the Pentium). Motorola 1-800-845-moto has designed a PowerPC 601 that is a bit more powerful than the Pentium and about 1/2 the price. When Motorola announced the chip Intel's stock dropped about 10%. The PowerPC 603 is a bit more powerful and 1/4th the cost of the Pentium. Looks like it will be sampling around the end of the year. A 603 was demoed at Fall 93 COMDEX. Also, NT was demoed at COMDEX (alpha version, but still). DEC's first Alpha chip is twice as fast and at least 6 months ahead of Intel's Pentium. DEC is sampling a 21066 which comes with built in cache controller, memory controller, video accelerator, and PCI interface. This makes it easy to build systems that sell for under $3,000. DEC has announced 275 Mhz chips and is rumored to have had a few chips run at up to 320 Mhz. Their faster chips can plug into the same sockets. DEC's Alpha runs NT. HP is coming out with a low cost chip that has the right byte order to run NT. There are persistent rumors that HP is working on an NT port. Also forecast of very low cost HP-NT systems. BINARY TRANSLATORS: Intel internally has a binary translator to clean up 486 code so that it runs faster on a Pentium. The performance of this about splits the difference between running 486 code and recompiling for the Pentium. John Crawford mentioned it during a talk he gave. There is no indication that they will make this available. 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. HP has a binary translator to move applications from their old architecture to their new HP-PA architecture. Echo Logic (908) 946-1100 is now selling a tool called FlashPort which performs binary translation from 68000 (Macintosh) to PowerPC. See Aug 1993 Byte. This translator is targeted to developers and not end users. Claims translated code is only 5% slower than if the source code is ported. Demoed in Jan 1993 at MacWorld Expo in San Francisco (Jan 11 MacWeek 1993). They have an architectural neutral intermediate form and from that they have backends to generate PowerPC, MIPS or Sparc code. So far they only had Apple's permission to translate the Mac-OS for PowerPC. They are going to do Snake/HP-UX and RS6000/AIX. Sounds like they could do others (say DEC/MIPS on NT as well) as long as Apple went along. They are at least thinking of doing x86 stuff as well. DEC 1-800-Digital has 3 binary translators. One for translating from VAX to Alpha, one for MIPS to Alpha, and one for x86 to Alpha. Info on the first two can be found in "Binary translation", Communications of the ACM, Feb 1993. It seems these work on the vast majority of applications. Some friends have tried a bunch of programs and they all just translate fine, including frame-maker. The performance hit is now minimal (after improving over the last year+). The latest I have heard is that using the binary translator results in code about 5% slower than porting the source code (MIPS to Alpha). This translator learns from previous runs how it should optimize the code. If DEC can get this kind of efficiency for the x86 translator (or is even close to this good) then Alpha has a very good chance of running x86 code faster than the Pentium (probably even if optimized for the Pentium). High end Alpha's could run translated 486 code much faster than a Pentium. The new 275 Mhz chips plug into the same socket (and can divide the clock down to get the same external speed). You can be some PC-makers/PC-buyers will be willing to pay the extra for the high speed chips. At least the first 2 translators DEC did were designed to be used by random end-users, the x86 one probably is too. The only trouble is that it uses a lot of memory (when we translated frame-maker we were glad we had 64 MB). DEC demo-ed their x86 to Alpha translator at the Spring Comdex (around 5/24/93). DEC has said they will have more than 300 native Alpha/NT applications by the end of 1993. Note that as of the end of October Microsoft has 2 NT applications. It seems highly probable that developers are already using DEC's translator. Quorum (415) 323-3111 has a binary translator to take Mac programs and convert them to run on Unix boxes. It does not seem to be geared toward the end user as much as DEC's is. Microsoft is using Quorum to port a number of their applications to Unix systems and then distribute/support them. Quorum calls their product an "applications adapter". As of 7/93 they say they have only qualified their software for word and excel but plan to have the top 10 Mac applications tested by the end of the year. They are still developing their environment (still having problems with programs that do not stick to the toolkit in proper way). Since they have only been able to port a few applications in many months, it seems far far harder to use than Echo Logic which seems to be able to port applications in a few hours. ARDI (or Ardi) - Abacus Research and Development Inc (505) 766-9115. Currently translate Mac to some intermediate code but will translate to native code. So this sort of should be under "emulators" at the moment but will be a translator soon. Cost is under $100 from Ardi, Dustin Discount or PC Connection. Even running the intermediate code with their "synthetic cpu" software, they claim it is very fast. Was only working on machines running NEXTSTEP but should be out for DOS machines in Oct. There was a NextStep demo that could be FTPed file from ftp.cs.unm.edu in pub/ardi. They are doing an Alpha version and maybe a PowerPC version. Multiport. Hunter Systems used to translate DOS applications to work on Unix machines (XDOS). They were about to do a Windows version and IBM was going to give them funding and somehow things fell through. Probably a big mistake on IBM's part. Anyway, Hunter has come back to life as Multiport. Still seems to be doing x86 to Unix but don't know much about them (don't have a location or phone number even). At ASPLOS-V (Oct 1992) some people from Tandem delivered a paper wherein they described their object code translation technology used to rehost their operating system from a CISC stack based architecture to new R3000 based computer systems. They ported the operating system and applications (sounds like any application works just fine). It attempts to do static translation but falls back to interpreter if need be. Note that like the HP translator this was going from a stack based machine (derived from HP architecture) to a RISC. Thus the architectural mapping is not as simple as, say, x86 to RISC. This one is particularly interesting as reliability is everything for Tandem applications. If they trust the translator, it must be good. MIPS used a translator to go from R2000 to VAX code to test the MIPS compilers before the first R2000 chips were running. IEEE COMPCON March 1986. MIPS uses binary translation from MIPS to MIPS in their Pixie program to add performance monitoring code to the binary. Apple used a translator to port QuickDraw from M68000 to Am29000 for a coprocessor card. >Saari, M. "68000 BINARY code TRANSLATOR", 1987 FORML Conference >Proceedings: Forth Modification Laboratory. Ninth Asilomar FORML >Conference and euroFORML '87; Pacific Grove, CA, USA and Heilbronn, >West Germany; 27-29 Nov. 1987 and 18-20 Sept. 1987; San Jose, CA, USA; >Forth Interest Group; 463 pp.; 1988; pp. 48- Describes a Forth program which translates BINARY code for the 68000 family of processors into BINARY code for the SPARC CPU architecture A binary translator can do global optimizations that a compiler can not do. It can optimized in library routines. So it is possible for a program to run faster after binary translation than when compiled from source. Note that if used with NT, a binary translator need not work for all applications as those that do not work can be emulated. So a translator that works with 80% of the applications could still be a market winner. In order for a binary translator to work you need to be able to provide the same operating system interface/environment on the new architecture. This is the hardest part. Because of this, Quorum and Hunter have more to do than the others or what needs to be done for windows to Windows NT (especially if NT's 16 bit windows emulator has some hooks in it). ------------------------------------------------------------------------ PC MARKET IS VERY COMPETITIVE: 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 top 4 PC makers are IBM, Apple, Compaq, and NEC. Three of these are coming out with RISC lines. I think Olivetti, DEC, Acer, and HP are in the 9th to 15th range and also brining out RISC lines. All of them together means that about half of the top 15 PC makers have already announced plans to make RISC PCs. Also, Workstation makers such as SGI, Intergraph, Sun who have not really been in the PC market are now targeting that market. So the RISC attack is extensive, and really just starting Q4 1993. 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 in CPU price can be significant (maybe $100 or 10%). Clone makers seem to have no loyalty at all. They buy cards from whoever sells the cheapest. Looks like they are doing the same thing with CPUs. UPGRADING TO RISC (DAUGHTER CARDS / MOTHERBOARDS / DROP IN CARDS): Much of Intel's profits come from selling upgrade CPUs. PC makers usually have an upgrade socket and Intel sells chips directly to users (getting full retail price that way). So for many PCs Intel ends up selling a total of 2 CPUS. When people upgrade CPUs they can sell their old one. If someone has a 486DX/66 and they upgrade to a RISC chip they could sell their old CPU for around $300. If a RISC upgrade card cost $600 and they can sell their old CPU for $300, then it only costs them $300 to get RISC/Pentium level performance. This is a fantastic deal. Since there are going to be $70 RISC chips I don't think the $600 for a daughter card that plugs into an overdrive socket is too low. Chip Store 800-820-6009 offered to buy a used 486DX/33 from me for $125. I suspect 486DX/66 is about $300 but did not really check. Anyway, if lots of people start upgrading from 486 to RISC then these used 486s coming onto the market will help prices go down. Imagine the following. Say one of the RISC makers creates a chip that can plug into the "overdrive" socket on a 486 system. Say for $500 you get the chip, some ROMS, and a copy of NT. You end up with a system that gives you the power of a Pentium for only $500. There is something on the order of 100 million systems with this upgrade socket. I really think a RISC upgrade would sell. Failing a chip with compatible pinout, a chip on a small card would not be too bad (either plugging into a 486 socket, an overdrive socket, a local bus, EISA bus, etc). I don't know if this is possible, but it may be possible to plug a RISC chip into the overdrive (with adaptor card) but still use the original 486 when you wanted to run x86 code. It would be ok if you could not use them at the same time, as long as you could switch back and forth. With present emulators a 66 Mhz 486 is probably faster than most PC priced RISC chips emulating an x86. However, in the long run you probably want to just trade in the 486. The installed user base is more sensitive to the price difference between a Pentium and a RISC chip. Since they only need to get a motherboard or processor card the price difference is drastic. Motherboard Warehouse (800) 486-9975 sells a Pentium motherboard for $688 plus $1672 for the CPU for a total of $2360. A RISC motherboard or CPU card will be $600 - $1000 or 1/2 to 1/4 this this amount. There is a very large installed base. Many of these people don't need a new screen, keyboard, or disk drive. They will be looking just at the prices for processor upgrades. Intel is far more expensive. If you subtract the trade in value of a user's old 486 you are then comparing $300 - $700 against $2060. Pentium looses big time. People who do not already have a machine might as well get a RISC machine since they also do not have existing software (assuming there is plenty of software available on NT). If used 486s go from $300 to $150 in 1 year, then that is about $0.50/day. This could be part of an ad campaign. Gives people some urgency, which should get them to act. Not exactly real though since at least that much should be coming off the RISC board every day. DAUGHTER CARD UPGRADES: A couple of companies have been thinking about upgrade cards. Tangent says they are thinking about it. DeskStation says some of the people buying their chipsets (which interface Rx000 to 486 bus) are thinking about it. From: shwake@nearside.UUCP (Raymond Shwake) > This might go into the category of "I heard", but one of our vendors >(they've built and sold PC clones to us) told us today that they've run >the Power PC chip on their own boxes using the CPU/expansion socket. Claims >it sits in an adapter socket which accommodates the different pin layout. >Claims it's damn fast too. DEC's PCI based 486 and Pentium systems can be upgraded to Alpha. I think it would be easy to plug a card into a "local bus" and remove the 486 as all of the signals on the 486 are on the local bus (as I understand it). MOTHERBOARD UPGRADES: Motherboards do not seem to be very expensive to make. Many places sell motherboards without CPUs for under $150. If RISC motherboards are triple that and then add CPU it is still cheap. Motherboard Warehouse (800) 486-9975 MIPS and PowerPC motherboards Should be out by the end of 1993 (Advertised in Sept then stopped). Sounds like $700 - 1,000. DeskStation (913) 599-1900 has a machine that uses an R4000 and Intel compatible peripherals called the "Evolution". They plan on making motherboards so that people can upgrade their own PCs. The hold up is that Windows NT does not yet have RISC device drivers for a wide enough range of PC devices. As this changes it will become reasonable for someone with an existing PC to buy just a motherboard and NT to move to RISC. Shablamm (408) 727-2210 MIPS Motherboard. Looks like it should be very low cost as it does not have a cache. Reply will have a PowerPC 603 motherboard upgrade for Macs. UPGRADE CARDS THAT PLUG INTO A STANDARD PC BUS YARC Systems 1-800-ask-yarc is selling a card to upgrade PCs to PowerPC. About the end of Sept they will also be selling a card to upgrade Macs. The current prices (around $5,000) are developer prices. There will be end user cards for under $2,000. The current cards (Aug) are 50 Mhz but the CPUs are 66 Mhz - only a phase lock loop chip is limited to 50 Mhz. In about a month you will be able to replace that chip and run the card at 66 Mhz. Other vendors could make PPC upgrade cards too. CHIPSETS FOR RISC UPGRADES DeskStation has a chipset for MIPS chips that handles a cache and also connects to 486 bus. The cache gives you good performance and the 486 bus means you can plug into a "local bus", an "overdrive socket", or a 486 socket. Nice. See microprocessor report around Nov 8. Someone has a Tigershark chipset (see microprocessor report around Nov 8) (could be Toshiba). This interfaces MIPS to 486. However, it does not have a cache between the 486 bus and the CPU so performance will be hurt. DEC's 21066 has a PCI bus. There are PCI to EISA etc chipsets. This means it should not be hard to make a Alpha upgrade card. IBM has a PCI chipset for the PowerPC. Again, from PCI you can get to normal PC buses. PORTABLES: In portables, power is everything. The 33 Mhz Intel 486 chips are showing up in portables. The reason why Intel 66 Mhz ones are not is that they use 8 watts. There are a few very small PC makers with 66 Mhz 486s but they are not really meant to run on batteries, the problem is that that one chip uses 8 watts. There are a number of 50 Mhz 486 systems using Cyrix or IBM chips which use less power than the normal Intel 486s. Intel has recently modified its chips to take Cyrix approach to saving power. In portables they compare things like MIPS/watt. Since Cyrix and IBM have had an advantage in performance per watt they have been doing well in portables. The new NEC/MIPS chip has twice the performance of a 66 Mhz 486 and is only 1.5 watts, so its MIPS/watt type rating is about 10 times that of Intel's 486. This will really kill Intel in portables. PDA/PEN BASED MARKETS: In the PDA and Pen-based markets there is not much benefit to running MS-DOS so there is not much need to use an Intel chip. The big winners in this area seem to be AT&T's Hobbit and the ARM processor - though many other chips are now targeting this market. These are cheap low power chips that are very fast. EO is using the Hobbit. Dell will be using the Hobbit. A pen-based computer that also has a keyboard (usually called a "convertible") does have some advantage to using MS-DOS and so x86. These are selling. WHAT IS COMING OUT OF INTEL NEXT (RUMORS): See July 12 EE Times page 4. A new version of the Pentium Processor, code named the P54C is expected to start shipping in 1st or 2nd quarter of 1994, with systems showing up late 2nd to 3rd quarter of 1994. The processor will be built on a 0.6 micro process vs. the current 0.8 micron process. The smaller geometry will reduce power consumption, allow the manufacturing of 3.3 V parts (at current speeds?), and production of 5 V parts running at internally at 100 MHz and externally at speeds of 33MHZ or 50MHz. More recently the speed numbers have been 90 Mhz (which fits with their 60 Mhz instead of 66 Mhz) then I have been hearing 80 Mhz. If it is 80 Mhz with 40 Mhz external, it might be slower than a 66 Mhz chip. This would not look good at, or after, fall COMDEX. The P24T is a 32 bit bus version of the Pentium. With a narrower bus you get reduced bandwidth and that should hurt performance. This is the same type of trouble that the Pentium 80/40 will have. Also Intel should be coming out with 100Mhz 486s. Most of the 486 sockets in PCs today are 33 Mhz. So Intel has to make this a DX3. This also means restricted bus bandwidth. To make this work well, they need a write back cache, which the existing 486 sockets do not support, but the overdrive socket seems to. But a write back cache means a fully new chip. Also, there are rumors that it is 3.3 volt. At 5 volts it might be too hot for most PCs. However, 3.3 volt would would make it incompatible with most existing 486s - also it is harder to hit 100 Mhz at 3.3 v (as of about Oct Intel's fastest 3.3 volt 486 was only 40 Mhz). They probably need to have 3.3 internal and 5 volts at the pins (doing their own power conversion on chip). Anyway, the point is this is not as trivial as going from 33 Mhz to 66 Mhz. If there is a 3 month lead time on Pentiums, then you (PC-buyer or PC-seller) really need to compare them to the other CPUs that you could buy 3 months from now when deciding if you want to order one. This means you have to compare current Pentiums to low cost Alpha's with built in PCI, cache controller, dram controller, graphics-accelerator, and 70 SpecInt92 in Q1 1994, chips and DEC's binary translator. You also need to look at the MIPS R4600 and MIPS R4200. So long lead times on the Pentium really hurt. Long lead times in fast moving fields are just deadly. There could be more than 6 month lead time on getting Pentiums from Intel at this point. Though end users are seeing less than 3 months. Fundamentally the problem Intel has gotten into is that the price of the Pentium (like $850 for 60 Mhz) is less than a supply/demand curves would have it be - so there is a shortage. Given that the Power PC had been announced at $375 and MIPS/DEC chips were more powerful Intel would look really bad if they charged more than $1,000/chip, but they probably could still sell all they could make (at least for the next few months). Cyrix, AMD, IBM, MIPS, HP, DEC, NexGen, etc are not going to under price their parts so much and so will NOT have similar shortages (some people think that shortages are fundamental to x86 architecture, which is silly). LOW COST RISC CHIPS: At a CPU price of $70 for the R4200, 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/NEC/IDT, 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/TI/SGS-Thomas sort of act like second sources for Intel, but with Intel lawyers always trying to stop them). Other companies such as DEC, Motorola and HP all have low cost chips coming too (Aug 9 1993 Hot Chips talked about MIPS, HP and DEC). PowerPC 603 is about the same performance as the PowerPC 601, and about the same as the Pentium - but will be 1/4th the price of the current Pentium. It should be sampling around the end of the year. Since binary translators will make software available it is not clear why anyone would pay 4 to 10 times as much for an Intel CPU. BUT UNIX CLAIMED SOFTWARE WAS PORTABLE TOO: In the past Unix has claimed to make software portable between different machine types. 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 in 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 available. 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/PowerPC based machines, which is already lots of software. Any user who buys NT applications gets versions for all different machine types - meaning that at any future date he could easily move to another machine type. The longer NT is out the more people there will be that can easily move to RISC. 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). Customers pay about the same price/performance for any Unix machine. It is amazingly close really on a graph. This implies that even though Unix machines are not really interchangeable, these are close enough that the prices act as if they were. BUT PEOPLE TOOK YEARS LEAVE 386 for 486: 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 does not include a monitor or harddrive. This price 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. Gateway (800) 846-2000 has a $2,995 Pentium system that seems loaded with features (includes disk, monitor, CD-ROM, software, etc). Page 62 of 11/15 Computerworld has a full page ad from Ambra (IBM) for a Pentium, PCI, 256 KB cache, PCI graphics accelerator, 8 MB ram, 340 MB disk, and 14" SVGA color monitor system for $2,799. At these kinds of prices it is not clear why anyone would buy a high end 486 these days. Sales should be dropping off. 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 and surplus of 486 chips. As of the end of Oct, Gateway was quoting 7 weeks to ship a Pentium system. --------------------------------------------------------------------------- EVEN IF INTEL IS GOING TO LOSE MARKET SHARE IT WILL TAKE A LONG TIME IBM mainframes have a lot of software that only runs on them. Just a few years ago there were people who said that IBM would be making 370 compatible machines for a very very long time. Today IBM is loosing money on mainframes. So while IBM may keep making them for a long time, and they still have the largest market share, they are no longer the cash cow they once were. The price of x86 chips can go down even if only a small fraction of the customers go to RISC. The unit sales might not go down much at all and Intel's high profits could still go down if the prices drop. Remember, the new "RISC" architectures have done very well in the mainframe, minicomputer, and workstation markets. RISC has done so well that in only about 7 years all old (well established) architectures in these markets seem to be either dead or on their way out. Up till now there has not been a RISC attack of the PC market. However, it now looks like SGI/MIPS/DEC/IBM/Apple/HP/Motorola/Sun/NEC/AT&T IDT/Acer/Olivetti/Siemens are all about to attempt it. A mainframe or minicomputer user has a much bigger and more complex/customized bunch of data/software to move over - so those moves were really more difficult than a PC move. Also, the technology developed to move minicomputer users to RISC machines will make the PC move much quicker. For good reason, emulating IBM's VM/CMS was never attempted, while emulating Microsoft Windows or the Mac toolkit is just not that hard (maybe by a factor of 100). Also, a single user can decide to switch to a RISC PC, while it took many people at a company to agree to move off the VAX architecture. And, note that the profits on IBM mainframes went down much faster than their market share went down (also very recently Apple has shown that you can keep market share and yet lose lots of money very suddenly). ----------------------------------------------------------------------- In 1985 Intel broke even, and in 1986 Intel lost money. So, in only 6 out of the last 8 years Intel made a reasonable profit. So, assuming the future is like the past in a random sort of way, :-), there is reason to predict a 25% chance of not making a profit next year. :-) --------------------------------------------------------------------------- Intel is spending a fair amount of money on massively parallel processors (MPP) machines. With Cray coming out with an MPP machine using Alpha chips it seems Intel may have a very hard time in this market. The Alpha floating point is very good. Cray understands applications software for supercomputers and has established supercomputer customers. Thus the MPP part of Intel could be a drain on profits. However, I don't think it ever made money (mostly lives off government contracts and Intel). Also, workstation clusters seem to be gaining momentum. --------------------------------------------------------------------------- Sales of Intel 386 seem to have stopped (Intel said Pentium sales are more than i386 in Q2 1993 and there were only 10,000 Pentiums that quarter). --------------------------------------------------------------------------- POSSIBLE SHORT TERM EFFECTS: 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. It seems the reason many "upgradable PCs" may not be is that Intel specified a plastic socket for the P24T that can not handle the heat. Ouch. 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 (Oct 1 there was about a 20% drop). MPP sales drop sharply (to Cray T3D) (though only few % of Intel's sales it hurts profits). 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 the best. Intel's best production fabs currently are 0.8 micron and they will have one 0.6 micron fab about January. 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. FAB CAPACITY Intel is the worlds largest semiconductor company in terms of profits and maybe sales. While sometimes people say "Intel has the largest fab capacity.", I don't think that is the case. Companies such as NEC make a lot of lower priced chips like DRAMs. So while their fab capacity is huge, they don't make as much money as Intel off each chip. If NEC were making 486s it might be able to make more than Intel can. In any case, since an R4200 takes 1/20th the fab capacity as a Pentium is should be trivial for NEC to make more of those than Intel makes Pentiums. In fact a tiny company like IDT may be able to make more Orion chips (faster than Pentium) than Intel makes Pentiums. So fab capacity can be measured in terms of dollars then Intel currently wins (but one would then be forced to say their fab capacity shrinks as prices go down). If it is measured in terms of Pentium equivalent CPUs of local design then NEC clearly wins. If it is measured in terms of transistors per month (or some such), then NEC and probably several others are ahead of Intel. ------------------------------------------------------------------------- From Computer Reseller News p3(1) March 22 1993 >HP plans a new line of low-cost, color workstations that will be capable >of running both Apple Macintosh and Windows applications. The new systems, >priced between $1,000 and $2,000, will compete with microcomputers based >on the INTEL 486 microprocessor. Third parties are reportedly working on >applications that will enable the systems to run Windows 3.1 natively; >Macintosh EMULATION products are also being developed. HP's systems will >be based on the company's Precision Architecture RISC 7100KLC >microprocessor, currently code-named Hummingbird, and will also be capable >of running applications designed for Sun Microsystems SPARCstations. The >HP PA-RISC architecture's advantage over the competition is its ability to >run multimedia instructions, which enables it to handle animation and >video, and provide superior multimedia performance. --------------------------------------------------------------------------- **************************************************************************** POSSIBLE WAYS INTEL MIGHT BE A FINE INVESTMENT: (a look at some opposing opinions) **************************************************************************** Intel's stock price already reflects an expectation that earnings will not stay at the recent high levels. This is shown in the relatively low P/E ratio. If you extrapolate the last 3 quarters you get about $5/share/year. With this assumption, and a stock price of $55, Intel's P/E is around 11. Given that the dow-jones P/E is about 24 now, this is very low. A stock price of $55 would be ok with a P/E of 24 and earnings of $2.30/share/year. If earnings are not going below that, then Intel's stock is already low. This is probably the best argument in Intel's favor. Note that last years total was about $2.50/share with almost half of that coming in the last quarter. The problem with this is that Intel may be losing money in a year and so the stock could go far lower. The price/sales is currently very high, so, if there were no profits a year from now, the stock could drop below $10. If improved yields allow Intel to up production of 486s as fast as prices go down they could keep up the total sales. For example, if yields go from 70% to 77% they would be making 10% more chips, and if prices dropped less than 10% they would me making more money. In spite of what Microprocessor Report says (something like 50% yield) rumors are Intel yields are much higher on 486 chips (probably because they can use slow ones and MR was looking only at 66 Mhz). It is probably getting hard to up the yields much more. If yields of 66 Mhz 486s go up and users switch from cheaper 33 Mhz 486s the average sales price for all 486s could go up even though the actual prices for any given Mhz were going down. The question is how fast do yields of 66 Mhz chips go up and how fast do prices go down. At some point yield improvements will see a diminishing returns effect (if on the same fab line) and price drops will be more significant, though Intel might be able to switch to another process by then. If Intel came out with 100 Mhz 486 chips very soon (ahead of expected 1Q 94 time frame) they could do well. (It seems they need to switch to smaller than 0.8 micron, and it seems that they had not been planning on switching the 486 originally - only the Pentium. Switching to a new process takes time). Maybe Intel comes out with a RISC chip. It would be amazing if they had kept this secret and yet had compilers, windows NT, applications etc all working. Doubt it very much. If they do come out with another RISC chip it will compete with the other RISC chips without any particular compatibility advantage. Also, other RISC vendors have second sources which PC makers like - Intel does not do that any more. Thus Intel would only win out if they had a much faster chips - which is not too likely. And even if they were able to sell the same numbers of chips as they currently are, they would have to be at the lower RISC price levels with numbers like $70 in the near future (so their sales be way down and there might not be any profits). The Pentium has logic for being used in a multiprocessor. Intel could put a RISC processor in the "overdrive" socket and have a dual-CPU system. Not clear NT can handle two different CPU types. However, if Intel can, other companies could also make RISC chips to fit into the overdrive socket. If the whole PC market has 40% growth in unit sales then even if others take part of the market Intel could keep growing. If they can keep their average chip price up and have 20% unit growth they could be losing market share and growing very fast. Microsoft estimates about 10% growth for the next year even though we had about 30% last year. Also, keeping the average chip price up when there is so much new competition seems very doubtful. If for some reason Windows NT were to flop Intel would be safer. However, reports are that it is very good. Reports are that 32 bit apps on with NT have 30% faster screen I/O and that disk is far faster. Since all interfaces/copies are done 32 bits at a time, instead of 16 bits at a time, it makes sense that the OS should be much faster. Windows NT could be so big that nobody wants it. Well, 16 MB is not that much these days as there are now samples of 8 MB chips, and SIMMs are costing about $25/MB. Memory should get cheaper as Koreans come online. Intel could get legal action stopping AMD, Cyrix, and NexGen. However, the courts seems to consistently rule that a company that has rights to Intel patents can make and sell chips compatible with Intel's even if they did not design them (Cyrix/SGS-Thomas or ULSI/HP). So if ULSI designs a 387 and HP manufactures it and sells it to ULSI, then ULSI can sell it with no trouble (see Electronic News June 21 1993 or also EE Times June 28 1993 for info on ULSI results in a rather high court set up to keep the supreme court from having to deal with high tech legal issues). Only real trouble Intel has been able to give AMD is about copying their copyrighted microcode - so AMD has redone the microcode. Also, these cases take many years. 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 Twinhead. 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-$200, which is not much compared to the CPU price differences. FLASH is an important emerging technology and Intel was leading in this market (AMD passed them). Though still small, it could become a very large market very soon. 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 for NEC to sell $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 has other companies inside the US making low cost chips (such as IDT and the R4600). 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. They may have recently dropped it. 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. Seems it will not sample till 1995. Also, it is probably twice as many transistors, so yields will be even more of a problem. Also some graphs of future products (P6, P7) compared to P5 show the P6 at least 2 years behind P5. ------------------------------------------------------------------------ Intel has made a graph where they show their chips closing the gap with the RISC chips. This graph has several errors. It claims Pentium systems were shipping 1/93 when they were not in fact shipping till 7/93. It also claims MIPS R4000 and R4400 systems were shipping far after they were really shipping. In a world where performance is doubling ever 1.5 years, getting a few data points off by half a year can close a gap. ------------------- August 9, 1993 Hot Chips 9:00-11:00 Session 1: High Integration Session Chair: Ruby Lee, Hewlett-Packard Co. . DECchip 21066 - Alpha AXP Architecture Processor for Low-Cost Applications Mark Rosenbluth, Digital Equipment Corp. . ICE: A high-performance MIPS microprocessor for portables Barbara Zivkov, MIPS Technologies Inc. . Hummingbird: A Low-Cost Superscalar PA-RISC Processor Stephen R. Undy, et al, Hewlett-Packard Co. . Optimized PentiumTM Processor 82430 PCIset for High Performance Local Bus Designs Dale Jorgensen, Intel Corp. ------------------- Computerworld, July 19, 1993, p110. "DEC CEO Robert Palmer told a group of analysts last week that his company will have to ship 4 million to 5 million Alpha AXP devices a year to pay for its investments in semiconductor technology and production facilities. Palmer did not indicate that he expects to have any trouble reaching that level over time, but some analysts said they remain skeptical. "A lot of us wonder where he's going to find that much business," said Chris Christiansen, research director for commercial systems at IDC. DEC is pinning much of its Alpha AXP volume hopes on WIndows NT-based PCs and is also looking toward the mobile and wireless market and embedded systems. Palmer did not offer a time-table for hitting the required shipment dates." ------------------- [ The following came out after Spring 1993 COMDEX ] [ Know there was a demo at Fall COMDEX but don't know ] [ much about it - except that it is fast. ] From: Richard Gorton gorton@tallis.enet.dec.com Alpha AXP Migration Tools - Digital Equipment Corp. + DEC IT SHOWS INTEL-TO-ALPHA DIRECT BINARY CODE TRANSLATOR Digital Equipment Corp is contemplating producing a translator that will take the binary code from shrink-wrapped Windows applications and spit out Alpha code. The result is faster than the standard SoftPC-based Windows and MS-DOS emulation, but slower than code specifically compiled for the RISC chip. The company launched a a direct VAX-to-Alpha binary translator last November and has been shipping an R-series-to-Alpha translator for the last three months. Nonetheless it still has qualms about the Intel Corp equivalent according to Richard Sites, the Alpha CPU architect who is demonstrating the technology at Comdex. He is trying to gauge the interest from developers, and also which set of Windows application programming interfaces it makes most sense for the translator to support: the old Win16 standard, or the intermediate Win32s version. Sites' demonstration takes a simple piece of Mandlebrot-plotting code written for the Intel machine and runs it in variety of ways on the Alpha machine. The times to complete give a rough guide as to what users can expect from heavy computation work: 66MHz 80486: 40 seconds 60MHz Pentium: 26 seconds Source code recompiled for native Alpha, running on an AXP 150: 14 seconds Intel code running in the standard Windows emulation on an AXP 150 (Sites' best guess - no benchtest run): 80 seconds Intel binary code translated to run on Alpha running on AXP 150: 24 seconds ---------------------------------------------------------- ADVERTISING SLOGANS. AMD is saying: MS Windows Compatible Cyrix is saying: Cyrix Instead Ads could ask if people want to buy a computer with any of the following written on it: inferior inside incumbent inside CISC inside Wonder just how aggressive Motorola, DEC and MIPS are. :-) They could try to get MS to let them say: microsoft inside But RISC vendors can probably get MS to let them say: MS Windows NT Compatible ---------------------------------------------------------- Micrografx (800) 733-3729 has a set of libraries called "Mirror" that a developer can link with to build a windows application that runs on OS2. It turns out that since OS2 is a 32 bit OS and faster than DOS/window that an application may run faster after porting to OS2. -------------- Novell NetWare on RISCs. August 31 1992 Open systems Today. HP had announced mid 1993 delivery date. DEC, Sun, IBM were all rumored to be developing ports. Converting servers to RISC machines could be one of the first things done since the effort/cost-saved is probably very good. Easier to bring up a new fast fileserver than to convert a bunch of users software to a new machine type. Microsoft has advertisements comparing their SQL performance to Novell and Unix machines with NT on a Pentium. They are much better in both performance and price/performance. They then say, "Imagine how much faster it would have been with NT on a RISC machine." Well, as soon as they can, I expect they will compare NT on a MIPS R4400 or Alpha to Sun on a Sparc or Novell on a Pentium. NTs big win is it works on RISC machines. Microsoft will want to stress this. This will be great for RISC sales. It has been said that Bill Gates wants lots of cheap fast hardware so there will be more people buying his software (ACE idea). This is probably true. The idea of some sort of grand Intel/Microsoft alliance is very doubtful. -------------- TI had a problem with Cyrix but now that the ULSI case makes it clear that TI would be safe making Cyrix chips (since TI has Intel patent rights) they probably are much more willing to go ahead. Also, Cyrix reached some sort of agreement with Intel on one case. So we may see TI produced Cyrix chips coming out soon. -------------- Date: Thu, 5 Aug 93 10:42:39 EDT From: CSG Hardware Technical Director 05-Aug-1993 1043 -0400 To: vincent.cate@furmint.nectar.cs.cmu.edu Apparently-To: vincent.cate@furmint.nectar.cs.cmu.edu Subject: Re: binary translation and NT Digital's AXP Migration Tools group demonstrated at Windows World the capability to transform Intel Windows NT programs into equivalent Alpha AXP NT programs using binary translation. The translated test program that was demonstrated executed 40% faster on the Alpha PC than on a Pentium-based system. Digital is the leader in the field of binary translation and has already released two binary translator products under the product name DECmigrate: one translates VAX/VMS programs to Alpha VMS and the other translates MIPS Ultrix programs to Alpha OSF/1. Both translators have been used by Digital, third party software suppliers and end users to rapidly move applications to Alpha-based systems. The Intel to Alpha binary translator is currently an Advanced Development project. No commitments to produce a product based on this technology have been made or implied. The intent of Advanced Development is to demonstrate the feasibility of a technical approach in order to receive feedback from potential users of the technology in advance of any related product development. Feedback on potential applications for Intel-Alpha binary translation should be directed to Andy Riebs, Senior Product Manager, AXP Migration Tools Group, 153 Taylor St. Littleton, Mass. 01460: DTN 227-3145, (508)-952-3145 or via Internet: riebs@tallis.enet.dec.com. ------------------- Pentium == Penultimate x86 implementation Is the P6 the last? Or is all the software going to be ported before the P6 comes out so that the Pentium is the last? ------------------- July PC Magazine article about IBM486SLC2-66 titled "The Perfect PC". ------------------- Motorola's announced quantity 20,000 MPC601 prices are $280 at 50MHz and $374 at 66 MHz. The $450 price is the ballpark price at 66 MHz in quantity 1000, as advertised in the Wall Street Journal and elsewhere. ------------------- EE Times Aug 23 1993: page 22: NEC out with MIPS R4000 100 Mhz systems in November less than $4000 NEC has a chipset to support R4000 expect to sell 100,000/month when in volume Acer said a dozen middle tier systems makers may start production because they are pessimistic about Pentium supplies but are waiting a bit to see how the market responds to MIPS systems SoftPC makes a 100 Mhz R4000 emulate x86 at 50 Mhz 486 speeds (don't really trust that claim) page 16 Intel has signed up Dell, Acer, and others to pay a 1% of retail price of a PC using a non-Intel x86 chip. This will amount to $10 to $20 price difference. Not much. Means PC makers don't have much trouble using non-Intel chips. page 70 MITI, Sumitomo predict no resin shortage -------------- From net: >According to Computerworld, July 19, 1993 Vol 27 No. 29, pp1 & 20 > >"Pentium is taking a temporary backseat at Intel Corp. in favor of a >revved-up version of the i486, which is due out in the fourth quarter, >sources close to the firm said. > The DX3 processor is expected to give near-Pentium performance >significantly lower costs, the sources said." > >The article continues to day that the 33/100 will be near Pentium >performance, until code is Pentium optimized for the Pentium. > >Their sources described two versions of the DX3: On at 25/75MHz and >33/100MHz. Also, that Intel *could* change one pin on the 33/100, and >sell a 50/100 DX2. ^^^^^^^ > >Prices of the DX3 are expected to be near that of the current DX2's >and the DX2 price's are expected to drop. Previously I have only seen "announced by the end of the year" type claims. It seems Intel had not planned on shrinking the 486 again and so jumped into it kind of late (after competition became clearer). I only know of one fab line Intel has that is less than 0.8 micron (0.6). It was doing something back as far as January 93 but did not think it was to be really online till Jan 94. Might be that shrunk Pentiums will not really be online till Jan 94. A 0.6 micron or less 486 should be fast and cheap. Could keep Intel profits up. Key issue is how much fab capacity they have at under 0.8 micron and how soon it is available. If they can only make 10% the of their 486s at 100 Mhz it could be that the drop the price of 66 and 50 Mhz chips was enough that they did not make more money. Intel has a large fab coming online in Dublin (I think 0.6 micron). This could help them alot but will not be in full production till mid 94. ------------------------ Sept PC Magazine (I think) make it look like Pentium systems are really pathetic. Of 5 system run with 3 windows benchmark aps, 3 machines were always slower than a 486DX/66, 1 was slower 2 out of 3 times, and only 1 machine was always faster and that was by an average of 28%. This is really poor. However, it also claims the Pentium has better floating point than an Alpha, so it is not clear how good these guys are at benchmarking. Also, they claimed that the ALR price of $2,500 was an upgrade card when it is really a system without monitor or harddrive. In article <1993Aug21.010820.12271@adobe.com>, zstern@adobe.com (Zalman Stern) writes: |> On another tack, I've heard that WOW runs Win16 apps faster on an R4000 than |> it does on a 486. According to "Inside Windows NT", the 486 version doesn't |> have to emulate the x86 instruction set. WOW = Windows on Windows which just means Win16 mapped into Win32 Not sure I believe this, but if true it means that today a person could be better off getting an R4000 system to run his software than a 486. From: lfern@netcom.com (T. David) >I noted that National seems to also be jumping on the x86 band-wagon. >They have been advertising for RISC and CISC designers with absolutely >no Intel experience to work on a x86 core (see misc.jobs.offered). So National is also planning on making a clean room x86 chip. Seems like they are way behind though. Would be more reasonable for them to just use Cyrix's design. ============================================================================ MAYNARD, Mass. -- September 10, 1993 -- Digital Equipment The DECchip 21066 architecture is the first in a family of low cost, highly-integrated chips for the Windows NT desktop market. More than 500 Windows NT applications running natively on Alpha AXP architecture systems will ship by the end of this calendar year. Based on the core of the existing 64-bit DECchip 21064 micro- processor, the new device includes integrated PCI interfaces, an on-chip memory controller and graphics accelerator. These features substantially reduce the cost of implementing systems based on the Alpha AXP architecture. Priced at $385 per chip in quantities of 5,000, the DECchip 21066 microprocessor operates at 166 MHz and delivers estimated performance of 70 SPECint92 and 105 SPECfp92. Samples are available now, with volume quantities shipping in the first quarter of calendar year 1994. ============================================================================ PC makers who plan to use Alpha: Olivetti Carrera Computers (800) 676-RISC (714) 707-5051 - makes MIPS motherboards too Tangent (also makes MIPS) Darius Technology Ltd PC motherboard makers that will use Alpha chips: Elitegroup Computer Systems Inc Modern Instrument Mentec ============================================================================ From the Oct 19th Microprocessor Forum schedule: Cyrix M1 - A superscalar x86 CPU that maximizes performance without recompilation [faster than Pentium on integer and the next day's schedule includes: IBM's RIOS-2 - second generation RIOS MicroSPARC 2 - A high-performance single chip SPARC MIPS T5 - A highly superscalar MIPS processor DECchip 21066 - design tradeoffs PowerPC - future direction [Just 2 inches to right comment is "You'll also hear about the first processor designed from scratch to implement the PowerPC architecture." This must be the very low cost PowerPC 603.] ============================================================================ PC magazine (~Sept) had some performance numbers for some NT systems (Alpha, MIPS, Pentium). For 32 bit "megapixle" test, the Alpha is about 3 times the Pentium, and the MIPS is about 5 times as fast (Alpha does not have byte updates of memory, and SGI does know graphics). For existing 16 bit apps I think numbers were like 2 to 5 times slower for the RISC systems emulating x86 than Pentium. ============================================================================ Sept 13 Infoworld, p8, "Digital acknowledges that it needs to improve Alpha's performance on existing DOS and windows applications to make significant inroads to Intel's desktop PC base." They don't really need to admit that. They are enough faster on new apps and not really so bad on existing apps. So, to me, this means they have decided for sure to push their binary translator. I expect to see it at Fall Comdex. ============================================================================ The SOMC (Federal Reserve Shadow Open Market Committee) said that "the momentary base -- consisting of bank reserves and currency supplied by the federal reserve -- rose 11 % in the year ending in August." They say the reason this has not yet shown up as inflation is that the public has increased its money balances by 5.5%. Around 1990 the monetary base was growing at the same rate as the GNP - so money and goods would stay at same ratio and no "inflation". There seems to be a couple year lag in monetary policy and inflation (seen this in a book or two and a graph). In 1991 the monetary base was growing far faster than the GNP and has been doing so since then. So it seems to me the safest of predictions to say that inflation will be up a lot over the next 2 years. At the first sign of significant inflation interest rates will jump and the stock market will crash. ============================================================================ NEW YORK -DJ- SunSelect, a unit of Sun Microsystems Inc. (SUNW), said it released the final master version of its Wabi 1.0 software to its OEM partners, including International Business Machines Corp. (IBM), Hewlett-Packard Co. (HWP), and Novell Inc.'s (NOVL) UNIX Systems Group. As reported, Wabi software allows Microsoft Windows applications to run directly on UNIX-based personal computers and other workstations at full performance without the need for MS-DOS or Microsoft Windows itself. In a press release, the company said the release gives the vendors the ability to ship their respective versions of the product to customers by the end of the year. (END) DOW JONES NEWS 09-20-93 ============================================================================ September 1993 PC World, page 71: > MIPS-Based PCs for NT: Fast, Inexpensive, Available Alternatives Caption: > PENTIUM PERFORMANCE AT 486 PRICES IS YOURS with the > AcerFormula 4000 and Windows NT applications Tangent Computer (800) 800-6060 $3995 Acer America (800) 733-2237 $3695 no-monitor, $4049 with monitor DeskStation Technology (800) 793-3375 $3995 --------------------- PC/Computing - page 111 (I think Aug) John C. Dvorak. "The PowerPC: For once A Non-Intel Chip Looks Like the Winner" People flocked from Zilog-Z80/CPM to DOS, and then DOS to Windows, and people moved rather fast. John expects such flocking action again but away from Intel CPUs. His guess is PowerPC (but he thinks MIPS chips are only for SGI - don't think he realizes the number of chip sources or system vendors). The Z80/CPM/S-100 sales did drop off really fast. Once people believed that the Z80/CPM combo was eventually going to loose out to Intel/DOS combo they did not want to buy a Z80/CPM system (it would mean buying something they believed was soon to be obsolete). It seems that RISC/NT/PCI could take over very fast. My own feeling is the flocking will be to NT and whatever RISC system currently offers the best price performance with little inertia/loyalty gained for any architecture. --------------------- John Dickinson, August 1993, page 98 (PC/Computing?): "Can Intel Stay on Top? Count on a Knock-Down Drag Out Fight" Says 486 prices will drop a lot. --------------------- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- CETIA makes a VME single board computer with PowerPC 601. They are in France. Phone is (33) 94088000. Also in USA at (415) 325-6170. MIPS is much faster than Pentium for NT applications. If the PC market grows by 20% each year, then at least 20% of the people buying computers are buying their first computer. As long as the applications they want are on a RISC system, they might as well buy one. NT applications ship with versions for MIPS/Alpha/Pentium. Even if most of the people the first year are using them on x86, at the end of the year there are a lot of people with software for non-x86 machines. As time goes on it is easier and easier for people to move to non-x86 machines. Motherboard Warehouse (800) 486-9975 was advertising MIPS and PowerPC motherboards in Sept Computer Shopper. First said they would both be out mid Oct but seems to have been delayed. NeTpower - (408) 522-9999 MIPS/NT machines ************************************************************************* From: sgolson@trilobyte.com (Steve Golson) This past Tuesday, September 28, Robert Palmer (President and CEO of DEC) spoke at The Computer Museum in Boston. Among other interesting tidbits, he held up the current Alpha microprocessor and said that their next generation processor "has 65% more transistors, in 65% of the area, and is 2.5x faster." He also indicated that DEC is "working on 1.5v circuits" for the next Alpha processor. ************************************************************************* It was interesting at the Intel museum in Santa Clara they show the performance of various Intel processors with bars sticking up into the air. Very fun to see these get bigger going from 4004 to the 486. Then they had a really large bar for the i860. The explanation was given that "If you can start with a clean slate you can build a much faster machine than if you have to be compatible with x86." By this logic (which I fully agree with) it makes perfect sense that the Alpha is much faster since it is a newer architecture. However, this contradicts what seems to be the Intel party line that architecture does not matter much. I also find it interesting that they seem to have canceled all future work on the i860. From: wmagro@baron.ncsa.uiuc.edu (William Magro) Newsgroups: comp.sys.powerpc,comp.sys.mac.hardware > >ctm@ardi.com wrote: >: About a dozen people who read comp.sys.powerpc have had a chance to play >: with Executor/DOS. I hope one or two of them will mention the kind of >: speed we're getting. I'm biased. >: >: --Cliff >: ctm@ardi.com > >I am one of the people trying out Executor/DOS. My first reaction to the >software was WOW, this is fast! I am running on a 40MHz AMD386. I know >from the Atari world that the Atari ST emulator (GEMulator) for DOS (built on >top of an interpretive 68000 emulator) takes a 33MHz 486 to get 8MHz 68000 >speeds. I seem to be getting about 80% of a 8MHz 68000 on my machine under >Executor. This indicates that the ARDI Synthetic CPU is considerably faster >than a standard (highly optimized) interpretive emulator. > >I should add that Executor/DOS was felt several times faster than my 8MHz >68000 based AtariST running native mac code (under Spectre GCR--reads mac >800K disks on a 720k MFM drive!). I attribute this huge speed increase >to the natively coded toolbox routines in Executor. If Executor/DOS on >my machine is an indicator of emulation speeds to come, I can imagine >being _very_ pleased with a PowerPC version (either from Apple or ARDI). > >-- >William Magro NeXTMail welcome >wmagro@uiuc.edu --------------------------------- DELL went from nothing to a very large company very fast. It is reasonable to suspect that a startup using MIPS chips could do a similar thing if the old PC makers stick with Intel. There are a number of startups using MIPS chips. --------------------------------- Profit margins are higher for Portables. The 486 architecture is really bad for portables. The R4200 will blow it away. --------------------------------- Looking at Apple switching customers over to PPC. They did well moving from Apple II to Mac. Sun moved people from 68000 to Sparc and did very well. DEC moved people from PDP-11 to VAX and did very well. IBM 709x to 360. In high-tech, moving to new tech may be painful, but frequently very profitable in the medium/long run. Looking at Microsoft switching customers to Win32 (and RISC). This is far easier than switching to Windows. Windows in 3 years passed DOS. ------------------------------------------------------- IBM is making microcode for a PowerPC to run x86 code. This is silly. This is a waste of time since binary translators result in better performance and no extra cost. ------------------------------------------------------- Electronics News (around Nov 5) talks about the IDT Orion (which is now the R4600) and the NEC R4200. Both are sampling now. The R4200 is 55 SpecInt, (think about 30 SpecFP), and $80. The R4600 is 68 SpecInt (around 60 SpecFP) and something like $240. It is fun to note that the R4600 is smaller than the R4200 but since it is faster they can charge more. Both are smaller than a 486 which costs around $30 to make. Both are very low power. I think both plug into R4000/PC or R4400/PC sockets. There should be demos of both at Comdex in 2 or 3 weeks. *********************************************************************** -------------------------------------------------------------- From: ctm@ardi.com Subject: Re: Binary Translators (was Object code postprocessors) Date: Wed, 3 Nov 93 14:29:09 PDT Sender: news@cobra.cs.unm.edu In article vac+@cs.cmu.edu (Vincent Cate) writes: ... >BINARY TRANSLATORS: > ... >ARDI (or Ardi) - Abacus Research and Development Inc (505) 766 9115. >Currently translate Mac to some intermediate code but will translate to >native code. Even running the intermediate code with their "synthetic >cpu" software, they claim it is very fast. Was only working on machines >running NEXTSTEP but should be out for DOS machines in Oct. Can FTP a >demo .tar file from ftp.cs.unm.edu in pub/ardi. Sounds like they will do >an Alpha version and maybe a PowerPC version. Thanks for the mention. Specifically we are shipping two products that can run some Macintosh programs: Executor/NEXTSTEP and Executor/DOS. The NEXTSTEP product will work both with original NeXT hardware (in which case the native CPU is used) and with PCs running NEXTSTEP. The DOS program requires a '386 or better and DOS (not Windows or OS/2). We have the synthetic CPU running fine on an Alpha, although we are still tweaking our "Mac toolbox" code to understand that Alphas have 64 bit pointers. Once we have Executor/X-Windows running on the Alpha, we should be able to port it in less than a week for any 32 or 64 bit processor that has gcc and gdb. We're doing the Alpha port first because DEC is nice to us (their binary translator group likes what they know about us). Executor is not a porting tool, it is for end users and does the recompilation on the fly (dynamically). Executor/DOS has been shipping for a week now; you can expect to see reviews in most popular PC periodicals in six to twelve weeks (lead time, doncha know). There will be no FTPable version of Executor/DOS because it's not copy-protected and doesn't use a license server (unlike Executor/NEXTSTEP). However, if you want to play with it, it is inexpensive ($99 from us, I think around $65 from Dustin Discount and probably the same from PC Connection). Currently Executor/NEXTSTEP has poor screen update performance because of the penalty of going through Display PostScript, while Executor/DOS just goes directly to the screen. More information can be sent via e-mail, NEXTmail, FAX or US mail if you send e-mail to questions@ardi.com or call us at +1 505 766 9115. You can also FAX us a purchase order at +1 505 247 1899. Always be sure to mention whether you're interested in/ordering the DOS or NEXTSTEP version. Mat Hostetter, the author of both our current synthetic CPU and the author of what will be our next generation (recompiles to native code for '386, Alpha, PowerPC, MIPS and maybe SPARC) syn CPU is the person who tipped me off to this thread in comp.arch. I think he's in a time crunch now, but hopefully, when he can spare some time, he will post a more technical description of what his synthetic CPU does. --Cliff[ord T. Matthews] ctm@ardi.com ---------------------------------------------------------------------- Something like 2/3rd of Intel's CPUs are sold to Taiwanese motherboard makers. Surprisingly Intel has shown no loyalty to these companies. Intel delivers chips faster to American companies, and at better prices. So the Taiwanese companies have no loyalty to Intel. They are very happy with AMD, Cyrix, and NexGen. They also seem interested in RISC. ---------------------------------------------------------------------- Subject: ****Comdex - MIPS Shows Fast R4200-Based PCs 11/18/93 Date: Thu, 18 Nov 93 12:52:08 PDT LAS VEGAS, NEVADA, U.S.A., 1993 NOV 18 (NB) -- At Comdex, MIPS Technologies and its partners are demonstrating newly developed notebooks, PC upgrade boards, and workstations based on MIPS' new R4200, a microprocessor billed as faster and less costly than the Pentium, Alpha, or PowerPC 601. The new MIPS platforms, and the reference designs that made them possible, were announced at a press conference headlined by speakers from MIPS and three hardware allies: NEC, ShaBlamm! Computers, and DeskStation Technology, Tom Whiteside, president of MIPS, told journalists that MIPS has created four R4200 reference designs: a notebook designed for high performance; a notebook designed for long life; a desktop PC designed for high performance; and a low-cost desktop model. R4200-based notebooks and desktop PCs, running Windows NT, are on display at Comdex in the MIPS booth, with an additional notebook in the NEC booth, Whiteside said. "We're very well positioned," he added. New benchmarks show that PCs based on a 150 megahertz (MHz) R4200 MIPS processor runs 202 percent faster than 60 MHz Pentium-based PCs, 297 percent faster than PCs with an Intel 80486 DX-2 microprocessor, and 34 percent faster than Alpha AXP-based PCs when the systems are running Windows NT, he asserted. The MIPS chip can also run Unix, he noted. Further, the R4200 costs one-tenth as much as the Pentium, and consumes one-tenth the power, according to the MIPS president. Already, about 150 independent software vendors (ISVs) have committed to support the new MIPS chip, said Whiteside. "And around 100 of these ISVs represent (major companies)." In addition, two chip foundries -- NEC and Integrated Device Technology (IDT) -- have licensed the R4200 technology from MIPS and are now manufacturing the new chip in sample quantities. NEC is calling its product the MR4401-75, while IDT's product is dubbed the Orion R4600. "We're producing the chip at lower cost and lower power consumption than expected," said Dave Corbin, director of marketing for NEC, The MR4401-75 is ideal for use in notebooks, as well as in game, set top, and many other devices, he added. Don Peterson, president of Desktop Technology, told the journalists that the MIPS 4400 offers greater performance than either the Pentium or the PowerPC 601, the only PowerPC chip shipping in volume quantities up to now. DeskStation Technology is using the MIPS chip for the Tyne Series, a new family of RISC (reduced instruction-set computer)-based workstations that is similar in size to a 486-based PC. Models in the Tyne Series will be priced at under $3,000, according to Peterson. "If any of you believe that PowerPC will offer a more compelling platform, nine months from now, I'll eat your hats," Peterson remarked. Sheblamm! Computing is employing the chip on an upgrade board for 486-based systems aimed at providing installed PCs with workstation-level performance, said Tom North, company president. North also noted that Sheblamm! is the acronym for "Sizzling Hit Accelerator Board Highly Affordable MIPS Microsoft Windows NT." The four reference designs for MIPS 4200 announced this week will complement existing reference designs for MIPS-based PCs, according to Whiteside. In addition to notebook and desktop PCs, the new MIPS and derivative core products are aimed at use in hand-held and other consumer products as in embedded applications such as laser printer controllers, X-terminals, and communications devices. (Jacqueline Emigh/19931118/Press contact: Steve Schick, MIPS, 415-390-2573) Date: Fri, 19 Nov 1993 00:15-EST Subject: Re: Fwd: ****Comdex - MIPS Shows Fast R4200-Based PCs 11/18/93 There were a bunch of errors in that article. The IDT R4600 is a different chip from the NEC R4200 not just a different name. Also NEC and MIPS have been calling it the R4200 so this "MR4401" is suspect. There is another totally different chip that is called the R4400. The R4200 does not yet run at 150 Mhz, that is the R4400. The benchmarks running twice the speed of the Pentium were with the R4400. Also, the "202" percent faster should have been 202 percent the performance or "102 percent faster" or much simpler, "2 times the performance". Still, the "R4200 costs one-tenth as much as the Pentium" is correct and that is the most important thing. MIPS is coming through as expected. Fall COMDEX had 200 Mhz R4400s, 95 Mhz PowerPC 601s, 320 Mhz Alphas, but I did not hear of any faster 486/Pentium chips. ------------------------------------------------------------------