Conference 7.286::digital

    SPECint95 Intel 200Mhz Alder P6:  8.09 (box about $6-8k well loaded)
    SPECint95 333Mhz EB164            8.08 (box about $26k well loaded)
    
    Well, fastest performance workstation if not measured by SPECint95.
    ;-)
    

Are those the original or revised (downward) P6 numbers?

			Steve

    
    Intel only got the SPEC92 data way wrong.  The SPECfp95 data is what
    you need to compare on..  EV56 should be much better there..
    
    

    Yup, with SPECfp95, the EB164 @333 box is almost twice as fast
    (12.1) than the P6 box (6.75) for four times as much money.
    That's (fp) where the real price/performance win is for Alpha.
    Kratz

> Yup, with SPECfp95, the EB164 @333 box is almost twice as fast
>    (12.1) than the P6 box (6.75) for four times as much money.
>    That's (fp) where the real price/performance win is for Alpha.
>    Kratz

Run that by me again???

2X performance for 4X the money is a price/performance WIN???!!!

That would mean that for the price of an EB164 box, you could buy
4 P6 boxes, giving yourself twice the overall compute-power of the one
Alpha.

How is this a win for Alpha again???

    Note: The AlphaStation 500 is similar to the EB164, but it
    is different and slightly higher performance due to fast
    cache and wide scsi IO.
    
    It's not identically the EB164.
    
    					RJD

    
    re -.1
    
     Maintenance and serviceability of one box verses four?
    
     -Mike Z.
    

    re .5
    You're a great straight man.  I owe you $5.  ;-)  

It is pretty disgusting.  Pentium Pro/200s with 32Mb/2Gb/4-6x/Matrox 2Mb/
no monitor can be had for under 5K.

To paraphrase an old campaign item: "Where was Alpha?"

    I've been asking myself about the price/performance reason for choosing
    an Alpha that's twice as much performance for quadruple the money. 
    When would that make sense?
    
    I think it would make sense if you are faced with buying very expensive
    software.  The difference in cost means that, technically, you get the
    same bang with two Alpha's ($48K) as with four Pro's ($24K) as far as 
    horsepower goes.  But, you may need to pay twice as much to license the 
    software.  Thus, if it costs $12K per machine for software, it's 
    basically break even.  (Total cost being $72K for either setup.)  If it
    costs less than $12K for software licenses per machine and if you only
    consider software and initial hardware costs, then the Pro gives you
    more bang for the buck.  If it costs more than $12K for software
    licenses per machine, then the Alpha.
    
    Here in Networks, where we use Alphas running U*ix, a single machine
    may run software with licenses costing in the range of $10K to $100K 
    per license.  With software this expensive, it can make sense to get 
    extra horsepower on fewer machines, even at significantly higher 
    hardware prices.
    
    Steve

  It may also make sense if you have a very expense wetware resource
  sitting at the keyboard/monitor.

  If I cost the typical $100K/year that Digital budgets for an
  "engineering resource", then it's a good investment to make sure
  my time is as productive as possible. A computer that runs, say,
  CAD tools twice as fast may still be a very good investment even
  though it costs four times as much as a slower computer, because
  it optimizes my productivity.

  The cost of the hardware resource is amortized pretty quickly
  when it makes the $100K/year resource more productive.

                                   Atlant


  Of course, this is all hypothetical, at least as far as *DIGITAL*
  is concerned. Digital shows absolutely no signs of actually in-
  vesting in this way. Inseatd, we tend to try to minimize all
  costs, even at the expense of enormous losses of potential
  productivity. But that's another (current) notes string.

    
    rep .0
    
    
    And I betcha in a month we'll market an NT only version, and knock say
    10K off the price and call it competitive.  Then it would be twice the
    performance for say twice the money.
    
    Alpha is the Ferrari of CPUs.  A Ferrari cost twice as much as a Turbo
    Porche Carrera, and delivers 10-20% more performance.  200MPH vs 165MPH
    for the Porche.  Not a good bang for the buck.
    
    I would think about the only place a 4x price, 2x performance machine
    makes sense, is when the Application you are running is going to take
    say 2 months to get you an answer, if you run it on Alpha it's only 
    1 month.  Otherwise as Harley Davidson has said for years, "Keep it
    Simple".
    
    Alpha makes sense for OSF for new customers, and our existing VMS 
    customers, for NT, it makes sense as a Compute Server serving lots
    of Intel, AMD, read inexpensive and fast, NT workstations.
    
    Just my opinion only, but I know where I'd spend my Capital equipment
    dollars.
    
    /art

   >  Maintenance and serviceability of one box verses four?
    
    They're still going to need another box to run their games & all
    the other Intel stuff. 
    
    Greg.

    As the author of a position statement on Alpha SPEC leadership, I
    addressed only technical issues  (ALPHASTATION 846).  But it's obvious 
    that the internal community is begging to know what the pricing
    strategy is.  
    
    Do we intend Alpha to be a price performance leader?  In a niche, or in
    broad areas?
    
    Now that this is boiling over in the DIGITAL conference, this would be
    a very nice time for marketing/product management to jump in here and
    help us understand the pricing strategies.  We don't need to know every
    detail... just tell us please whether those who read the strategy as
    "Find a [small] niche" are reading it correctly!
    
         /john

    Maybe we should sell/license our floating point expertise to Intel?  An 
    Intel chip with Alpha FP performance would be quite interesting.


    							-Paul

    Well, I'm glad to see some people raising the real issues of how and
    why Digital, SGI, IBM, HP, and SUN are able to all grapple for a
    multi billion dollar workstation market rather than sit around thinking that
    five thousand dollar x86 based machines are the answer to everything.
    (Although they are the answer for most things; hence the high volumes
     but less margin).
    
    People buy these machines usually for some combo of the
    following reasons:
    
    1. Their OS or code to run (NT, OpenVMS, or UNIX)
    2. They can't afford the lost time when their lower performance
       machines take too long to finish a job or "fall over" under
       severe load.
    3. They have to deliver on a short schedule, money becomes less
       of an issue when your engineer already costs $65k a year or
       more and the software license for his/her design tool cost
       $20k or so. - Plus your customer is demanding output. You
       don't want to be sitting in front of low performance and
       waiting for computes.
    
    	So, the market is there. Part of it is ours to be had.
        These customers would be rather bummed if we tried to
        sell them P6 based machines to do the job. They are well
        aware of the cost/performance trade-off and nobody is making
        them pay 4x for 2x performance. It makes sense for them based
        upon their cost and time constraints. The market can bear it.
    
    	Also, besides none of our competitors being able to offer
    	performance as high as the AlphaStation500/333 on the desktop,
        they can only offer less performance for a higher price - and
        they still hold a bigger chunk of the market. Things may change,
        but that's the way it is now.
    
    
    

    re: Note 4368.15 by PERFOM::HENNING
    
    > Now that this is boiling over in the DIGITAL conference, this would
    > be a very nice time for marketing/product management to jump in here
    > and help us understand the pricing strategies.
    
    It would be nice, but why should they?
    
    Ken....
    

    Re: .15

>   It would be nice, but why should they?
>    
>   Ken....

    Perhaps so that our sales reps can respond in a consistent, if not
    plausable, way when asked why our Alpha systems price/performance
    numbers are half that of Pentiums.  If we can explain to a customer
    why they should think of us and not watches when they here the word
    digital we should be able to explain our pricing.

    The previous replies not withstanding, I can't explain why our Alpha
    systems cost so much.  If those who set the prices can't explain them,
    who can?

    Bob

Paul:

> Maybe we should sell/license our floating point expertise to Intel?  An 
> Intel chip with Alpha FP performance would be quite interesting.

  I thought we did. But it was pacakged in the convenient form
  of Dr. Dileep Bhandarkar. :-(

                                   Atlant

    When I was working on the technical position paper, I consulted with
    several business people.  They seemed to think that I was asking them
    to "pre-announce" the Q3 price reductions, which of course they
    couldn't do. 
    
    But I hope that it ought to be possible to say "our strategy is to win
    CAD/CAM and Mumble/Niche" --  or -- to say that our strategy is much
    more broad-based that and that we will price accordingly (over time,
    plus or minus the fluctuations of the market, etc.)
    
    Why should they?  To help all the troops know what the targets are.

    
    
    re -.1
    
    Targets.
    
    I view everything in life as a target. That makes it easier
    understanding, people, animals, digital..... Of course a few
    years ago, something happended to give me a shooters view of
    the world. '^)
    
    -Mike Z.

    
    
    Congratulations on the AS500.  I'll offer the following observations
    in the context of the previous replies:
    
    1.  I suspect that we will be publishing SPEC numbers for the AS500
        for which the SPECint numbers are marginally better than the
        8.09 of P6.  It just requires a 1% improvement in any of the
        eight SPECint benchmarks.  Marketting should be able to claim
        the fastest workstation in the world.
    
    2.  No Windows NT Alpha workstation is going to be broadly competitive 
        with a Windows NT Intel solution until FX!32 ships.  If you can't
        be broadly competitive, the only recourse is to identify viable
        niche markets and make as must profit there a possible.
    
    3.  The AS500 will not be broadly faster than P6 platforms.  "Up to
        2 times faster on some floating point intensive applications" is
        not broadly faster.  On the other hand, EV56 is broadly faster 
        than than P6.
    
    4.  Maverick has been in the pipe for a long time.  Its a cost
        reduced Alcor.  I suspect that it was designed to compete
        with the products of Sun, HP, SGI and IBM -- not Intel.  I suspect
        that beating out those competitors will be a healthy business.
        Selling Maverick in the PC market place would probably require
        selling it below cost.
    
    5.  Any movement of customers from the proprietary software
        environments of Sun, HP, SGI or IBM to the Windows NT environment
        is good for Digital, even if it is to Intel processors.  In the
        future, we can sell these customers Alpha processors, when we
        have a solution that is broadly competitive with the Intel
        solutions.  So, sell those Sun, HP, SGI and IBM customers
        Windows NT P6 sytems, if that is what they want,  -- and sell 
        them Alpha systems in the future.
    
    6.  I look forward to a high volume, low cost ev56 system that includes
        FX!32.  We just don't have that product yet.  Based upon the FX!32
        announcement, this can't happen before some time this summer.
    
    
    Hugh
    

    Sigh.
    
    One, Mr. Kratz estimates the performance of the Maverick based on the
    EB164.  Out on a limb on several counts there.  Maverick beats Pentium
    Pro 200MHz.  Let me repeat that.  Maverick beats Pentium Pro 200.
    No matter which SPECfoo9x you pick.
    
    Two.  Mr. Kratz, who until two weeks ago was shouting SPECint92 to the
    world, quietly ignores it now.  He does this because Intel has admitted
    to a minor faux paux.  They claimed that they had a bug in a compiler
    (yes they very much did) that resulted in overstating SPECint92 by 10%,
    and they have just corrected their numbers.
    
    They relased their new numbers today.  SPECint92 for the 200MHz PP
    is 318.4 *NOT* the originally claimed 366.0.  (By the way, Kratz,
    that's about ****15%****, not 10%.  Maybe they used an old Pentium for
    the math?)
    
    Three, Intel now has to answer some other questions, such as how do
    they do so well on SPECint95 while doing relatively poorly on
    SPECint92.  This is a question that Mr. Kratz would not like to have
    asked.
    
    Four, a *well* configured PP200 system is not the same as a PC.
    A *well* configured PC has 32MB of memory.  A *well* configured PP200
    has 64-128MB of memory.  And to even hope to catch up on floating point
    performance a PP200 would have *4*, count them *4* PP220's in them.
    That assumes that you have an application that is relatively
    parallelizeable, *AND* the ISV parallized it.
    
    Argue that floating point is unimportant all you want.  But don't argue
    that people who need the floating point of Maverick can get by with a
    5K 32MB PP200.  They can't.  Price out a "well configured"
    (64-128MB memory, *not* 16 or 32MB) four banger PP200, it'll be
    costing out at a higher price than a Maverick.  (Hint, a 64MB
    *TWO*xPP200 from Intergraph (minus the graphics) costs out at $23K,
    monitor not included.)
    
    -----
    
    If you want to run Excel, buy a Pentium and have fun with Windows 95.
    (And if price/performance is what you want, buy one from Dell, they
    just dusted everyone on the latest "hot" Pentium 166MHz review in
    PC Magazine - for a *lot* less money than Digital's 166MHz Pentium.)
    
    								-mr. bill

    Bill,
    Why should *I* have to explain why P6 does better (relative to Alpha)
    with the new SPEC95 suite again?  Isn't that Alpha's problem?
    
    I'm glad Maverick is faster than a $5k P6 box (which doesn't need
    near the memory that Alpha does).  The purpose of my entering notes
    in this string was to expose some reality to Alpha's "price/
    performance leadership" claims.  98% of this company is brainwashed
    into thinking that Alpha has such leadership.  Every once in a while
    I like to be the sheep in the pen that lets out a growl.  It isn't
    real popular with the other sheep, but it does wake 'em up.  The
    more people understand where Alpha really stands, perhaps there
    will be just a slight bit more pressure to improve.  I think that's
    an area that we'd both like to see.  I think Alpha is a great
    product.  I also think that Alpha's worst enemy is not Intel, not
    SGI, not Sun... but ourselves.
    .02 Kratz

    P6 does better (relative to *EVERYBODY*) on SPECint95 vrs SPECint92.
    
    I believe *all* their benchmark data deserves closer scrutiny.  This
    might be nothing but FUD, and their results could be just fine thanks,
    but given they've admitted to being wrong once after they got caught,
    it may be only a matter of time before they admit to being wrong again.
    Do you realize how rare it is for a vendor to pull SPEC results?
    (Doing my best Senator Al impersonation "It's the paaaatearn that
    this enginear is concearned about.")
    
|   I'm glad Maverick is faster than a $5k P6 box (which doesn't need
|   near the memory that Alpha does).
    
    Sigh.  Oddly enough, Intel doesn't run SPEC and BAPCo on $5K 32MB
    Pentium Pros.  But you know better than Intel.  (And oddly enough,
    Intel keeps using 1MB L2 caches with their Pentiums when doing
    performance testing, but you know better than Intel.)
    
|   The purpose of my entering notes in this string was to expose some
|   reality to Alpha's "price/ performance leadership" claims. 98% of this
|   company is brainwashed into thinking that Alpha has such leadership.
    
    Stuff and nonsense.  I can tell you flat out that you are quite wrong,
    and that people know all too well *exactly* where we stand on performance,
    price/performance and price.
    
    								-mr. bill

    Well, I was glad to see somebody finally slap Kratz around. The
    attitude is unbearable.
    It was also nice to see Kratz issue SOMETHING positive about Alpha.
    
    Again - our competition is selling lower performance for 20k,30k,40k
    and $50K and making a bundle. We are offering higher performance in
    that range for less money. I can't do much else right now. I do not
    control prices of EV chips or systems. 
    
    We do have protos running at high freq with EV56. They're next.
    
    					RJD
    
    

    Hey, I'll get in line to slap Kratz around, too.  But then I'll go buy
    him a beer (if he promises not to bicycle afterwards) since it actually
    *is* useful to have someone really challenging Alpha - hard - on pricing.
    
    The incremental cost of making 500,000 chips vs. making only 100,000 is
    peanuts, compared to the upfront investment required to make the first
    100,000.  Lots of us in engineering would love to know why we don't
    lower the price and make it up in volume.
    
    But maybe that's too naive of a business perspective....
    
    		/john

    well John, maybe it doesn't take more than, say 10,000 employees to
    make chips...
    
    Mark

    rep .28
    
    HLO has lowered the price of their CPUs, if you run NT they will
    sell it to you for a lower price than if you are putting it in an
    OSF/VMS platform.  What is that?
    
    Alpha is great for OSF/VMS, it can beat Sun/HP in the price performance
    arena.
    
    But for NT, it's can't compete in the price performance arena.
    It can compete and win in the performance arena, but at a 2x price.
    Alpha makes sense for NT as a server, for lotsa Intel/AMD NT boxes,
    and for those Niche applications where you gotta have the speed man,
    like CAD/CAM, but not for everyones desktop to run MicroSoft Office
    etc.
    
    And the market bears this out, in the 3 years that Alpha has been out
    with NT being out for 2 years, we don't dominate the market, not even
    close, so what does that tell us.   
    
    If Alpha had the volumns we need for success, would we have just
    rented/leased/whatever Fab 6 capacity to Cirrus logic, I don't think
    so, but what do I know!  I know that if I was the Capital Equipment 
    manager for a group/division, I'd put the Alpha computes on select
    desks, and the rest would have cheap/fast Intel/AMD boxes.
    
    And as far as getting more out of an employee, if I give you the answer
    before you ask the question, does it help you, or does it just prepare
    you for a stint on Jeopardy.  Humans need time to digest answers, we're
    not computers that respond immediately to stimulus.  And with machine
    approaching supercomputer speed/throughputs of 5-10 years ago, they are
    approaching "silly" fast for most normal everyday compute stuff.
    Everybody doesn't need to compile their code on their own machine, they
    can use a Server, etc.
    
    As for bashing Kratz, at least he has the guts to call them the way he
    see's them, even if he has PCBU colored glasses on!  And I don't think
    that the individuals who would stand in line to "slap him around" would
    be so bold in person, aren't notes great, we can all be Mike Tyson
    without the body shots, just cheap shots!  All the venom directed at
    him must mean he's preety close to the mark!
    
    But Kratz, I don't like the sheep analogy, perhaps we could all be
    sharks and have some blood thrown in once in a while just to keep
    most of us sharp and well fed. 
    
    
    /art
    

    I tell you what gets me excited ...  It's seeing an Alpha 3D graphics
    workstation package sell for under $4K.  I see this advertized
    frequently in my 3D rags.  It makes me feel like someday I may get to
    take that leap from my Pentium to an Alpha.  Someday ...
    
    Steve

> And as far as getting more out of an employee, if I give you the answer
> before you ask the question, does it help you, or does it just prepare
> you for a stint on Jeopardy.  Humans need time to digest answers, we're
> not computers that respond immediately to stimulus.  And with machine
> approaching supercomputer speed/throughputs of 5-10 years ago, they are
> approaching "silly" fast for most normal everyday compute stuff.
> Everybody doesn't need to compile their code on their own machine, they
> can use a Server, etc.

  No, it's not that you can calculate an Excel spreadsheet in
  0.001 seconds instead of 0.1 seconds. You're right; that's
  not interesting anymore.

  It's that the speeds of modern computers let you do things in
  ways that were completely infeasible before. The "Graphing
  Calculator" on the Apple PowerMacs is a very good example
  of this. Whereas the old "Calculator" applet was just that
  (a calculator or scientific calculator), the "Graphing Cal-
  culator" will redraw *IN REAL TIME* the equation you're
  manipulating, even if the result is a 3D graphic. This
  allows you to look at things in completely different ways
  than before.

  Virtual Reality is another example of this. For example, the
  ability to manipulate, via a data glove, a molecule *IN REAL
  TIME* allows chemists to do things they could never do before.
  "How will molecule A bond to molecule B?" "I don't know; let's
  push them together and watch as it happens!"

  It's these (excuse me) "Paradigm Shifts" that really suck up
  the available computes, not the ability to do old stuff faster.

                                   Atlant

    Re: would be so bold in person - 
    
    Um, I should point out that Kratz & I used to be in the same group,
    he's fun to be around and to talk to, and the only kind of slapping
    referred to here is dialectical not physical. 
    
    But the beer is physical, not metaphorical, on the day that Megaphone
    Kratz (cf. 4031.9 in this conference) succeeds in getting Alpha pricing
    to be truly aggressive.
    
    	/john

    I did read somewhere that Palmer asked Caldwell to explain the poor
    SYSmarkNT scores, so it (4031.9) may have got thru.
    
    Bad news for Alpha tho: it looks like that new "Windows NT" magazine
    standardized on using SYSmarkNT.  The good news for Alpha: somehow
    they butchered up the test on the HP Pentium Pro 150 in the Feb issue
    and got a really poor (425 as I remember; shoulda been @500; we get
    a 504) score.  Regardless, that's one rag that Karen Q. in PR should
    probably keep Alpha away from (at least the current Alpha XL's which
    are in the lower 300 range; EV5/EV56 is needed there).
    Kratz
    
    P.S. I can't ride after beers either John; hills become pure hell!

    I can't believe that _anybody_ is justifying Alpha price/performance. 
    Sure we're still the fastest, but every time our margin over Intel
    erodes we just come up with more "justifications": 
    	
    	1. We don't need to be twice as fast, 50% is enough 
    		(when Alphas first shipped)
    
    	2. We don't need to be 50% faster, 30% is enough
    		(after recent Intel TPC C results)
    
    	3. We can eke out a few % faster than Intel in SPECxxx
    		(earlier in this note)
    
    	4. We'll put an Alpha on every Power User's Desk
    		(Jensen timeframe)
    
    	5. We'll target Alpha Workstations towards very specific niche
    	   markets where they need the fp performance (cad, etc.)
    		(XL Dream Machine announcement)
    
    The proof is in the pudding.  Alpha workstations are not selling well. 
    We continue to lose market share in workstations.
    
    Thank goodness are server sales are doing well.
    
    Speaking only of workstations:
    
    	1. In the UNIX arena we have too little market share and still miss
    	   key applications
    	2. There is no VMS desktop market worth talking about
    	3. In the NT arena our desktop marketshare is dropping.  Our
    	   pricing is too high v. Intel, which absolutely is the competition.
    
    When our certs indicate an increasing market share then I'll know we've
    hit the right combination of performance, pricing, marketing and sales.
    
    We may believe that we have machines so fast that people will beg to
    buy them, but this is NOT happening.
    
    The only way to ensure mind boggling price/performance is to ship
    machines with the latest chips sooner than we have been and at prices
    lower than we have been.
    
    If we can not do that, then we'll have to fix our marketing or our
    sales.  
    
    Regardless of the fine quality of the individuals in all these
    organizations, something is not working.
    
    End of Soapbox.  Go ahead beat me up.  I want Alpha to succeed (and it
    IS in the server marketplace, so I know it can be done).  Let's stop
    fooling ourselves.  You can not win if you believe your own propaganda.
    
    Ken B.
    In the UNIX arena
                                    

|    	3. We can eke out a few % faster than Intel in SPECxxx
|    		(earlier in this note)
    
    Sigh again.
    
    Alder PP200      - SPECint92 of 318.4
    Maverick EV5/333 - SPECint92 of over 400.0
    
    That isn't a "eek out a few percent faster".  You don't even want me to
    talk about how badly we beat a uniprocessor P6 200 at FP.
    
    
    Compare us fairly.  A >$5K Micron PP200 competes (I think rather poorly)
    against <$4K P166s.  It does *NOT* compete against Maverick.  If we tried
    to sell a Maverick to a Micron PP200 customer, we'd be wasting our
    time.  If Micron tried to sell a Micron PP200 to a Maverick customer,
    *they* *ALSO* would be wasting their time.
    
    
    A 23K Intergraph 2xPP200 competes against a 23K Maverick.  Now, ask
    yourself if 30-ish% at int and fp is enough to move someone from
    Intergraph NT to Alpha NT (especially considering the horrible industry
    practice of quoting Unix performance for Windows desktops) and you've 
    got a very valid question.

    
    But ranting about 5K Microns leads to INACTION.  People rightfully
    tune you all out because you keep asserting (incorrectly) that
    nobody will buy Mavericks because everyone will be buying 5K Microns.
    It leads to INACTION because the message sounds very much like we
    are wasting our time until an EV56 at >433MHz is in a $5K box.
    Assuming that Intel doesn't come out with a PP233 with 512K L2 cache.
    
    (And if you got your wish, I suspect that you would be right back
    saying we fail since it isn't at $4K or $3K, or perhaps join the
    search for the lochness market which is out there somewhere at $0.5K.)
    
    								-mr. bill  
    

    I find it interesting that for years we (DEC) were "beaten up" in the
    trades for not having RISC, when our CISC VAX must shurely be out of
    headroom. Now that we have ALPHA, we are debating performance with a
    CISC (Intel) chip. I know I'm facing the wrong way (looking backwards)
    but I wonder where a VAX chip would be on the performance curve with
    todays chip feature size.
    
    Rob Wall
    [and could NT be made to run on it?!?]
     

re: .37

There was a time when when it was an industry truism that a RISC architecture
would always be at least twice as fast as a CISC. When the fist MIPS systems
came out (e.g. DECstation 3100 et al) it looked as though that was true.

As there were people within Digital who wanted to build RISC systems, that
message was promoted all over the Corporation in order to ensure that the
RISC program went ahead.

However, the likes of Intel decided that that truism was flawed and put
a lot of effort into their own architecture. THe result was the performance
gap between RISC and CISC was in practice much smaller than expected.
Each organization was driven the outcome they wanted and both succeeded!

Could faster VAX systems be produced?

Well we have continued to put out ever faster VAX systems!
VAXen now are the best they've even been. I'm sure the VAX teams could
do even better if they so desired. The supposed defect in the VAX
architecture was the small (512 byte) page size.

M  

> The supposed defect in the VAX architecture was the small
> (512 byte) page size.

  Actually, that's not the source of much of a performance penalty.
  Rather, it implied a memory penalty, both in main memory overhead
  (to hold the page tables) and translation-lookaside-buffer size.

  And, as a second-order-effect, bigger memories mean slightly
  slower speed.

 -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

  The progress that both Digital and Intel made at making CISC
  architectures take advantage of RISCy techniques is really
  very striking, viewed in retrospect. And the business benefits
  of *NOT* forcing *ANY* software conversion were equally striking.

                                   Atlant

    >However, the likes of Intel decided that that truism was flawed and put
    >a lot of effort into their own architecture.
    
    Did they decide it was flawed, or did they decide that their installed
    base was too large to abandon?
    
    >Could faster VAX systems be produced?
    >
    >Well we have continued to put out ever faster VAX systems!
    >VAXen now are the best they've even been. I'm sure the VAX teams could
    >do even better if they so desired. The supposed defect in the VAX
    >architecture was the small (512 byte) page size.
    
    You are right to point out that faster VAX systems have been 
    continually produced.  I've often wondered where the VAX architecture
    would be today had we decided to invest the $$$$ we invested in
    Alpha on VAX hardware and compilers.  Intel sure has done wonders
    for x86.  They started with a simpler instruction set, though.
    
    The 512-byte page size is not the major defect in the architecture,
    though.  It is a relatively minor one, and probably one of the
    more easily fixed problems (especially now that OpenVMS has the
    pagelet concept).  
    
    The VAX has a variable length instruction set.  This conserves
    memory, but makes multi-stage pipelining rather difficult (you can't
    grab the instruction four instructions down the line without
    interpreting the previous three instructions).  This coupled with
    an instruction set that wasn't designed to promote pipelining, means
    that pipeline stalls are more likely, and we can't maximize the
    use of the CPU's resources.  Also, since the instruction set is more
    complicated, it becomes difficult for the compiler to optimize to
    take advantage of the processor's pipelining capabilities (the way
    RISC compilers can).
    
    My 2 cents,
    Jim

    > where the VAX architecture would be today
    
    same as always, 32 bits.  Maybe the question ought to be:  Where will
    VAX computers be in 10-20 years?  Can you say, "PDP-11"?
    
    Mark

> same as always, 32 bits.  Maybe the question ought to be:  Where will
> VAX computers be in 10-20 years?  Can you say, "PDP-11"?

  How many bits is that ugly Intel architecture again?

                                   Atlant

    Anyone see the advertisement in PC Week 15-Jan?  Thanks, Steve Lionel,
    for showing it to me.  This makes my day for exposing the fact that
    many of the competitors' claims are for systems that are not in the
    general market or don't exist yet!
    
    Digital 
    Semiconductor                Just a reminder that the 333 Mhz Alpha
                           is simply the fastest processor that you can
    buy, period.  Other chips may
    quote estimated performance              W A R N I N G:   C H I P S
    and theoretical speeds,                           C O M P E T I N G 
    but only Alpha has the                          W I T H   A L P H A
    real-world, in-system numbers that                 A R E   M U C H,
    really count.  And every time the           M U C H   F A R T H E R
    other guys even get close, the next               A W A Y   T H A N
    generation of Alpha chips makes              T H E Y   A P P E A R.
    them play catch-up all over again.
    For the chip specs, visit http://www.digital.com/info/semiconductor
    or call 1-800-332-2717 ext. 333.  And see why, as always, Alpha 
    pretty much has the road to itself.

    There is an preliminary web site to go to for AlphaStation 500:
    http://www-wbs.eng.pko.dec.com/high_perf/hp.htm
    
    	It's under construction.
    
    				RJD

    re .36
    
    I don't mean to imply comparing radically unlike boxes (Micron to
    Maverick).  You make a valid point.
    
    as for:
    
    
    >    A 23K Intergraph 2xPP200 competes against a 23K Maverick.  Now, ask
    >    yourself if 30-ish% at int and fp is enough to move someone from
    >    Intergraph NT to Alpha NT (especially considering the horrible
    >    industry  practice of quoting Unix performance for Windows desktops) and
    >    you've got a very valid question.
    
    _That_ is much closer to my real question.
    
    As you well know, the only thing Alpha has going for it is performance
    (including large memory access).  In all other respects it's an
    inferior computer (non-standard, doesn't run all the software, 
    requires another set of binaries, etc).  So it's performance or at
    least price/performance is what counts. Period.
    
    Lest it be thought that I'm proposing that we cut prices and sell at a
    loss, not at all.  Companies have been making money (lots of money) in
    the commodity market for a long time.  They do so by becoming
    efficient producers and sellers. 
    
    This is what we MUST do.  We must learn how to minimize manufacturing
    and selling costs.  This allows us to sell at lower prices and still
    make money.  Of course we need to do this and still allow a decent
    investment in research and development.  I didn't say it was easy.
    
    BTW - it may not make sense to compare a Micron and a Maverick, but it
    certainly makes sense to compare a Micron and a Celebris XL 266 and
    it's not a pretty picture (except in a few niches markets, but is that
    enough to survive?)
    
    Summary:
    	reduce manufacturing costs to allow us to sell Alpha boxes that are
    X percent faster than the competition (whatever that is in the
    particular market) and sell for Y % less money, yet still make money
    for us.  As X goes up, Y can come down.
    
    
    ken

    
    PC Week, what volume number what page, what add?
    
    -Mike Z.

    Well, I'll side with Kratz.
    
    Re: .27
    
    >our competition is selling lower performance for 20k,30k,40k and $50k
    
    And how about the 'sweet spot' at under $10K on the desktop?
    We offer an AlphaStation 200 4/166 at $8500 US, but it doesn't
    include a monitor ($1700 for 17") or a keyboard ($75).  I can
    understand not including the monitor, given that it's a high price
    item, but a keyboard?  (By the way, the keyboard is mandatory)
    HOW INCREDIBLY ASININE!
    
    >We do have protos running at high freq with EV56.
    Great!  But so does Deskstation.  And Deskstation was shipping
    an Alcor equivalent last march.  We ANNOUNCED Alcor in June.
    Deskstation's box cost $15k vs. $30k for Alcor in equivalent
    configurations.  I'm willing to bet that they included a keyboard.
    My point is that protos do not a system make.  The PC industry is
    shipping boxes almost as soon as new chips are announced.
    The non-PC computer systems builders typically do the pre-announce
    game.  Just because they (Sun, HP, SGI, IBM) do it, doesn't mean
    that we can be lax about improving our turnaround time for new
    systems.  Time IS money.  If we had been able to sell Alcor systems
    three months earlier, how much more money would we have made?
    
    Re: .36 with a subject of "Play the cards dealt."
    
    No.  I don't want to play the cards dealt.  I want us to
    improve the deal.  Especially since we get to look at the cards
    we are dealing ourselves.  If we aren't capable of doing that, we
    ought to get out of the desktop system building business.  Period.
    
    >Alder PP200      - SPECint92 of 318.4
    >Maverick EV5/333 - SPECint92 of over 400.0
    >
    >That isn't a "eek out a few percent faster".  You don't even want me
    to
    >talk about how badly we beat a uniprocessor P6 200 at FP.
    
    Really?  And will I be able to buy one of these Maverick systems
    before the end of CY '96?  I can go buy a desktop PP200 from a number
    of system houses right now.  Is Maverick a desktop box?
    
    Intel has done a wonderful job of boosting their performance,
    given how badly constrained they are by the errors of their
    forbearers.  They can't fix the really brain-dead stack-oriented
    floating point architecture they have, due to backwards compatability.
    I'm surprised that they didn't add some new floating point
    instructions to the P6 to remove the top-of-stack bottleneck.
    
    
    Bottom line - some of us believe that we, as a corporation are
    badly 'in need of a clue' when it comes to timely delivery of
    systems, which, in my opinion, is a fundamental reason we aren't
    gathering market share on the (UNIX) desktop at a fast rate.
    

>    Really?  And will I be able to buy one of these Maverick systems
>    before the end of CY '96?  I can go buy a desktop PP200 from a number
>    of system houses right now.  Is Maverick a desktop box?
    
    Please read the basenote again - Maverick is a desktop box, you can
    order it today, first deliveries 16-Feb-96.

        ....and then there's the approach that we are competing not with Intel
    but with Sun, HP, where we have a much better price/performance and
    technology advantage.  Although I agree with this point, I don't think
    we can afford to be complacent about it.

    Unfortunately, we currently miss out in a segment of this competition -
    the large sale, low cost market for UNIX workstation clients
    (100's-1000's of <$2k seats, local disk for swap space and root fs
    only, everything else nfs-mounted).  "Adequate" performance meets a
    sanitary requirement, while low price, quality, and maximum software
    availability are the prime factors. Sun and HP eat this one up (we only
    meet the quality requirement), unless Windows is acceptable (then Intel
    wins).  Apparently, we can not match their prices and still make an
    affordable sale.  Buying such business is a prohibitively large
    investment in any account.

    Unfortunately, the use of these machines on the desktop often dictates
    the choice of file and backend servers where we normally have the far
    superior and more cost-effective product.  The same architecture as
    that of the desktop is often chosen so that only one software
    environment needs to be developed for/under and supported (this latter
    is a significant cost in large organizations, where formal support
    tools and procedures need to be developed for each different platform). 
    Also, the large volume of these architectures in the low-end market
    also influence the availability of a customer's current applications
    and development tools from software vendors.

    I have to admit I don't know the size of this market in this day and
    age, but currently I am engaged at a customer who fits this
    description, and the lack of a Digital presence in a VERY large rollout
    under way (>5,000 HP seats plus 100's of mostly HP and a few Sun
    servers) is, well, um, it makes me feel like something is missing.  Is
    this a market that we should/could go after with Alpha, with a
    bargain-basement product that's at least as good as theirs in the same
    price range and is at least marginally profitable?

    (Of course, there is bright news too - currently, DECmessageQ is being
    used on most of those new machines and is the raison d'etre for my SI
    engagement in the account; it continues to come out on top in the
    customer's evaluations of MOM, and also the sales rep has successes
    with large UNIX servers in other divisions where the large volume UNIX
    desktop is not currently a factor.)

Re: .46

January 15 (Vol. 13) , the "PC Week Inside" section (probably not in all 
copies), page A4.

				Steve

    
     Is there a notes conf. on the ALPHASTATION 255? I'm looking for a
    conference that I can direct technical questions to.
    
    thanks,
     Bob Corrigan

    
    rep -1
    
    wrksys::alphastation
    
    /art

    
    Thanks Art.
    
    Bob

    So we want the server business and we figure we either don't need the
    desktops, or the desktops will follow.  
    
    A server company is a minicomputer company that's been EMPOWERED and
    RE-ENGINEERED.  You will sell 10-20 desktops per server (traditional
    number in the commercial space is 8-16 seat clusters.  When you sell
    the seats AND the servers, you get that magic feeling called account
    control.  With account control, your cost of sales drops and your
    margins increase.
    
    If I recall, a company out of Washington state went straight for the
    desktop, and now they dictate the computer business.
    
    Sorry for the rambling, but its late.  Here's the real point:
    
    A unified product line from the desktop to the computer room is THE
    goal if you are in the hardware business.  It used to be called
    MicroVax - Vax ####, or MV2000-MV10000.  Scalable OS, unified
    instruction set, common networking, layered products and applications.
    
    Just because these boxes now produce pretty pictures and run at 200+
    MIPS doesn't mean that the goal for that desktop to computer room
    product range should be abandoned.
    
    $0.02
    
    gc

re: Note 4368.54 by PMRV70::CROSBY



    >So we want the server business and we figure we either don't need the
    >desktops, or the desktops will follow.  

    This seems to be the strategy with Alpha. We develop CPU's (EV4, EV45,
    EV5 and EV56) with servers in mind, these CPU's are also placed in high
    priced workstations.  When we come out with a faster CPU, we use a
    shoehorn to put the previous generation CPU in a low cost desktop. We
    have not and will not win on the desktop with this strategy.   We need
    to design an Alpha CPU for the desktop if we want to be taken seriously 
    in that market

>    If I recall, a company out of Washington state went straight for the
>    desktop, and now they dictate the computer business.

    Microsoft only dictates the desktop for now.  They are just aother
    player in the rest of the market. It will be quite some time before the
    business world trust Microsoft for all it's mission critical computing.
    
    -Bruce

RE: .35

>>>        2. There is no VMS desktop market worth talking about

The last time I checked, 45% of workstation sales were with OpenVMS.  

RE: .42

>>> How many bits is that ugly Intel architecture again?
  
Intel has successfully managed the transition of the x86 market from 8 
bits, 16 bit and on to 32 bits.  With this kind of record, they should be 
able to handle the transition from 32 bit to 64 bit in the P7 timeframe.


							-Paul

    
    Re: -1
    
    |>>>        2. There is no VMS desktop market worth talking about
    |
    |The last time I checked, 45% of workstation sales were with OpenVMS.  
    
    45% of what sales? 
    All industry workstation sales or DEC's workstation sales. 
    
    jc
      
    
    

>>>    45% of what sales? 

Approximately 45% of Alpha workstation sales at Digital are for OpenVMS.

						-Paul

    re: -1
    
    Thanks. Kinda thought so.
    
    Now, the next question is this: 
    What % of all workstations are Alpha based? 
    
    Thanks,
    jc 

> Intel has successfully managed the transition of the x86 market from 8
> bits, 16 bit and on to 32 bits.

Have they indeed?  How come I've been seeing articles in the trade press
for over a year about the perils and pitfalls of managing the transition to
32 bits?  (See a recent "Datamation".)  How come I have to choose between
"32-bit disk access" and "32-bit file access" on my PC?  What's this win32s
nonsense, anyhow?  And how come I have to choose between 16-bit and 32-bit
apps sometimes?

The hype here has definitely outrun the reality, and at my site, where
we're trying to work with the stuff, we're only too aware of that.  Intel
and MS definitely don't have the act together.

___Pete

P.S. Yes, I know that this may not be how the great public market sees it.

>>>>    45% of what sales? 
>
>Approximately 45% of Alpha workstation sales at Digital are for OpenVMS.

since most INDUSTRY workstation sales are in the UNIX market (as I recall a
number of years ago something like 85% of the units sold, though I don't know
any recent figures), this would make me worry.  It implies that all these
people who want to run UNIX *aren't* buying ALPHA, otherwise VMS would be a
much smaller percentage.

-mark

    
    rep, last few
    
    Quote:
    
    "What statistics reveal is subjective, what they conceal is vital".
    
    /art

> since most INDUSTRY workstation sales are in the UNIX market (as I recall a
> number of years ago something like 85% of the units sold, though I don't know
> any recent figures), this would make me worry.  It implies that all these
> people who want to run UNIX *aren't* buying ALPHA, otherwise VMS would be a
> much smaller percentage.

Digital has a very small percent of the UNIX workstations market, too small
to justify the investment in UNIX workstations.  We basically have two
choices at this point:

1. Get out of the UNIX workstation market.  This would allow diversion of
resources towards support UNIX servers and NT workstations, which both
seem to be doing well.

2. Get serious about gaining marketshare in the UNIX workstation
market.  To gain market share we have to have some clear
differentiators.  Our current 10-25% price/performance advantage is
not enough to offset the relative paucity of applications and the pain
to migrate off another UNIX platform.  The only compelling
differentiator I see is better price and hence price/performance.  We
must get to a 50-75% price/performance advantage.  This implies
shrinking of margins.  But if we are serious about UNIX workstations,
we have to endure these poor margins until we have a significant
market share.  With significant market share will come applications,
and this will result in further growth in market share.  A year ago
when we were losing money, this was not an option.  Having reported a
$148M profit for Q2FY96 today, this route may be an option again.

Based on our current workstations strategy, I believe we have implicitly
chosen option 1 (get out of the UNIX workstation market).

Ram Rao
UNIX Consultant, Indianapolis, USA

> Based on our current workstations strategy, I believe we have implicitly
> chosen option 1 (get out of the UNIX workstation market).

Which one might argue is stupid, since if you produce competitive NT
hardware that can also run Unix (just like x86...), the primary cost
(beyond server support that you're already doing for Unix) doesn't seem
like it should be a large incremental cost.  (Of course, if you divide
up overhead incorrectly and artificially "punish" volume systems relative
to people-intensive "high margin" systems [see previous note on Compaq],
you might think something is unprofitable.)

Further, you're going to lose non-server only apps if you don't have
the volume of systems (*including* workstations) to justify having your
platform be a porting target.  As usual, Digital still seems to think
hardware is what matters.  (i.e. cutting out workstations may have
"unintended consequences" on all Unix products on the platform.)

Digital also seems to think it can just wander in and out of markets
and people will still take it seriously.  One of the criterias to being
a believable supplier is to have sustained presence.  (This is the thing
that irritates me when I see a total absence of advertising in places
like PC World for well over a year, for example.)

Franchises don't come easy, and reputations that make/break franchises
can be broken by actions in other areas.  (e.g. if you back out of Unix
workstations, why should we take your server offerings seriously?  How
do we know you won't dump those too?)

re Unix workstation market,

things could get interesting once Linux for Alpha gets a foothold.  Now all we 
need to do is get the hardware prices down...

Chris.

Thought writers of this note might appreciate another metric - Sun uses
    cost per rendermark in discussing their role in "Toy Story"; funny
    how we never seemed to get same press out of Johnny Mneumonic. What's
    interesting near end of note; they never estimate how many
    Alphastations this would have taken (assuming software available).
----- Begin Included Message -----

Sun and "Toy Story" --  Some Facts
 
In creating Disney's newest film, "Toy Story," Sun and Pixar Animation broke
new ground. Following are some key facts that can help you tell Sun's story
as it relates to the production of this film.
___________________________________________________________________________
 
* "Toy Story" is a milestone in animated film. It is the first completely
computer-generated full-length (78 minutes) film in history. In contrast,
"Jurassic Park" had about 4 minutes' worth of animation.
 
* What viewers see on the screen--the life-like characters, the light and 
shadow, the surface textures, and so on--was done on a networked cluster 
of 117 SPARCstation 20s, one SPARCserver 1000 server and a SPARCstorage Array. 
The systems did what is called rendering.
 
* The systems were also used earlier in the film's production for lighting
tests, again working on textures and surface qualities.
 
* The SPARCstorage Array was used to store the final rendered frames
before they were transferred to film.


* RenderFarm Facts: Ran 24 hours/day, 7 days/week from early January 1995
to August 1995. No processor failed during this time.

* The film comprises about 34 TRILLION bytes of data. Until now, producing 
a film like "Toy Story" was cost-prohibitive. Would have required about 
300 Cray computers or about 16,000 DEC VAX11/780s. Sun drastically reduced 
the cost by bringing down the cost per rendermark. 


----- End Included Message -----

If Johnny Mnemonic hadn't been such a dud, we might have gotten some good
press from it (though Alphas weren't involved.)

				Steve

> * "Toy Story" is a milestone in animated film. It is the first completely
> computer-generated full-length (78 minutes) film in history. In contrast,
> "Jurassic Park" had about 4 minutes' worth of animation.

Just as I suspected ... those *were* real dinosaurs!

    How many Turbozillas equate to .65's SPARCosaurus (tm someone) farm ?

re : Note 4368.67 by QUARK::LIONEL 

>If Johnny Mnemonic hadn't been such a dud, we might have gotten some good
>press from it (though Alphas weren't involved.)
    
    Digital Semiconductor played a tape with film clips from films and TV
    shows thatused Alpha systems to produce computer graphic generated
    between shows in it's booth at COMDEX.  I think the clips came from 
    Amblin Entertainment.  
    
    -Bruce
    

    My understanding is that Johnny Mnemonic used Digital Pentium PCs for
    its animation.
    
    			Steve

    
    A quick note on the workstations of which 45% are VMS-based. 
    
    A great many of those machines are going to resellers, who use them
    as MV3100 replacements at the low-end. I know, I've sold over
    200 Mustang LXs this year for just *that* purpose. The buyer gets
    a 2, or, 4-user update for OpenVMS.
    
    I have yet to see a VMS single user workstation. ala SUN....
    
    		the Greyhawk

    To .72:
    
    It's very difficult to know exactly which OS is used on a workstation.
    In fact, many NT workstations are used in the edu market to install
    OpenVMS.... Just because the price is lower... Apparently our licence 
    policies allow these kind of practices with the so called DECcampus
    program.

    16,000 VAX 11/780s, huh?  I wonder how many TurboLasers that equals...
    

>    16,000 VAX 11/780s, huh?  I wonder how many TurboLasers that equals...

Assuming that's floating-point work, one Turbolaser CPU is approximately
500 SPECfp92's, or approximately 500x a VAX-11/780.  So it would take
approximately 32 Turbolaser CPU's -- or about 1/3 of the "Turbozilla"
cluster that was shown at the Supercomputing conference in December.

    re .73
    I don't know anything about the DECcampus program, but I have seen some
    of the hoops we have jumped through to PREVENT systems sold with NT
    licenses from being used with OpenVMS or Digital Unix. You may want to
    take a closer look as the customers to whom you refer may be in
    violation of blah blah (all the legal stuff).
    
    Rob Wall
    

>    I don't know anything about the DECcampus program, but I have seen some
>    of the hoops we have jumped through to PREVENT systems sold with NT
>    licenses from being used with OpenVMS or Digital Unix. You may want to
>    take a closer look as the customers to whom you refer may be in
>    violation of blah blah (all the legal stuff).

	There are (and have been) products you can buy this way leagally:

		- buy WindowsNT based AlphaStation or AlphaServer
		- buy separate Digital UNIX or OpenVMS license

	And you find out that the combination is lower priced than 
	buying same system with UNIX or OpenVMS factory installed...

	Silly but it happens... And this is true even for "non DECcampus"
	customers. 

		-jari

    Web pages:
    
    http://www.imc.das.dec.com/alphastation1
    
    http://www-wbs.eng.pko.dec.com/high_perf/hp.htm
    
    
    					RJD

    
From:	WRKSYS::COTE "Performance Workstations Marketing 508-493-1538  08-Mar-1996 1203"  8-MAR-1996 12:04:02.00
To:	@mav-5x5
CC:	MTS$::"GEO::CAROLE-ANNE HOUCKE"
Subj:	updated information in share area.

Folks:

Many people have indicated that they cannot print the postscript files that 
I have put into the share area.  Therefore, I put some of the MS-Word files
there as well.  

Here is the list of files.

wrksys::USER$$01:[APSSHARE.MAVERICK]


	500OPSOC.DOC -  500 options document
	500SOC.DOC - 500 SOC document (both 333 and 266)
	500SOC.PS - 500 SOC file (both 333 and 266)
	DISSOC.DOC - distributor SOC document (255, 500, 600)
	DISSSOC.PS - distributor SOC file (255, 500, 600)
	DIST_PKG.PS - distributor price file numbers - there will not be an 
		      editable document for this file.
	PRICE.PS - package file numbers - there will not be an
		   editable document for this file.
	SALESUP.DOC - Sales update article document
	SALES_UP.PS - Sales update article file

Hope that this helps.

Regards,
Bruce

)0 Worldwide News                                 LIVE WIRE
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
 New AlphaStation workstations, PowerStorm ...               Date: 11-Mar-1996
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
Page   1 of 3  
                New AlphaStation workstations, PowerStorm graphics  
                             deliver top performance 
   
         New entry and mid-range desktop workstations and graphics options 
   from Digital deliver the industry's best overall performance and best 
   application performance for the money.  The new products are designed 
   to allow customers working with mechanical and electronic design, 
   software development, and other engineering and scientific 
   applications, to significantly improve their time-to-market 
   capabilities.
         The new products include:
   
         o  The AlphaStation 255 workstation series -- AlphaStation 
            255/233 and 255/300 -- with prices starting at $7,399, 
            delivers the best price/performance among all entry-level 
            workstations;
   
         o  The AlphaStation 500 workstation series -- AlphaStation 
            500/266, 500/333, 500/400 -- with prices starting at 
            $15,863, provides the best price/performance among all 
            mid-range workstations and best overall performance;
   
         o  The PowerStorm family of PCI-based workstation graphics 
            options is supported across all Digital workstations, 
            delivers the best mid-range graphics performance and 
            utilizes the common OpenGL API found in most third-party 
            applications.  
         
         "With today's announcement, Digital is restating its commitment, 
   in a very bold way, to the workstation market," said Harry Copperman, 
   vice president and general manager, Systems Business Unit.  "With our 
   branded PowerStorm graphics, we're renewing our commitment to providing 
   superior graphics.  Digital is once again setting the pace and 
   providing leadership solutions in the technical computing market."
   
   Sizzling performance plus investment protection 
   
         According to benchmark results, Digital's new AlphaStations and 
   PowerStorm graphics outperform the competition in compute- and 
   graphics-intensive application areas.  For example, running Parametric 
   Technology's Pro/ENGINEER software, a new AlphaStation 500/266 
   workstation demonstrated 33 percent better application performance than 
   the Sun UltraSPARC 140; 61 percent better application performance than 
   the HP C100; and 78 percent better application performance than the SGI 
   Indigo2 R4400/250.  Also running Pro/Engineer, the new AlphaStation 
   500/333 demonstrated 85 percent better application performance than the 
   high-end Sun UltraSPARC 170E; 85 percent better application performance 
   than the HP J210; and 117 percent better application performance than 
   SGI's Indigo2 R4440/250 workstation. 
         The need for superior application performance is supported by 
   customers such as Bob Irwin, IS director at Moog, Inc., a manufacturer 
   in the aerospace industry.  "We evaluated workstations from Digital, 
   IBM, Hewlett-Packard, Silicon Graphics and Sun running EDS Unigraphics 
 
)0 Worldwide News                                 LIVE WIRE
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
 New AlphaStation workstations, PowerStorm ...               Date: 11-Mar-1996
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
Page   2 of 3  
   software to compare the actual performance of each," said Moog.  "We 
   were extremely pleased with the results of Digital's workstations, 
   which outperformed the other vendors' offerings."
         Nadir Khoshniyati, CEO of microCAD Solutions, one of the largest 
   VARs in North America specializing in the mechanical engineering 
   market, said, "The price/performance of Alpha workstations is just 
   incredible.  Our customers are looking to see how fast applications run 
   on different platforms and they are really impressed with the 
   performance of Alpha."
         Customers who invest in the AlphaStation 255 workstation will be 
   able to upgrade to an AlphaStation 500 within the same enclosure. 
   Because AlphaStations support Digital's 64-bit operating systems -- 
   Digital UNIX and OpenVMS -- and Windows NT, a customer's hardware 
   investment is safe, even when migrating to a different operating 
   system. 
         Each AlphaStation 500 workstation includes a 2 MB fast L3 cache, 
   256-bit memory bus, up to 8 GB of internal storage in five drive bays, 
   four PCI slots, support for dual-fast and wide SCSI-2 channels, 
   Ethernet and Fast Ethernet, and up to 512 MB of RAM.  The AlphaStation 
   500/266 workstation delivers 7.29 SPECint95 and 10.5 SPECfp95; the 
   500/333 delivers 8.8 and 11.6 SPECint95 and SPECfp95, respectively; and 
   the 500/400 delivers 11 SPECint95 and 14 SPECfp95. 
         AlphaStation 255 workstations offer 1 MB of secondary cache, a 
   128-bit memory bus, 2 PCI, 1 PCI/ISA, and 1 ISA slots, and support up 
   to 6 GB of internal storage in four drive bays. The AlphaStation 
   255/233 workstation delivers 3.8 SPECint95 and 5.09 SPECfp95; and the 
   AlphaStation 255/300 workstation delivers 4.51 and 5.71 SPECint95 and 
   SPECfp95, respectively. 
         All AlphaStation workstations include bundled multimedia 
   capabilities, including CD-quality audio, and offer an industry-leading 
   three-year warranty.  The AlphaStation 500/400 workstation will be 
   available in June.  All other models are available immediately. 
   
   About PowerStorm  
   
         PowerStorm graphics were specifically designed for 2D, advanced 
   3D wireframe and solids applications, and for applications requiring 
   the highest performance for realistic motion, advanced hardware shading 
   and texture mapping. 
         Digital's PowerStorm family includes:
   
         o  PowerStorm 3D10, ideal for price-sensitive applications, 
            such as basic graphics, text and menus.  3D10 is priced 
            at $399 and is available immediately. 
   
         o  PowerStorm 3D30 offers the industry's fastest Xmark 
            performance, the best price/performance of any entry 3D 
            wireframe product and the best 8-bit 3D wireframe 
            performance at any price.  3D30 is priced at $795 and is 
            available immediately. 
   
         o  PowerStorm 4D20 has the fastest 3D wireframe performance 
            at any price and is the most cost-effective solids 
 
)0 Worldwide News                                 LIVE WIRE
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
 New AlphaStation workstations, PowerStorm ...               Date: 11-Mar-1996
 qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
Page   3 of 3  
            modeling solution. The PowerStorm 4D20 is priced at 
            $2,495 and is available immediately. 
   
         o  PowerStorm 4D60T will deliver high-end performance for a 
            mid-range price. 4D60T will be optimized for OpenGL 
            throughput, offer texture mapping, and deliver leadership 
            price/performance for under $15,000.  It will be 
            available this summer. 













































 
    

    reality check:
    AlphaStation 255/233 has "the best price/performance amoung all
    entry level workstations".  Entry price is $7399.
    
                         	SpecInt95	SpecFP95
    AlphaStation 255/233:       3.8		5.09
    P6@200                      8.0		5.82
    
    Entry P6@200's have better performance for less money.  I suppose
    you could always claim that the P6 competition, while faster and
    cheaper, isn't an "entry level workstation".  Kratz

    	I agree that new EV45 machines are not the best idea.
    	Maybe they mean VMS and UNIX when the term "workstation" is
    	used?
    
    					RJD

The blurbs for these workstations don't admit to the existence of VMS at all.

				Steve

    To be ethically and morally correct, I think we should start putting
    an asterisk next to some of these Alpha price/performance claims.
    In the tiny print at the bottom, we can put something like:
    "For machines that run VMS from companies with Digital in their name"
    or
    "We're actually lying thru our teeth on this one, but odds are you'll
    believe us anyhow and never read this disclaimer".
    
    As is tho, some of these claims are getting hip deep in sheep dip.
    .02 Kratz

    
     Yes, I feel I am being brainwashed. When I start digging, I see this
    performance "gap". I say we let loose the marketing on customers in a
    more truthfull way. This battering of the digital mindset does no one
    any good.
    
    Sigh.
    
    -Mike Z.

    Not that the PCBU isn't guilty of trying to pull the fast
    one either: I just got a PCBU marketing Flash on the Celebris
    XL 6180 and 6200 that uses the old SPECint92 numbers that
    Intel pulled in December. 

    Yeah, ok so we may massage the message a little, but at least our boxes
    are a nice new improved colour (just like SGI's so it must be OK).
    
    NOT!
    
    Instant doubling of the weight of the SOC - will that be a new blue
    TLZ07 or a traditional-colo(u)r one, miss ?

    That's a very snazzy feline at http://www.alphastation.digital.com.
    My kitties approve :-)
    

Is that the center picture that's unviewable except by Netscape?

			Steve

    re .89
    
    Java in action...

I don't think it's Java - it's a "server-push" series of images.

			Steve

re http://www.alphastation.digital.com/:

Well, I'm using Netscape V2.0 with Java (on Digital UNIX), and 
I just see the "broken image" in the middle.

Bob

    
    Glad it is not just my Netscape that is broken...
    
    John
    

I've been discussing the problem with Marcia McCann, who created the page.

			Steve

    I'm running Netscape 2.0, and I see a cougar (? not up on my wild
    animal names) leap out of the screen a couple of times.  I had
    looked at the page source, out of curiousity, and it seems to be
    being done by a cgi file.  So, I thought that wasn't Java...
    

    I saw it, when the page first comes up. If you aren't looking at it
    then, you miss it. And it is a panther, which is a black leopard.
    
    Deb

    I just assumed it was Java. Oh vell. You're right, it looks like
    it's a Perl script. 

    re: reality check:
|                         	SpecInt95	SpecFP95
|   AlphaStation 255/233:       3.8		5.09
|   P6@200                      8.0		5.82
    
    What OS was used by Intel to generate the P6 numbers?
    What OS is used by P6 customers?
    
    How come the answer is different?
    
    								-mr. bill

    Sorry, Bill, notes collision.
    You need to get a Pentium Pro Performance Brief (or read one).
    Novell/AT&T Unixware V2.0 is the OS used by Intel's SPEC95 numbers.
    Unixware V2.0's market share is bigger than Digital's Unix, but
    apparently not big enuf to satisfy you.  The WIndows NT performance
    numbers are done with WIndows NT V3.51.  Digital is also guilty of
    mixing OS's in performance reports: witness the Jensen reports.
    Digital Semi also has ads with SPEC numbers from Unix and "runs
    NT like a racehorse" copy, never mentioning that the numbers aren't
    from NT.
    Kratz
    

    Oh, trust me, I've read the Pentium Pro Performance Brief.  Over and
    over again.
    
    Unixware V2.0 as a server, yes.
    Unixware V2.0 on a workstation?  Oddly, workstation apps don't appear
    to be available on Unixware V2.0.  (Quite often they *are* available
    for Windows/NT.)
    
    Those "reference" compilers are finally available, right?
    
    								-mr. bill

    Bill,
    Tip on it's side, it's a server.  Call it what you want.  Actually
    the underlying OS doesn't impact SPEC95 numbers that much; a P6
    would still kick an Ev45 product for less money using SCO instead
    of Unixware.  Your (belated) point on the compiler is a good one tho.
    If it occurs, cry foul when the 6 month period for compiler
    availability after publication is passed (that's still the rule, no?).
    Kratz 

    No, it's not call it what I want.  You can't put a server on a desk and
    call it a workstation.  Novell is pushing Unixware as a server OS.
    And it's software that turns an office heater into a workstation,
    not it's location.  Most people buy these things to do something.
    (OK, we all know the folks who just heat their desk with them.)
    
    You can get Pro/Engineer for Digital UNIX.  You can't get
    Pro/Engineer for Unixware.  You can get Pro/Engineer for
    Windows/NT.
    
    Now I can see you believing folks will sell boatloads of P6s running
    Windows/NT that compete with AlphaStation 255 running Windows/NT.
    I can even understand you believing folks will sell boatloads of P6s
    running Windows/NT that compete with AlphaStation 255 running Digital
    UNIX.
    
    But to claim that P6 running Unixware competes with an AlphaStation
    255 running Digital UNIX, well, you are dreaming.
    
    
    So, what are the SPECint95 and SPECfp95 metrics for a P6 running
    Windows/NT?
    
    								-mr. bill

    >So, what are the SPECint95 and SPECfp95 metrics for a P6 running
    >Windows/NT?
    
    As I'm sure you're well aware, SPEC hasn't released the '95 suite
    under NT yet.  No surprise: the Unix-based SPEC consortium is
    scared sh*tless of NT.
    
    BTW, The trying-to-be-competitive-with-Intel Alpha XL's don't even
    run Digital's Unix... for that priviledge, you get to shell out more
    megabucks for the same box.
    Kratz
    
    
                                                                        

    
|   As I'm sure you're well aware, SPEC hasn't released the '95 suite
|   under NT yet.  No surprise: the Unix-based SPEC consortium is
|   scared sh*tless of NT.
    
    Nice spin.  Too bad you are wrong.
    
    								-mr. bill

    >Too bad you are wrong
    
    SPEC95/NT out, or you think my assertion that NT is a threat to
    Unix is wrong?
    Kratz

    No, SPEC95/NT is *not* out.  The SPEC consortium, who you incorrectly
    call "UNIX-based" wants very badly for SPEC95/NT to be finished and
    shipping.
    

    Another vendor, in all fairness, put a fair amount of work into
    the SPEC95/NT port.  They've made a good start, but they
    promised to be finished, and they aren't.  They've asked for help,
    they've asked for someone else to finish.  (Most of the work is
    in the finishing.)
    
    
    Surprisingly, somebody is going to finish the SPEC95/NT port.
    Just as somebody is going to finish the SPEC95/VMS port.
    
    								-mr. bill

    >Surprisingly, somebody is going to finish the SPEC95/NT port.
    >Just as somebody is going to finish the SPEC95/VMS port.
    
    Same 'somebody', by any chance?
    
    ...petri

    Looking forward to seeing SPEC95 ported to NT.  Then Digital
    Semi won't have to do ads with SPEC numbers (SPEC92 of course)
    with text copy that talks only about NT (the "Runs NT like a
    racehorse" ad, to be specific).  Nowhere in the ad did it mention
    those SPEC numbers are from Unix.  SPEC reprimanded Digital for
    that misleading ad if I'm not mistaken.  Kratz

    Are you finished yet?
    
    Getting all these facts wrong doesn't slow down your pot shots at all,
    does it?
    
    								-mr. bill

    Then speak the truth, oh Godlike creature.
    Take a look at the ad yourself Bill.  SPEC *was* miffed.
    Shak'll remember it if you can't.
    Kratz

                                                                         
|   SPEC reprimanded Digital for that misleading ad if I'm not mistaken.
    
    You are mistaken.
    
    A couple of SPEC members asked about the ad.  The answer was the
    metrics were based on measurements on Unix, sorry the ad didn't fine
    type that somewhere.  (Intel's earlier P6 pre-announcement promo
    inserts in the PC magazines didn't fine type (*UNIX) anywhere
    either, btw.)
    
    Listen, you are talking to someone who sweats everytime an (est) isn't
    in the right place, so *I* was miffed about that ad.
    
    Someone you know well in our group, when looking at a HELP WANTED AD
    in the March 10th Globe, their first reaction -- "Oh no, do we have a
    fair use violation"?  (No, we didn't.  But to worry about a help wanted
    ad, it seems almost silly.)
    
    But a SPEC reprimand of Digital Semiconductor's racehorse ad?  Sorry,
    just didn't happen.
    
    								-mr. bill

    Ok Bill, but reread .99 where you criticized me in an internal
    note for doing literally the same thing Digital did in an
    external advertisment.  It's bad when I don't mention the OS
    used in an internal note, but ok when Digital doesn't do it in
    an ad for the world?  K

    Damn, I lost the pool.  Who had backpeddle?
    
    
    You've called just about everyone else a liar.  So, yeah, when you
    don't measure up to your own standards, I think it is important to
    point that out.
    
    When you repeatedly want readers here to believe you have knowledge,
    and you get things so very wrong, I think it is important to point that
    out as well.
    
    
    But no, it is not good that Digital Semiconductor did not fine print
    the ad.
    
    And no, it is not good for you to keep quoting UNIX numbers
    for Pentium PRO workstations that run Windows/NT.
    
    And no, it is not good for you to pretend that every Pentium
    PRO workstation has performance identical to an Intel Alder.
    
    And no, it is not good for you to quote performance numbers for a
    high end Pentium PRO workstation and quote prices for a low end
    Pentium PRO workstation.
    
    And no, it is not good for you to....
    
    								-mr. bill

    >And no, it is not good for you to keep quoting UNIX numbers
    >for Pentium PRO workstations that run Windows/NT.
  
    Bill, Pentium Pro workstations can run DOS, Unix (lots of flavors),
    OS/2,..., etc, not just Windows NT.  Some benchmarks come from the
    Unix world (like SPEC, which is only Unix and VMS at this point),
    some are based on other operating systems.  Sounds like you're
    confusing Pentium Pro machines with the Alpha XL's, which runs
    Windows NT *only* (lots of versatility there).
    
    As for Alder numbers, you're right, an Aurora (as used in Gateway,
    Dell, etc) or our design is more appropriate.  But since neither has
    external cache, and they all use the same chip set, the difference in
    performance is minimal (see http://www.bapco.com for example:
    Alder @200: 648; our P6@200 does a 639).  That's a little over 1%
    difference; sorry for misleading folks.
    Kratz
    
    
    

    
|   That's a little over 1% difference; sorry for misleading folks.
    
    You just won't quit.  Bapco is *not* a bash the memory hierarchy
    benchmark.  To imply, as you just did, that we are only talking about a
    1% difference in SPEC95 results by pointing to Bapco is the *HEIGHT*
    of arrogance.
    
    								-mr. bill

    Bill,      
    Bottom line: you pick the benchmark and pick any P6@200 box, and it'll
    beat anything Ev45-based.  And in the case of BAPCo and some other
    benchmarks like SPECint95, it'll even cream *MOST* of the current
    crop of Ev5-based boxes.  If you don't like the workstation-type
    application support of the particular operating system under which
    the Unix-based benchmark numbers were obtained, I guess that's your
    problem.  It's still a real SPEC95 number; you had your chance to
    officially challenge it (and the BAPCo numbers for that matter).
    Regards,
    Kratz
    BTW, I refuse to stoop to namecalling; you're on your own there. 

    In the case of SPECint95, your Pentium PRO will lose to *most* of the
    AlphaStation 500 family (the topic of this note).  Those are our
    EV5-based desktop boxes.  It'll get *dusted* on SPECfp95.  Even
    spotting your Pentium PRO running Unixware with "reference" compilers.
    
    I look forward to the challenge of a fair fight - Intel Pentium PRO
    vrs Digital AlphaStations, both running Windows/NT and benchmarked with
    SPEC95.
    
|   BTW, I refuse to stoop to namecalling; you're on your own there. 
    
    Oh really?  See .111.
    								-mr. bill
    

    since when did the p6-200 specfp/95 get downgraded by about 14% to 5.82
    from their initial fact sheet--was that another itty-bitty
    overstatement or is its algorithms messed up again 
    
    
    
    rick

    >In the case of SPECint95, your Pentium PRO will lose to *most* of the
    >AlphaStation 500 family (the topic of this note).  Those are our
    >EV5-based desktop boxes.  It'll get *dusted* on SPECfp95.
    
    No, our little trist started with .99, which if you reread it, has
    nothing to with EV5.  I like how you completely changed the machine.
    From April 1996 "Byte" magazine, p.40 and Feb 1996 "Byte", p.155:
                          	PPro200/256 (*ANY*)	Ev45@275
    Bytemark V2.0 integer (NT):      3.5                 2.8
    Bytemark V2.0 fp (NT):           2.8		 2.45
    
    You're more than welcome to try it under Digital Unix... the sources
    are on their web page.  As for Ev5, we're in COMPLETE AGREEMENT!!
    I've been trying to get EV5 moved into the mainstream (i.e. cheap)
    for 6-9 months; see my note to Palmer (4031.9) when he solicited input
    for his employee DVN.  The problem I had, if you reread .99, was calling
    an EV45-based system "best price/performance".  I can put together
    a P6/200 system that runs Bytemark (or just about anything else)
    faster than EV45 for about half the $7k entry cost of the AlphaStation.
    But as is, we're making it really hard on resellers when customers
    come up to them with BAPCo, Bytemark, etc results and wonder how we
    can claim "best price/performance" with some of these machines...
    Kratz

    re: .119
    
    No, I don't believe SPECfp95 got "downgraded."  Just consider the
    reliability of the source.  (The number Kratz quoted is incorrect.
    I should have assumed that all numbers were suspect, but who would
    ever dream that somebody would understate an Intel number?  Thanks
    for catching that.)
    
    re: .120
    
|   And in the case of... benchmarks like SPECint95, [P6@200] will
|   even cream *MOST* of the current crop of Ev5-based boxes. 
    
    You wrote that, I didn't.  You were incorrect.  A Pentium PRO at
    200MHz beats an EV5 at 266MHz on SPECint95.  (A *remarkable*
    achievement.)  Every other EV5 in shipping systems outperforms
    the Pentium PRO.  To put it another way, the fastest Pentium PRO
    beats the slowest EV5 at SPECint95.  (Again, a *remarkable*
    achievement.)
    
    
    Can we, Digital, *ever* brag about a Unix workstation without you
    taking shots at it?  Just what is the proper way to say "it's
    faster than a Sun, it's faster than an HP, it's faster than an
    IBM, it's faster than an SGI."  What?
    
    								-mr. bill

    Hi Kratz,
    
    In .120 you mention in passing the $7K price which I will assume is the
    AlphaStation 255 UNIX entry level (eg PB470-aa).
    
    I believe PC industry practice is "street pricing" - no discounts, what
    you pay is what you see. Give or take a little bit. Am I right ?
    
    Workstation practice is NOT street pricing. Any worthwhile customer
    would be able to get a SERIOUS discount - maybe 30% or more - either
    from Digital or from a box-shifting reseller. Same for HP, SGI, Sun,
    etc. DEC tried street pricing a few years back but has since returned
    to keeping purchasing managers and resellers happy (and potential
    customers confused about real prices).
    
    So the $7K list price you quote should be a lot less in real terms.
    Does that change your picture at all ? 30% off doesn't _quite_ halve
    the price (not even with a Pentium doing the sums.... ha ... ha), but
    it does make quite a difference.
    
    Also, look at it another way - cut out the DEC bits which don't need to
    be DEC, and start with a "system building block". Just add $$$$.
    
    Today's list price of the "system building block" equivalent of that
    AlphaStation, for NT, is $2k7 (PB47A-CA). To which you need to add the
    bits to make it usable. For NT, be it Intel or Alpha, they don't need
    to be Digital bits. How does that compare with PCBU pricing ?
    
    For UNIX, which is by definition what "workstation applications" run
    on, the system building block (PB47A-AA) is $3k7 today. Being UNIX, the
    bits you add are somewhat more restricted, but that's one of the
    necessary consequences of going for a single-vendor solution (one phone
    number to ring).
    
    So: *do* Compaq, Dell, GW2K, and the PCBU routinely offer 30% discounts ?
    
    regards
    john

Unless you want to get a lot of I/O thruput through the P6 box.  Intel
announced that the Orion chipset has a "bug" which limits PCI DMA write
posting rates to ~5MB/s (instead of 60-70MB/s it should be).  This makes
P6 boxes completely unsuitable for servers.

    re: Orion bug
    
    According to Intel, this situation happens when you mix PIO and bus
    mastering controllers. If you have a straight SCSI disk farm, you
    shouldn't run into this problem.
    
    As I was walking through the local PC marketplace here in Singapore I
    saw a number of dual-processor PP motherboards for sale. I'm not sure
    what the street price is for PP/200's these days, but even the thought
    that I could build my own Pentium Pro NT server with standard parts off
    the shelf had me salivating. Although I wish Alpha had the same level
    of openness and market availability, it's just not going to happen ...
    
    Geoff

    Re: .123
    
    And that has already been fixed in the chipset Intel is currently
    shipping.
    
    ...petri

    re: street prices:
    April Computer Shopper shows P6/200 motherboards at the $2500 level;
    the only ad I could find with Alpha motherboards was Motherboards
    International where EV45 boards start at $3500 and EV5/300 boards
    are in the $7-9k range.  While Digital Semi may offer motherboards
    to wholesalers/resellers for less money, it doesn't seem to make
    it to the retail level.  It would be nice to see EV5/300 boards in
    the $2500 range; then we'd actually have a helluva price/performance
    story! 
    
    P.S. all of our (Digital's) 200Mhz have the Orion bug fix.  
    Kratz

    I was interested in street prices of systems, not boards, as most of
    Digital's volume customers are buying computers (not hobby kits) today.
    Can you address street prices of PC systems too ? Wrt Compaq, Dell, GW2K,
    HP PC, others ? Wrt HP PA/RISC, SGI, Sun is already covered, thanks.
    
    The Motherboards International Alpha prices have restricted appeal; an
    AlphaStation 255 NT SBB from Digital with onboard goodies, warranty,
    etc, lists for less $$$ than their advertised Alpha board. Pay even
    less if you get a discount from Digital or a reseller.
    
    Admittedly the AS255 has non-standard form factor and restricted
    expandability. But that's the tradeoff. Maybe there's a niche for
    someone to fill with BabyAT or ATX boards and big boxes and lots of
    slots, priced less than the AlphaStation 600. If someone wants to "just
    add hot peripherals" the 255 SBB is a reasonably low-risk (but also
    low-return) approach. Doing it to make money would be VERY vulnerable
    to Digital's next downward change of price (but one or two
    organisations seem to do it as a sideline).
    
    And how many corporates buy from Computer Shopper anyway ? 
    
    be seeing ya
    jw

    Be my guest if you want to compare a GW2K P6/200 at under $7k
    fully loaded with what an AlphaStation255 prices out to be with the
    same line item goodies (64Mb, 4Mb Matrox, 21", 2Gb fast SCSI on 2940U,
    4x SCSI CD, NT, MS-Office,...), minus 30%.  The 64Mb and 21" are
    going to be the killers.  Kratz
    
    

    re: .127
    
    >I was interested in street prices of systems, not boards, as most of
    >Digital's volume customers are buying computers (not hobby kits) today.
    
    While you are correct that corporate customers don't buy parts, they
    *do* buy systems from vendors who buy parts, aka Dell, GW2000, etc.
    And if "hobbyists" like me are now empowered to build high end Intel
    systems, what do think mainstream vendors and VAR's are capable of
    building? I'm not throwing stones at Alpha, but Intel is *everywhere*,
    always in your face, and the entire product range is accesible by just
    about anyone from hobbyist to corporate purchasing agent. Whereas it
    takes special detective skills and a lot of patience to get Digital to
    agree to sell you an Alpha, hopefully at a discount that will make it
    even mildly affordable ...
    
    Geoff

    Compare an AlphaStation 255/233 and AlphaStation 255/300 to:
    
    	Sun, HP, IBM, SGI
    
    Is that so hard to understand?
    
    I mean, we all *know* that those folks haven't sold a single UNIX
    workstation since November.  Kratz has been reminding us of this
    "fact" for months and months and months.
    
    -----
    
    BUT, let's take a closer look at the "workstation" that Kratz thinks
    the world will buy....
    
    First, the GK2K P6/200 is over 7K.  You forgot a sound card and
    ethernet and minimal service.  Speaking of service, price out a
    three year service contract on that workstation?
    
    Also BTW, say you want a GW2K P6/200 with 32MB memory, 2MB Matrox,
    17" color, 1GB fast IDE drive, floppy and CDROM?  How much will that
    cost?  $7K again, because they won't sell you anything but the fat
    system. 
    
    Another BTW, hope you don't need more than 64MB of memory, because
    you'll have to wait for it.  And then you'll only be able to max out
    at 128MB of memory.  And if you can't wait, you get to toss that
    bargain 64MB of memory when the 32MB SIMMS start hitting the street.
    That memory doesn't sound like such a bargain now, does it?  (I guess
    you can get some money back by selling your used SIMMS in Malden.)
    
    Final by the way,  Kratz, after you add all those adapters in the
    i/o riser, how many free slots are left on that system?  And remember,
    this is the bottom of your desktop workstation family, you want a
    boatload of much faster workstations, even willing to spend more than
    7K for them.  Gosh and golly, that is Gateway's fastest workstation?
    
    								-mr. bill
                                                              

    
     This string is proof positive that it takes a very shrewd person to
    come up with a configuration that will meet their business needs 
    "now" and "in the future".
    
    Only problem is that 90 percent of the PC buyers don't think like this,
    so when they buy... they just keep on buying. It is a lot easier to
    explain to the boss.. "we need another server", "A PC will suffice"...
    than to get it right the first time. Pretty soon you have multitudes
    of them. Hey.. it only cost 7k to start with.
    
    So, personally, I feel a low cost entry slot from Alpha would get our
    ball rolling. If you can do it on x86, Pwhatever, you can apply the
    same domino theory to our stuff.
    
    -Mike Z.

        re: Kratz & Mr. Bill string...

    Picture both of you coming into one of my or other people's customers to 
    represent both sides of Digital's "cohesive" XL Workstation strategy.

    Yow!!  90's version of the thrilla in Manilla!

    Yet another instance of Digital drawing defeat out of the jaws of
    victory.  Digital's problem isn't whether its Intel or Alpha W/S is *the*
    superior platform.  The problem is that there's not enough of either of
    them in the hands of customers.
    
    Maybe the two of you ought to get together over a couple of beers, have
    a few laughs, exercise your considerable talents, and realize that
    the enemy ain't us.
    
    My $0.02

    Bill,
    You need to give Gateway a call (800) 555-2099 and ask them about
    pricing and config options.   Your last note is full of, er,
    inaccuracies.  P.S. They are very accomodating to config and memory
    stick size if you ask.
    
    >Gosh and golly, this is Gateways fastest workstation?
    At more than twice the performance of the Alpha EV45 (SPECint95,
    BAPCo,...) for about the same or less money, you're right, perhaps
    it wasn't a just comparison.  But the last note does have a good
    point: we can beat SGI, HP(RISC), PowerPC with either Intel or Alpha;
    it doesn't really matter which the customer picks.
    .02 Kratz 

|   >Gosh and golly, this is Gateways fastest workstation?
|   At more than twice the performance of the Alpha EV45 (SPECint95....
    
    *Less* than twice the performance at SPECint95.  It's the constant
    drip drip drip of all these overstatements that really end up bothering
    some of us.  (And if somebody wants a system that has more floating
    point than either EV45 or P6/200 can deliver....)
    
|   But the last note does have a good point: we can beat SGI, HP(RISC),
|   PowerPC with either Intel or Alpha; it doesn't really matter which the
|   customer picks.
    
    Oddly, whenever and wherever anyone has a parade over Alpha beating
    Sun, HP, IBM or SGI, some of us are learning that we should expect
    rain.
    
    								-mr. bill

    re: SPECint95, see .99
    3.8 is less than half of 8.0, even using an old Pentium ;-)
    Where's the overstatement?

    
    4.51 is not half of 8.09, no matter how you do the math.
    
    								-mr. bill

    The SBU Americas just announced a special upgrade on the Alpha XL 266.
    Buy an Alpha XL 266 today, and upgrade to a 21164A @366mhz in 1996.
    Cost of the upgrade is $3995. For more info check:
    
    www.alphastation.digital.com/promo/alpha_nt/nt_dbl.html
    
    Digital can now offer an the best 1-2 NT-workstation punch in the industry:
    
    Celebris XL - w/ 200mhz Pentium Pro 
    "best price/performance for integer based apps"
    
    Alpha XL - w/366mhz 21164 upgrade
    "incredible performance for fp based apps", and at a price point that
    blows UNIX systems from HP, Sun and SGI out of the water.
    
    ... both Celebris XL and Alpha XL (w/366) are available for under 10K.
    
    Bottom line is we have an unbeatable NT workstation combo with Celebris
    and Alpha XL.
    
    ... let's go sell some. I'm sure the stockholders would be
    happy no matter which one we sell, as long as we sell Digital.
    
    Regards,
    Tim
    
    

    Ah yes, the Celebris XL. Nice box, esp with EV5. But it's "The machine
    with no channels" at least in the UK. PCBU won't actively sell it
    because it's got an Alpha in it. SBU won't actively sell it because it
    has a LOT less margin than they are used to (they like to be thought of
    as the "BMW" of the industry i.e. not cheap). For the SBU it may appear
    to be profitable at local level to sell an XL, but by the time HQ
    deduct all their taxes, it looks like a negative-margin product. Does
    the same apply elsewhere ?

RE: .136

	Maybe if you use a Pentium to do the math?

								mike

    re .137    
    >Celebris XL - w/ 200mhz Pentium Pro 
    >"best price/performance for integer based apps"
    
    Nice to see somebody can see thru the fog... betcha ya don't 
    win any friends over in Hudson for that statement tho.  You'll
    have to start checking for car bombs too now ;-)
    K  
    

    re .139
    Bill switched to a higher clocked EV45 from note .99, that's why
    he's claiming 4.51 is better than half of the P6's integer
    performance.  Regardless, it's still pathetic.
    K

|   Nice to see somebody can see thru the fog... betcha ya don't 
|   win any friends over in Hudson for that statement tho.  You'll
|   have to start checking for car bombs too now ;-)
    
    Nah, I doubt anyone from Hudson or anywhere else in the company is
    nipping at the heels of every mention of Celebris XL 200MHz Pentium
    PRO.
    
    Imagine for a moment that within an hour of sharing good news
    about a Celebris P6/200MHz Windows/NT workstation in this or
    any other conference someone enters replies along the lines of:
    
    	Enter cheap shot mode>
    
        Best price/performance?  Yeah, only if you don't look at
    	a GK2K P6/200 which has the same performance for less money.
    	Maybe we need some fine type, best price/performance of
    	Pentium PRO workstations with the Digital LOGO?
    
    	Besides, Pentium PRO, give me a break.  Pentium machines get
    	better price/performance (more than half the integer performance
    	at less than half the cost), why would anyone would spend
    	the money on a Pentium PRO workstation?
    
    	And with 486 machines falling to near "internet appliance"
    	prices, the price/performance nod might even go there.
    
    	Exit cheap shot mode>
    
    I would like to see Digital sell boatloads of Intel Pentium PCs,
    boatloads of Pentium PRO Workstations, and boatloads of Alpha
    Workstations.
    
    Perhaps we can get *someone* here to acknowledge that there is a reason
    for Digital to sell the best damn UNIX workstations in the industry?
    Perhaps we can get *someone* here to acknowledge that there is a market
    for hot floating point for NT.
    
    								-mr. bill

|   Bill switched to a higher clocked EV45 from note .99, that's why
|   he's claiming 4.51 is better than half of the P6's integer
|   performance.
    
    You priced a cheaper Pentium PRO and claimed it had top-of-the-line
    Pentium PRO performance.  You generalized the performance claim from
    a 233MHz EV45 to any EV45.
    
|   Regardless, it's still pathetic.
    
    No, it's not.  Example:
    
    			AlphaStation 255/300	SGI Indy R5000SC 180MHz
    	SPECint95	4.51			4.1
    	SPECfp95	5.71			4.4
    
    I'll leave it up to you to discover that the AlphaStation is $2K
    cheaper.  More performance, less money.  *NOT* pathetic.
    
    								-mr. bill

    re: SGI comparison
    Agreed: Alpha is the one to beat in the RISC Unix crowd; you won't
    see me buying SGI stock anytime soon.
    K

    Can we call it a truce, folks?
    
    			Steve

    Probably - assuming we can agree on the clock speed of the truce.
    

Any chance of NOTES developing kill-file functionality?

    only when somebody forgets to backup the file.
    
    Mark  :-)