Path: utzoo!yunexus!geac!syntron!jtsv16!uunet!lll-winken!lll-tis!ames!ncar!gatech!rutgers!apple!voder!pyramid!prls!mips!mash From: mash@mips.COM (John Mashey) Newsgroups: comp.arch Subject: Architecture Wars, again; survey Message-ID: <7316@winchester.mips.COM> Date: 30 Oct 88 02:07:11 GMT Article-I.D.: winchest.7316 Lines: 177 A bunch of people have emailed me random questions about the "BRAWL" I mentioned before, as well as related topics. Rather than emailing everywhere, I'll post some answers, trying pretty hard to be fairly objective: >Also, betting on a new architecture at the beginning of a cycle [i.e., >in the Z8000/68K/X86/X32 etc wars in the early 80s, and the current >BRAWL (Big RISC Architecture War & Lunacy) is very exciting,.... >Consider, choosing an architecture >is like an odd form of Russian Roulette: you pick a chip and pull the >trigger, then wait a year or two to see if you've blown your brains out. >Fortunately, the BRAWL will be over before the end of the year, >which will make life saner. Q: WHAT WAR ARE YOU TALKING ABOUT? A: There are at least 2: War1: Systems design-ins (esp. UNIX) War2: Embedded controllers (anything from airplanes to laser printers) The war is similar to the early-80s design-in battles: get there first into new application areas and get enough successful design-ins for your architecture that make it one of the "standards" (whatever THAT means). Q: WHO ARE THE SIDES? A: This is very complicated, especially as the turf is split up in ways where -some are publicly committed -some are committed, but not publicly -some are not yet committed, but they're now down to choice amongst at most 2 architectures. -some sell computers and/or chips to companies who are otherwise even publicly committed to somebody else (!) -some people's business causes them to play with multiple sides (This is like Diplomacy or RISK, if you know those games). War1: (major sides) MIPS & friends (R3000) Semi partners: Performance Semiconductor, Integrated Device Technologies, LSI Logic; +2 more (large ones) to be named soon. Sample of announced, committed design-ins: DEC, Tandem, SGI Motorola & friends (88000) Sample of announced, committed design-ins: Data General, Stratus, Tektronix Sun & friends (SPARC) Semi partners: Fujitsu, LSI Logic, Cypress, Texas Instruments; BIT Sample of announced, committed design-ins: AT&T, Xerox, Arix There is a bewildering variety of licensing arrangements (hardware, software, "architecture", designs, etc) that surround all of this. An "architecture" license lets somebody build hardware that follows your architecture, usually where they might want to do something different (beyond semiconductor partners). Public examples include: SPARC: Prisma (GaAs), Solbourne; MIPS: DEC; Motorola: DG. Finally, the last wild-card in this war is the Intel N-10, which is lurking around behind the scenes, but not announced. There probably can't be more than about 2 real winners in this one; all the sides have differing combinations of strengths and weaknesses; we all beat each other up as often as possible (like, next Wed's Santa Clara IEEE meeting, for example, with MIPS, Moto, and Sun.) War2: (I don't understand this as well, as it's handled more by our semiconductor partners; also MIPS, right now, mostly plays in the high end of this.) Still, here's a cut: AMD (29000) (aimed at embedded; medium-high to high end) Intel 80960 (embedded; medium to medium-high, as far as I can tell; maybe Intell-ites will say something). MIPS (R3000) (in high end; in defense applications; where software is deciding factor; where general-purpose FP is important) (CISC chips of various ilks) [If somebody who knows can describe this better, please do.] This war is probably more prone to having multiple survivors. Q: WHY DO YOU LIST THEM THE WAY YOU DO? A: (MIPScentric): In War1, we used to fight with SPARC a whole lot for the systems design-ins; now we bang heads more with the 88K; we've only once seen the 29K in that arena. In War2, we most often run into the 29K, and others occasionally. Presumably, as you go down the performance range, AMD runs into other people. Q: WHY DO YOU CLAIM THE WAR WILL BE OVER BY THE END OF THE YEAR? A: This is an overstatement, in one sense, and accurate in another. Also, the 2 wars are somewhat separate, although they interact (if you can get volume in one of the Wars and reduce your costs, you might get an edge in the other war, or with customers who'd like the same chips&software for both systems & embedded applications.) War1: The first battle (get systems vendors to pick sides) will be over: ALMOST ANYBODY WHO DOESN'T ALREADY HAVE A RISC, AND WANTS ONE, WILL LIKELY PICK SIDES VERY SOON. This doesn't necessarily mean that they'll announce their choice; that means that typically the only other people who know will be the architecture vendor who won a given customer, and the one who came in second. (The winner must keep quiet, even though they'd rather not; the one in second place has little motivation to blab! :-) Then there will be an implementation frenzy, and the second battle will really get rolling, i.e., it's possible to win some parts of the first battle, and still lose the war, if you make promises to get the design-in, but things don't come together. [My boss calls this the distinction between the "air-war" and the "ground-war". The "air-war" is fought in the press, including rather long-term pre- announcements; the "ground-war" is fought in getting real products built and shipped.] War2: This is a little different, in that software is a major part of War1, and acts differently in War2. That is, a systems vendor, having chosen a RISC, will tend to stick with it, just as they did with CISCs, given the software investment, and even visibility of the architecture to users. Non-reprogrammable devices (laser printers, etc) need software also, and buyers certainly have some loyalty, but decisions can be and are made on a more project-by-project basis. Thus, this war won't be settled for a while. Q: WHO'S LEFT? A: Well, some important computer vendors are not publicly committed to a RISC: - for some, RISC isn't particularly relevant to their business. - some of them are making up their minds (and usually pretty soon) - some of them are committed, but not publicly Following is a quick sample (U.S. only, because I'm looking at nice magazine-provided list, grossly ordered by computer revenue: I had to guess a couple places from electronics-revenue numbers). *'d ones are the obvious up-for-grabs (at least publicly; I know better in a couple cases), and mentioning only ones where it seems RISC-micro choice is a relevant issue. IBM: ROMP (PC/RT) & followon DEC: MIPS *Unisys: - (contrary to press early this year; read Datamation September 1, 1988, page 53 if you don't believe me) HP: PA (Precision Architecture) *Honeywell: - *NCR: - *Apple: - *Wang: - Xerox: SPARC AT&T: SPARC (contrary to recent silly rumors, I think) CDC: - (MIPS, but only sort of, via Silicon Graphics) Data General: 88K *Prime: - (complex, because announced RISC_based hardware platforms are OEMed workstations and include many vendors; there are various side deals, incl. software, and MIPS is probably more in there than others, but would be hard to call from public data.) Tandem: MIPS Sun: SPARC InterGraph: Clipper Apollo: PRISM (DN10000) Tektronix: 88K (stops around $600M computer revenue in calendar 87) As you can see, some of the ones left are pretty interesting; I don't think I'm giving away anything to say that many of the *'d ones are basically 88K vs MIPS fights at this point. Q: WILL RISC CHIPS ELIMINATE CISC CHIPS? A: No, X86s, 68Ks, etc, will be with us forever. As discussed earlier in this newsgroup, you can usually keep tuning a CISC architecture to go faster: it just costs you more $ and time to do it. Hence, RISCs should maintain a performance advantage, even in the next round where everybody stuffs more things on single chips. The other place where the performance advantage is maintained is that some of the CISCs just don't have enough registers: good optimizers are still worth having, but they don't turbo-charge performance as much as they do for RISCs. Q: WHO WILL WIN? A: Well, that's the $64K question, isn't it? Stay tuned. -- -john mashey DISCLAIMER: UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com DDD: 408-991-0253 or 408-720-1700, x253 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086