vsnyder@jato.jpl.nasa.gov (Van Snyder) (02/08/91)
Now that the 68030 is obsolete, and the 68040 is about to be, but the 68050 isn'ty enven announced yet, how about a TT follow-on with an 88000? -- vsnyder@jato.Jpl.Nasa.Gov ames!elroy!jato!vsnyder vsnyder@jato.uucp
vsnyder@jato.jpl.nasa.gov (Van Snyder) (02/08/91)
The 68030 and 68040 are obsolete because:
1. At the same clock rate, even the 68040 is less than half as fast as
the 88000.
2. Motorola has only two REAL 680x0 customers left: Apple and Motorola. NeXT
might pick up some slack, but not more than 2-3% in the foreseeable future.
3. Motorola has already announced the 88110, twice as fast as the 88000, with
the CAMMU built in (not on a seprate chip).
4. Motorola has publicly boasted that 88110 will be operating at 200 Mhz by
the end of 1992. They might not make it, but... I'd rather have a 200
MIPS single-chip computer than a 15-MIPS one.
5. Right now, 88xxx's are more expensive than 68xxx's, but that won't last
forever.
6. Motorola has several REAL customers for 88xxx: Data General, Tektronix,
Sequent, Opus, and, of course, Motorola.
--
vsnyder@jato.Jpl.Nasa.Gov
ames!elroy!jato!vsnyder
vsnyder@jato.uucpstephen@oahu.cs.ucla.edu (Steve Whitney) (02/08/91)
In article <1991Feb8.010109.396@jato.jpl.nasa.gov> vsnyder@jato.Jpl.Nasa.Gov (Van Snyder) writes: >The 68030 and 68040 are obsolete because: >1. At the same clock rate, even the 68040 is less than half as fast as > the 88000. Are you talking instructions per second here? I'm sure you realize that the individual 68040 instructions are more powerful than individual 88000 instructions (in general). >2. Motorola has only two REAL 680x0 customers left: Apple and Motorola. NeXT > might pick up some slack, but not more than 2-3% in the foreseeable future. I don't what your definition of REAL customers is, but I _know_ that there are a _lot_ of computer peripherals which use 680x0 processors. No, they're not personal computers, but there are a lot of 'em. >3. Motorola has already announced the 88110, twice as fast as the 88000, with > the CAMMU built in (not on a seprate chip). Is the CA for content addressable? I'm not familiar with this term. >4. Motorola has publicly boasted that 88110 will be operating at 200 Mhz by > the end of 1992. They might not make it, but... I'd rather have a 200 > MIPS single-chip computer than a 15-MIPS one. And how much will this beauty cost? You can get some pretty fast ECL chips right now. Darned expansive (and power hungry) though! >5. Right now, 88xxx's are more expensive than 68xxx's, but that won't last > > forever. >6. Motorola has several REAL customers for 88xxx: Data General, Tektronix, > Sequent, Opus, and, of course, Motorola. I'm sorry, but I still don't understand your definition. > >-- >vsnyder@jato.Jpl.Nasa.Gov >ames!elroy!jato!vsnyder >vsnyder@jato.uucp -- Steve Whitney "It's never _really_ the last minute" (())_-_(()) UCLA Comp. Sci. Grad. Student | (* *) | Internet: stephen@cs.ucla.edu UCLA Bruin--> { \_@_/ } GEnie: S.WHITNEY `-----'
bcc@Eyring.COM (Brian Cooper) (02/08/91)
In article <1991Feb8.010109.396@jato.jpl.nasa.gov> vsnyder@jato.Jpl.Nasa.Gov (Van Snyder) writes: >The 68030 and 68040 are obsolete because: [several arguments deleted] >6. Motorola has several REAL customers for 88xxx: Data General, Tektronix, > Sequent, Opus, and, of course, Motorola. ^^^^^^^^^ ^^^^^^^^ I think it`s funny that you should propose this as an argument against the 68040 -- the 88000 is better than the 680x0 because Motorola backs the 88000? I think that Motorola makes lots of boards for the 680x0 as well! Also, didn't I see recently that Tektronix decided there was no future in the 88000 for them and they decided to get out? They closed down a division or something? Going by usenet alone, the 68k traffic has been higher than the 88k traffic for some time now. (P.S., I like both chips, but I doubt that the 88K will ever make the 68K go away ... and I don't think Motorola wants it to.) Most of the 88K boards I have seen used a 68K to handle the I/O anyway; the interrupt handling destroys the advantages of pipelining and caching.
vsnyder@jato.jpl.nasa.gov (Van Snyder) (02/09/91)
1. Speed. I don't count how many instructions are executed per second. I
haven't looked at the output of a compiler more often than twice per year
since about 1970. I look at how fast a given program gets executed. At
the same clock rate, the 88000 gets done twice as fast as the 68040.
2. REAL customer: A stable company with a good reputation that buys several
hundred thousand or more of some particular part, e.g. 68030 or 88100, per
year, year after year. By this measure, Atari is not a REAL Motorola
680x0 customer -- too small a player (several tens of thousands at most).
3. CAMMU means "Cache And Memory Management Unit", presently a separate chip.
4. I know that 200 Mhz means 5 ns clock, and that a Y-MP has a 6+ ns clock.
But it also has MILES of wires. The Fujitsu VP-200 has a 2+ ns clock.
The 88xxx is built on a 1+ cm die, so no wire is longer than about 2cm,
about .17 ns. The Y-MP has 64 bit paths, and TERRIBLE arithmetic. The
88110 has 80 bit data paths, and IEEE arithmetic (much better than the
junk Cray calls arithmetic). With SuperScalar and Multi-Pipelining,
the 200 Mhz 88110 will be faster than a Cray 1-s, and at $1000 (initially)
will cost al LOT less. When the price inevitably comes down to $50,
who would prefer a 68090 that costs the same (give or take $5), but runs
half as fast at the same clock rate? (The "half as fast at the same clock
rate" part is inevitable: The 680x0 instruction set REQUIRES hazards in
the pipeline that INEVITABLY slow it down, except at the expense of
ENORMOUS numbers of transistors. To achieve the same speed as an 88xxx,
a 68xxx will need 2-4 times as many transistors, and therefore ultimately
COST A LOT MORE.)
5. My first hand calculator was an HP "Digital Slide Rule", 6 digits, add,
subtract, multiply, divide. No square root. $600+. They GIVE you a
better calculator now if you buy a toaster at K-mart. I don't think the
price of 88xxx chips will be FOREVER greater than the price of 68xxx
chips, if for no other reason than the 88xxx has fewer transistors!
--
vsnyder@jato.Jpl.Nasa.Gov
ames!elroy!jato!vsnyder
vsnyder@jato.uucpplinio@boole.seas.ucla.edu (Plinio Barbeito/;093091;allsites) (02/10/91)
In article <1991Feb8.204957.10770@jato.jpl.nasa.gov> vsnyder@jato.Jpl.Nasa.Gov (Van Snyder) writes: >1. Speed. I don't count how many instructions are executed per second. I ... > the same clock rate, the 88000 gets done twice as fast as the 68040. Running what software? Not existing ST software. To run ST software without modification requires that the 88000 system be able emulate 680x0 instructions, or that all of the developers port their software to it (a dubious proposition 8-P ). Assuming an emulator was written, you'd need to crank much more than twice the dhrystones to keep up with the 040. For present 68000 machine owners, the 68040 is not an "obsolete" upgrade path in terms of speed. It will run their existing software much faster than any processor that will fit in their case (see, this excludes the Cray :) ). What software did you have in mind? >2. REAL customer: A stable company with a good reputation that buys several ... You don't need real customers if you have real volume; the kind of volume intel gets from the jillion little clone companies buying 80286's. As long as the new 486 runs DOS programs, intel doesn't need one real customer to make money off of it. The same could be true of Motorola, if Mac clones (for example) became ubiquitous. As evidence of this, the success of Sun's SPARC technology is due in large part to the fact that Sun didn't keep it proprietary, actually encouraging third parties to produce it, assuring a large installed base, assuring a future demand for new and improved Sun workstations and software. >3. CAMMU means "Cache And Memory Management Unit", presently a separate chip. This seems a good way to design a fast system, even a system that is easy to implement into a final product, but it is not the way to make an economical personal computer, where you try to reduce the chip count as much as possible. Instead of buying one chip, you have to buy three. This is likely to cost more than buying one chip. See item 4 (PRICE?) below. >4. I know that 200 Mhz means 5 ns clock, and that a Y-MP has a 6+ ns clock. ... > a 68xxx will need 2-4 times as many transistors, and therefore ultimately > COST A LOT MORE.) As someone mentioned, the price of a memory system to support a 200MHz clock is prohibitive. A cheap processor with an expensive memory system and many expensive memory chips costs more than one expensive processor and a simple memory system and many cheap memory chips. Notice that the 68040 includes an FP and cache on-chip. That effectively reduces chip count and final system cost, if you were going to include the FP and cache anyway. If you weren't, you might be interested in the fact that Motorola plans to release a low cost version of the 68040 this year ...I think it's called the 68EC040. No FP, smaller cache, fewer transistors: lower price. Just what the doctor ordered for upgrade boards. Of course, right now it's just as available as the 200MHz 88K :-). ... > I don't think the > price of 88xxx chips will be FOREVER greater than the price of 68xxx > chips, if for no other reason than the 88xxx has fewer transistors! Another way to look at it is that they will forever cost about the same and the 68xxx will be slower. But that does not mean people will deem it obsolete in terms of their own needs. For most people's personal needs, a 68030 is not just adequate, it's ideal. Unlike the 68000, it has an on-chip cache (256 words of data, 256 words of instructions), so that even if it's driven up to 33MHz and used with slow (cheap) memory chips, it may not necessarily need a large external cache to give a reasonable speed improvement over 16 or 25MHz versions. Furthermore, a 16MHz 68030 is dirt cheap NOW. And if you could get a 50MHz 68030 at a similar price, you could have your cake and eat it too. For reference, my definition of 'obsolete': A product is obsolete if there is something better which can be used for the same purpose. That's why I think that if people can't use the alternative because it costs more than they are willing to pay, or if it isn't fit for the same purpose, then the original product is not obsolete. >vsnyder@jato.Jpl.Nasa.Gov >ames!elroy!jato!vsnyder >vsnyder@jato.uucp plini b -- ----- ---- --- -- ------ ---- --- -- - The proper place for the capslock key is not where the key 'control' used to be. plinio@seas.ucla.edu * Boelter 4442 lab: 206-1982, Boelter 3760: 825-7413
chen@digital.sps.mot.com (Jinfu Chen) (02/12/91)
In article <1991Feb8.010109.396@jato.jpl.nasa.gov> vsnyder@jato.Jpl.Nasa.Gov (Van Snyder) writes: >The 68030 and 68040 are obsolete because: >2. Motorola has only two REAL 680x0 customers left: Apple and Motorola. NeXT > might pick up some slack, but not more than 2-3% in the foreseeable future. Don't forget HP/Apollo, Unisys, NCR, Atari, Commodore, Force Computer, just to name a few. And both unit/dollar sales of 680x0 based products are much higher than the 88k lines. There is no doubt 88k is superior than 680x0, but there are still large customer base sitting on 680x0 codes. Disclaimer: Motorola doesn't know we have net access. -- Jinfu Chen (602)898-5338 Motorola, Inc. SPS Mesa, AZ ...uunet!motsps!digital!chen chen@digital.sps.mot.com CMS: RXFR30 at MESAVM ----------
mjs@hpfcso.HP.COM (Marc Sabatella) (02/12/91)
> >The 68030 and 68040 are obsolete because: > >1. At the same clock rate, even the 68040 is less than half as fast as > > the 88000. > Are you talking instructions per second here? I'm sure you realize that the > individual 68040 instructions are more powerful than individual 88000 > instructions (in general). Not only that, but I doubt the original claim. The 68040 comes pretty close to one instruction per cycle if you perform some instruction scheduling. It can almost certainly average under 2 clocks per instruction. Does the 88000 average under 1? Anyhow, what counts is how fast an "equivalently equipped" computer with the two chips will run. A 25 Mhz 68040 can generate a SPEC rating of around 11, or more with an external cache. It cranks about 30,000 Dhrystones. Will a 25 MHz 88000 get over 20 SPECmarks or 60,000 Dhrystones? >2. Motorola has only two REAL 680x0 customers left: Apple and Motorola. NeXT > might pick up some slack, but not more than 2-3% in the foreseeable future. I think you are forgetting the workstation market. HP (& Apollo) has a rather large market share (#2, behind Sun), almost entirely based on 680x0 workstations. -------------- Marc Sabatella (marc@hpmonk.fc.hp.com) Disclaimers: 2 + 2 = 3, for suitably small values of 2 Bill and Dave may not always agree with me
d_alvear@jhunix.HCF.JHU.EDU (Dom Alvear) (02/13/91)
In article <1991Feb7.213305.26568@jato.jpl.nasa.gov> vsnyder@jato.Jpl.Nasa.Gov (Van Snyder) writes: > >Now that the 68030 is obsolete, and the 68040 is about to be, but the ^^^^^^^^^^^^^^^^^ Why is it that everybody in the world thinks that a processor is obsolete when it's successor is announced or becomes available??? I mean, come on, the 68040 is such an expensive chip that developing a machine for it would be cost prohibitive. And if the 68030 were truly obsolete, then why are OTHER companies, i.e. NOT ATARI, still using them? (Apple, commodore, etc?) Finally, sure there are newer processors available, but have YOU, or anyone else for that matter, completely exhausted the utility and power of the lowly 68000??? Realistically, the 68030 is still a powerful chip, and it really is NOT obsolete. It is older than the 68040, but it surely hasn't been exhausted of its potential yet... >68050 isn't even announced yet, how about a TT follow-on with an 88000? >-- >vsnyder@jato.Jpl.Nasa.Gov >ames!elroy!jato!vsnyder >vsnyder@jato.uucp Dom Alvear (no .sig) P.S. On a completely unrelated note, how does one send an article to the net without doing a followup. My newsreader doesn't seem to have a submit option...
piet@cs.ruu.nl (Piet van Oostrum) (02/13/91)
>>>>> In message <7340079@hpfcso.HP.COM>, mjs@hpfcso.HP.COM (Marc Sabatella) (MS) writes:
MS> I think you are forgetting the workstation market. HP (& Apollo) has a rather
MS> large market share (#2, behind Sun), almost entirely based on 680x0
MS> workstations.
Make that #1 (in the Unix workstation market). You as an HP'er should know
that :=).
--
Piet van Oostrum <piet@cs.ruu.nl>