Conference 7.286::digital

    
    Intel would probably sue us.

    >Why don't we buld our own x86 clone chip using all the tricks and
    >techniques
    >learned from Alpha? Make it the baddest, fastest Intel Pentium (with
    >MMX of course) clone on the planet.
     
    OK--  I'm not a semiconductor designer, but...
    
    Intel's Pentium revenues far exceed DIGITAL's Alpha revenues.  Intel's
    total revenues significantly exceed DIGITAL's total revenues.  Intel
    has more capital to do chip development and has a significant headstart
    in producing a faster chip.  It seems unlikely that we could make do
    what .0 modestly proposes, even if we diverted the entire Alpha R&D
    budget to do it.  Given Intel's head start, by the time our chip
    reached market, Intel will have leap-frogged us.
    
    DIGITAL has been able to generate chips that out perform Intel's
    offerings BECAUSE we are not constrained to supporting the x86
    archecture.  
    
    As far as the mode bit thing...  It would probably be
    cheaper to build a CPU board with both processors on it.  
    
    Jim

    Let me get this straight...
    .0 suggested
    
    >>>Why don't we buld our own x86 clone chip using all the tricks and
    >>>techniques
    >>>learned from Alpha? Make it the baddest, fastest Intel Pentium (with
    >>>MMX of
    >>>course) clone on the planet.
    
    >>>We could even get fancy and extend the Alpha architecture by adding an
    >>>Intel
    >>>layer (and a mode bit ;-} )
    
    And .1 responded
    
    >>>Intel would probably sue us.
    
    So Intel would sue us for using our own technology to do a better
    job of pirating technology that they allegedy pirated from us in the
    first place?
    
    Sounds like a recursive lawsuit; the press would love it.
    
    
     

    About the time we would come out with our FAST x86 clone, Intel will
    have come out with a chip that breaks with the 80186 legacy and
    all the attending baggage! 
    
    Question is....would this be bad for DEC, or bad for Intel?
    

    re: .5
    
    If we could get in the place where Intel lacks compatibility, but we
    have it, we could seriously increase our marketshare, IMO.
    
    I wonder if AMD & Cyrix already have this as a goal...?
    
    -- Russ

Intel would just plow us under with their manufacturing capability.  They 
could end it all for AMD or Cyrix any time they want, but I don't think 
either one is a big enough bother to make it worth the antitrust suits 
that could follow.

Bruce

 >  Sounds like a recursive lawsuit; the press would love it.
 
No, the lawyers would love it.  

Just imagine, if we could actually get this going, we may be able to suck
all the lawyers in the universe into a black hole, never to be seen, heard, or
billed from again!

I can dream can't I?

    re: .6
    
    Manufacturing capability isn't the key here.
    
    If Merced requires a migration, that means there is a potential market
    for a Merced-like processor which doesn't require a migration.
    
    -- Russ

|    If Merced requires a migration, that means there is a potential market
|    for a Merced-like processor which doesn't require a migration.

Somehow I do not think the Merced will require migration thanks to excellent
marketing. It will probably appear as the best of two worlds. The fact that
the vast majority of users will not see any performance gains with Merced until
code takes advantage of it will probably go just as unnoticed as MMX being
nothing more than an enabling gimmick. 

It is not impossible that our biggest opportunity could become HP's transition
from proprietary to Merced. I've seen some reports already suggesting that
their NT annoucement and Merced work is creating a lot of uncertainty. We
also know, for a fact, that porting to 64 bit is not painless. Perhaps we
can lure some of their customers away to either 64bit NT or 64bit UNIX? I
mean, if you've got to bite the bullet, why not start with robust code on
the UNIX side, today, and get a head start on the NT side very soon?

>Per

I'd rather see us help AMD with the K6 (improving
the FPU performance)  and work a deal for royalties
ples an agrrement to use idle Fab 6 time to manufacture
the improved K6.

    The key area is not the CPU itself but the supporting chips for the
    motherboards. Intel has had this well sown up and the competitors dont
    have strong penetration. Also, Intel have not made fundamental
    improvements in their chipsets families for some time, just reworked them 
    at the same price points. If we look at the success some board
    manufacturers have had at clocking boards over 66Mhz, despite Intel's
    protests, there is certainly some feeling that Intel is "dogging it" in
    the supporting chips space.
    
     To have any chance of success, you have to support all the major CPU 
    chips, not just AMD or Cyrix or Intel alone for that matter.
    
    SMP has been the achilles heel of competitors against Intel. It is 
    something we know how to do. There is currently not good support for
    SMP on any third party chipsets. Apparently AMD have an SMP
    architecture but noone producing to it.
    
    Then, if we could help AMD with their floating point performance, who
    knows what may happen. The K6 is considered pretty good in everything
    except this. 
    
    
    Just a thought
    - John

>    Manufacturing capability isn't the key here.
    
    
If Intel decided to price Merced at $100 ea, in quantity?  They could, 
and they'd be making a decent margin, comparable to our SYSTEMS margin, 
if not better.  Add in what they are doing with PROCard, or whatever the 
marketing term for the next gen processors comes out with (single edge 
connecter or something?) and you'll soon see that they've got people 
locked into their manufacturing process, which no one else can match.

>    If Merced requires a migration, that means there is a potential 
>    market for a Merced-like processor which doesn't require a 
>    migration.

Pretend you are a customer.  Digital tells you that it's easier to 
migrate to an Alpha from your PP.  Inte ltells you the same thing.  Who'd 
you believe?

Bruce

Re: .0 

I think that would be a great idea. Particularly with some variations on 
that theme that ongoing litigation might permit.

I suspect it will not happen. A few years the ago the IEEE proposed a
standard for an Instruction set. This would have been great for customers
and vendor's alike had it been successful. It would have really opened up
the CPU market. However within Digital's chip design community this 
proposal was attacked in a widely circulated memo. It invited all
standards reps to oppose the putative standard. 

As I recall from the memo, a standard instruction set would restrict the
chip designers ability to innovate. As of course would choosing the de facto
standard instruction set. 

re: Compatibility

I think Intel have learned their lesson on compatibility. The Pentium Pro
strayed from the strict compatibility that the "market" mandates, as a 
consequence they have had to introduce the Pentium II which restores 
full compatibility. Just as the PC standard slipped away from IBM's control
I suspect the Instruction Set standard has slipped from Intel's control.
I would be surprised if they make the same mistake with Merced.

Software rules!

Mark
 

    re: .12
    
>>    Manufacturing capability isn't the key here.
>    
>    
>If Intel decided to price Merced at $100 ea, in quantity?  They could, 
>and they'd be making a decent margin, comparable to our SYSTEMS margin, 
>if not better.  Add in what they are doing with PROCard, or whatever the 
>marketing term for the next gen processors comes out with (single edge 
>connecter or something?) and you'll soon see that they've got people 
>locked into their manufacturing process, which no one else can match.

    At $100 a pop, they'd get converts.  But that's not the issue.  The
    issue is that every existing PC would need a hardware AND SOFTWARE
    upgrade to get to Merced.  That gets painfully close to "throw the
    existing box away and start over" -- one of the same hurdles that Alpha
    faces.
    
    Then, there is the incompatible third-party software problem.  Now,
    granted, most vendors probably will jump on Merced quickly (unlike
    Alpha) since there is little doubt that it is the main desktop of the
    future (at least, unless someone tries to alter the picture).  But
    there will probably still be a good period of time until ALL the
    desired apps are running on Merced.  Heck, on the Internet, I just saw
    someone's list of drivers missing from WNT.  It was HUGE!  And WNT
    isn't exactly the new kid on the block!
    
    There is a distinct window of opportunity between the time of the
    release of Merced and the time it finally becomes suitable for every
    desktop in the business world.
    
>>    If Merced requires a migration, that means there is a potential 
>>    market for a Merced-like processor which doesn't require a 
>>    migration.
>
>Pretend you are a customer.  Digital tells you that it's easier to 
>migrate to an Alpha from your PP.  Inte ltells you the same thing.  Who'd 
>you believe?
    
    No.  The idea is NOT Alpha vs Merced.  The idea is some super
    x86-compatible chip vs Merced.  The idea is NO software migration.  Nor
    FX!32 incompatibilities.
    
    The choice needs to be between running your current software as-is on a
    new super-charged box, or facing a hardware and software migration with
    Merced.

    Why not lure them to 64 bit VMS? (re .9).
    
    We ought to make capital out of the abandonment of x86 code
    save as a new "compatibility mode" (whether formally a "mode" or
    not). Processes running the slow instruction set will tend to
    suck up the machine...remember how it worked on the vax 780?...
    so that customer sees no benefit till the x86 code is all, or VERY
    nearly all, gone.
    
    Push that, and point out that the application backlog for
    slow instruction set processors (validated by Merced having
    a faster instruction set) is meaningless, and that an
    alternative system that already has some fast instruction set
    apps (Alpha) exists. And even has some OSs that don't need to be
    rebooted all the time, and don't get broken into or remotely
    crashed all the time...
    

    Creating a partnership with AMD is an interesting idea.
    
    In an earlier entry, I criticized .0 without offering alternatives.
    Given that we already have Alpha and FX!32, if we had the $$$ required
    to develop (from scratch!) a world-class x86 processor, I believe the
    $$$ would be more effectively spent encouraging/assisting software 
    vendors to create native Alpha versions of their products.
    
    My 2 cents,
    Jim

    Re: .1
    
    >> Intel would probably sue us.
    
    They just did. 
    

    
    Intel threats to cutoff the supply of Pentinum to Digital.
    The news says it might happen after the third quarter.

    I know nothing about chip design :')
    
    but... :'D
    
    I remember reading heaps about the PowerPC 615 chip being developed by
    IBM - the idea was that it had the instructions for both the PowerPC
    and 486 built onto the one chip - delivering the "compatability mode"
    required for x86 apps. As far as I know, IBM have basically denied its'
    existence, and as it hasn't appeared so far, my guess is that they may
    have found it 'too hard' (or maybe they didn't want to tick off Intel?)
    
    Perhaps something of this nature in an Alpha chip for the PC class
    market? Useing something like Cyrix or AMD's technology (licensed of
    course ;'), built into a version of the Alpha chip giving the best of
    both worlds.
    
    As I said, I know nothing about chip design, but maybe... ?
    
    H

IBM already had the rights to produce x486 from a previous agreement with 
Intel.

Watching this thread and some others in this conference, I see a lot of 
people thinking that Alpha compatibility with x86 isn't good enough at
the current level.   Probably true....

However, I have a Cyrix 6x86 at home, and it's not 100% compatible with Intel
either.  Win95 doesn't detect it as a Pentium without some help (a little
shareware program fixes that) which causes many games to yack on install.  There
are also other little incompatibilities which are well documented.  K6 is
similar in that it's not exactly either a P5 or a P6.

In order to be truly compatible you'd need to clone 100% of the behavior of your
favorite Intel processor including all undocumented behavior and bugs (aka
errata).  Not an easy undertaking technically and more luxury automobiles and
vacation homes for the lawyers...

Note that various incantation of Intel machines don't even behave the same way
so an argument could be made that P6 isn't 100% compatible with P5.  Try
converting a floating point number to integer and compare the results between
the two....

  As an aside, we did partner with AMD to take their 486 design and
  get it fabricated at Digital's South Queensferry, Scotland (SQF) fab
  (now owned by Motorola).

  I went to a talk on the work involved -- it was quite something,
  we took a design in someone else's CMOS process and modified it to work
  on our CMOS process. Got great speed (for the day) and made a lot of them
  for AMD.

  -- John Faricelli

>  I went to a talk on the work involved -- it was quite something,
>  we took a design in someone else's CMOS process and modified it to work
> on our CMOS process. Got great speed (for the day) and made a lot of them
>  for AMD.

While I agree it was no small task, I'm not sure about the "great speed".  
SQF was making AMD's 486DX2 @ 80mhz and DX4 @ 100mhz.  At the same time, AMD 
was making the stuff at the same speeds in their own FAB's, as was IBM with 
it's blue lightning stuff and Intel.

-Bruce

    
    Some other not-quite-so-modest proposals I'd like to see implemented
    
    	- A decent employee purchase plan, (cost +5%) that covers ALL
    	  products sold by the company.  Toner cartridges to TurboLasers.
    	  Limit it to one of each item/month, and exclude
    	  service so that it won't easily undermine our resellers.
    
    	- Anyone with the title "Vice President" or "President" (pick
    	  relevant titles for your country) is REQUIRED to purchase
    	  at least one mid-range desktop computer OUT OF THEIR OWN money
    	  every year.
                                                        
    I believe these would have the following effects:
    	- Employees actually have an opportunity to buy competitive
    	  products, instead of end-of-life products.
    	- Employees are effectively encouraged to innovate
    	- Volume of products sold rises (economies of large scale
    	  reduce our costs)
    	- Enterpreneurial employees will figure out how to sell
    	  their 'old' stuff, and upgrade to the new.  A consequence
    	  of this is that we expand our mindshare through personal
    	  contact (arguably the most effective form of advertisement)
    	- Employees may use the products that they buy to innovate,
    	  and create new opportunities for Digital as a corporation.
    	- Vice Presidents very quickly get "we have to have affordable
    	  and competitive products" religion.  Our $50K workstations
    	  might actually have $50K worth of value, then.
    	- Vice Presidents very quickly find out what it's like to
    	  actually be a customer of this corporation.  (Customer Focus
    	  takes on a new meaning)
    
    

TALLIS::GORTON for President.