Conference 7.286::digital

    re .0
    
    I agree - for PC Magazine it was quite positive.  And yes, they are
    clearly WIntel oriented.  But keep in mind, WIntel is the desktop
    market and about 90-95% of the server market.
    
    With recent announcements for Alpha, especially the 21164PC chip, there
    is a glimer of hope to make a dent.  Digital has at least four advantages 
    over the failed PPC and MIPS: 1) Alpha does have a performance advantage
    over Intel (please, no ratholes on this!), 2) there ARE a lot of native
    apps for Alpha-NT, 3) FX!32 promises to fill a lot of gaps, and 4) we
    get very good reviews for our Windows NT service and support.
    
    When you combine the above around Alpha-NT workstations that will
    compete head-to-head in price with WIntel, you have a compelling story. 
    Keep in mind Intel seems to be falling short of everyone's expectations
    right now; Merced is a year behind schedule and Klamath is delayed as
    well with rumors of performance not meeting expectations.  This may be
    perfect timing for Alpha-NT!

	My neice writes for PC Magazine.  I wonder if she wrote the article.
	Is it on a Web Site ?

	Regards,

	Chuck

    
    
    Re: -1
    
    It's the *cover story* and there are several authors credited.  It's
    not on the PC Magazine web site yet.
    
    Kevin
    

As soon as there's a web site that carries the article, I think
we should get a copy for our very own http://www.digital.com/ or
even better... reserve an advertising spot on altavista.digital.com!

-Richard

[Posted by WWW Notes gateway]

    Computer Shopper also has an article online this month on FX!32. 
    www5.zdnet.com/cshopper/frames/techmain.html

    Computer Shopper and PC Magazine are both part of ZDNet.
    
    -Dom

    It is now online.  See http://www.pcmag.com/features/grworks/rev3.htm.
    
    John
    

    
    A few issues to keep in mind when discussing Intel vs Alpha for
    servers.
    
    Reference NT 5.0 whitepapers on Feb'97 Technet CD's.
    
    NT 5.0 is going to introduce a whole pile of additional encryption
    capabilities. So, while it is very difficult to compare Intel and Alpha
    for normal file / print as there is comparitively little cpu
    involvement in moving an IO from disk to PCI to NET to desktop, this
    will change significantly in NT 5.0.
    
    The new sequence will be disk to PCI to cpu encrypt/decrypt to NET to
    desktop.
    
    While not all IO's will involve encryption/decryption, from white
    papers from MS, the typical workload in NT 5.0 will be significantly
    more CPU intensive.
    
    Add to this the 64 bit features in NT5, desktop standards reaching
    server speeds (govt dept here figures no more money for next 2 years,
    so their new std desktop is mix of P166's and PPRO 200's with 32M
    memory), TCP/IP IPV6 now in beta (even getting MS attention in white
    papers) which is 64bit aligned architectually, more Alpha native apps 
    appearing very rapidly (Word, Excel, VB and VC++ soon) and the bottom 
    line question to ask Customers is :
    
    "How long do plan this new server to last ? By spring 1998, PPRO's will
    be desktop std for large companies, so at what point do you plan 
    to view 32 bit as a client platform and 64bit as the server platform ?"
    
    Why wait ? Invest now in 64bit servers and you will automatically be
    ready for new NT5 VLM and encryption issues ..
    
    <time to don hard hat..>
    
    :-)
    
    / Kerry
    

    Actually, CPU utilization on Alpha servers isn't any better than
    Intel...  for example, see our Microsoft Exchange server testing
    results at
    http://frodo.eng.pko.dec.com/info/alphaserver/solutions/mail/
    flash_497.htm
    
    2150 users on a 1CPU Alphaserver 1000 (400Mhz Alpha): 72% utilization
    2100 users on a 1CPU Prioris ZX (200Mhz Pentium Pro): 59% utilization
    
    Perhaps Intel is better suited for the CPU intensive encryption in
    NT V5?  ;-)
    K
    
    
    
    

    
    
    Some sample quotes from the PC Magazine article:
    
    
    "When going head to head with the other Windows NT workstations,
    Digital's entry is the clear speed demon. While running our AutoCAD
    test, for example, the Personal Workstation 433a beat the x86 systems
    by huge margins. We also found when running Microsoft's Softimage that
    the Personal Workstation 433a crunched numbers better than the others
    did. But remember, you need native Alpha apps, which are hard to come
    by." 
    :
    :
    		
    "If Windows NT is your OS but the CPU doesn't need to say
    "Intel," Digital's Personal Workstation 433a, based on the same
    vendor's 433-MHz Alpha 21164 processor, is an outstanding machine.
    Under the applications we tested with native Alpha ports, the Digital
    box blew away the competition. The challenge for this platform,
    however, isn't hardware--it's software. Of the ten applications we ran
    on our test machines, only four were available for the Alpha (we ran
    two productivity apps in emulation mode under Digital's FX!32). Digital
    is investing heavily in development for this platform, but the
    challenge will be in finding the ISVs to support it."
    :
    :
    
    "Native software application support is critical for the best
    performance of the very demanding applications found in the workstation
    environment. A cursory examination of our performance test results will
    verify this fact. For tests in which there were native Alpha versions
    of the apps available, the 433a performed significantly better than the
    other tested Windows NT machines (the dual Pentium Pro-based Compaq and
    HP workstations). Under AutoCAD, the x86-based Windows NT systems took
    almost 50 percent longer to complete the test than the Alpha did.
    
                 
    "On tests where some cross-comparison was possible, such as
    Pro/Engineer, the Alpha unit was nearly twice as fast as the Sun unit
    and shaved a third off the SGI O2's timing. On the Softimage Mental Ray
    Render test, the Alpha unit shaved almost 30 percent off the timing of
    its next-closest competitor, the HP Vectra XW. This kind of performance
    improvement is why users will purchase Alpha-based workstations;
    however, it's also their Achilles' heel. To get the most out of an
    Alpha-based system, you need applications compiled for native Alpha
    code, and there simply aren't that many of these apps available. (We do
    not publish results for this machine on the PV-Wave and Photoshop
    tests, because the software has not been ported to the Alpha
    platform.)"
    
    
    

    
    
    
    We should be screaming "Alpha blows away the competition!  Thousands
    of native applications now available!" and pointing to the PC Magazine
    web site and an online list of native Alpha apps!
    
    Take credit for every "powerhouse with room to grow" compliment and
    talk up the "new native applications available" part. 
    
    Maybe even show some of the magazine performance charts on our web
    site and in printed advertising! 
     
    There are *WAY* too many positives in this review to not get some
    good publicity from it!
    
    We need to do it NOW, not a month from now!
    
    Kevin
    
    
     
      

    re: -.1
    
    Yawn... Strech... Yawn...
    
    We'll get around to it when our marketing people don't have anything
    else to do.... 
    
    You don't push a thing like marketing....
    
    We'll market no good news before it's time...
    
    HAS ALMOST PAST...
    
    Until then I'm going to be screaming about this to every single
    customer I have...;-)
    
    This is great news PC Magazine even refused to automagically 
    give the Editor's choice award to a Wintel system...
    
    We are making some progress... Now if we would only advertize 
    our stunning performance for native applications over the 
    WINtel workstations we'd be golden.. We might even attract 
    a couple more native ports...
    
    WE wouldn't even have to say that we don't have all the applications
    ported to alpha (after all it's OUR commerical...)
    
    We should be the dominant workstation in the autocad, proengineer, and 
    PVwave space...
    
    But of course we'd have to convince a few more software houses to 
    do the "Port"/Recompile on Alpha...
    
    Give them an Alpha, Give them an Engineer.. Get the applications on 
    Alpha at any cost... Every major WNT app vendor should be on our 
    platform or have to explain why they aren't doing a port.. To 
    thier customers!
    
    They still point at Digital and say "I'd have to buy a workstation,
    software, and I have no expertise..." Digital should have to work 
    this.. and work it now...
    
    Not all the 25k applications have to be there... just target the 
    best seller in each software catagory...
    
    JMHO
    
    John W.
    

    I agree with .11 ie. establish best of class in each category and you
    will automatically get the others as they will not even be in same
    game ... 
    
    As an example of this, Oracle did the initial 64bit db benchmark. Prior 
    to this other db vendors were saying "some time in future, but no demand 
    yet..". HOWEVER, as soon as Oracle released their huge numbers, the
    others realized that they had just been sent down to the minors. They
    came to our door and started kicking at it for us to let them in to.
    
    Rumour has it that Oracle is looking at same strategy for NT 5 and VLM. 
    It will establish them as the leaders in NT, UNIX and OpenVMS platforms
    in end-to-end database solutions.
    
    [p.s. using Exchange as a comparison for cpu performance is not valid,
    because the reality is that Exchange 4.x does not scale very well on
    Intel or Alpha platforms. There were known issues in 4.x that Microsoft
    has started to address with Exchange 5.0. To put reality into this
    picture, we recently had a situation whereby the Customer wanted to
    move 32,000 users to a new mail system. Local Microsoft techie in chg
    of proposal is rumoured to have proposed Exchange with 100+ Intel
    servers, because he was not "comfortable" with Alpha. HP came in with
    OpenMail and proposed 4 servers. Whats wrong here ?]
    
    My line to Customers these days is that 32 bit is great for client
    configurations, but best value based on cost / performance and 
    ability to take advantage of future software optimizations is going
    to be 64bit based.
    
    Keep in mind that NT5 VLM capabilities is only one of the area's that
    will take advantage of 64 bit - others include IPV6 (64 bit aligned
    architectually and now in beta), numerous ecryption algorithyms,
    graphics design and rendering and watch for 64 bit java capabilities
    and applets very quickly.
    
    Oh yes - if Cust thinks P7 will give them 64bit capabilities in 1999,
    point out to them that P7 will run x86 app's in 32bit emulation mode
    until those app's are ported to the P7 IA-64 from Alpha. This could
    take an additional 6 months (optimistic) to 18 months (realistic)
    beyond 1999. Since 850+Mhz to 1Ghz EV6-7 Alphas will likely be the 1999
    or year 2000 competition, those P7 guy's have their work cut out for them !
    
    [boy and this started out as a quick reply :-)]
    
    Time to buy DIGITAL stock..
    
    Regards,
    
    / Kerry
    
    

    Well according to the word on the external streets, if it's Autocad
    we're using to compete on Alpha, we better HAD be very quick about it,
    because the next release will be Intel only.
    
    Try AltaVista news search for +Autocad +R14 +Alpha.
    
    regards
    john

    Yes, thats true the next release of AutoCAD (14) will be Intel only but
    the CAD tool in the article is Pro/E and Alpha is the way to go by
    FAR!!! Pro/E and Alpha will be around for the forseeable future and it
    the PW433a or PW500a blows the doors off of any Intel box, just look at
    the acticle posted in the Miata note conference.
    
    			Bart Lasky 

    What's the relative market share for AutoCAD vs. Pro/E?
    

    re: .13
    
    Using search argument of:
    
    +Autocad +R14 +Alpha
    
    Resulted in:
    
    Word count: R14: 1199; Autocad: 4395; Alpha: 54178
    No documents match the query.
    
    -----------
    
    I remember some questions about this in one of the news groups.  Is
    it actually true that Autocad is dropping Alpha support?  Or is it
    just another one of those vicious internet rumors?
    
    Hugh
    

    Re: .15 (Market share question)
    According to Daratech, Inc. (market research firm), the 1996 revenues
    in the mechanical CAD/CAM/CAE market:
    
    PTC (Pro/E)		$660M
    IBM/Dassault	 580M
    Computervision	 298M
    SDRC (I-DEAS)	 275M
    EDS/Unigraphics	 231M
    Autodesk (AutoCAD)	 222M
    Cadam		 155M
    MacNeal-Schwendler	 147M
    Matra Datavison	 114M
    
    So it looks like Pro/E has about triple the dollar volume of A-CAD. 
    PTC also had the largest number of license renewals, recurring
    licenses, and maintenance $$$. ("Maintenance" is the term for service
    contracts, etc., which allows you to call PTC for tech support. It also
    covers upgrades, I believe. It does not necessarily imply that Pro/E
    crashes and needs to be fixed, for example.)
    
    Rich Krause

RE: .11

	It's about time we drop "Whatever it Takes" and change it
	to something like "Digital: The fastest computers on the planet"
	or something like that. Enough with the damned "branding" stuff.
	We'll make our "brand" by marketing and advertising our wares.
	Microsoft does. HP does. Even the old stogy IBM does.

							mike

    .18 and if ever we made a machine clocked at 666 Mhz, even *our*
    marketing people should be able to come up with some neat names and
    ads...
    
    
    .nah..

    
    	Yeah, the best I'd expect is "The devil is capable of anything it
    	takes".
    

,-----even being nice, caring and loving?

PTC moved into the S&P 500 last week.

AutoCAD may sell more, but Pro/E and Pro/Jr make more money.

    But is the 4:24 time for the Alpha on the Pro/E test in the
    PC Mag review a convincing enough win over the 4:44 time posted
    by the Intel boxes (HP and COMPAQ)?  That's exactly 10%.
    
    P.S. Did PV-Wave drop Alpha support too?  It used to be there
    like AutoCAD.
    Kratz 

    re: AutoCAD vs Pro/E
    
    Much of this discussion misses the point, IMHO.
    
    The problem is that there are gobs of AutoCAD users out there who will
    now have little or nothing to gain by investing in an Alpha box.
    
    Since we've made a marketing plan by defining all the business that
    we're NOT interested in (e.g., "we don't do mom & pop shops", "we don't
    do low-end workstations", etc.), we should be going like GANGBUSTERS to
    win and keep the business that we do want (e.g., compute-intensive
    CAD/CAM workstations).
    
    We should be fighting to keep AutoCAD in our arsenal of apps.  Give 'em
    an engineer for free.  Do that "whatever it takes" stuff to keep us
    from losing marketshare in the niche that we've designed for ourselves.
    AutoCAD's discontinuing the Alpha version of its software sends a loud
    and clear message to its users:  "Alpha just ain't worth it."  No
    amount of our work by our internal spin-doctors will succeed in making
    this a positive message.
    
    If we want to win, we've got to fight to win.  Covering over the dead
    bodies on the battlefield with blankets that read "This is not a dead
    body, we won!" just doesn't cut the mustard.
    
    -- Russ

    I asked the market share question primarily because I've heard of
    AutoCAD for years and years, but not Pro/E.  I would think as platform
    vendors, we'd care more about total number of units shipped, as opposed
    to revenues, though.  Is there any information comparing total units
    shipped for each of these products, regardless of application (the
    numbers quoted where for mechanical CAD, right?).
    
    tim
    

    .25 Yup, bingo.
    Along the same lines, if a native Alpha version had made it into
    Photoshop V4, that would have been good for several thousand
    high-end commercial desktops almost overnight (most of them
    former Apple owners...)  .02 K   

    .26 I agree. 
    
       erik

>>    But is the 4:24 time for the Alpha on the Pro/E test in the
>>    PC Mag review a convincing enough win over the 4:44 time posted
>>    by the Intel boxes (HP and COMPAQ)?  That's exactly 10%.
    
      In the most recent Pro/E benchmark run by the Pro/E magazine
      (March '97) an Alpha personal workstation 500a blew every
      pentium pro based system out of the water by a factor of *2*.
      
      The PW500a ran the suite in 56 minutes, PW433a in 64 minutes,
      and various Pentium pro's in 101-111 minutes.
    
      This benchmark is the industry standard Pro-E benchmark, and is
      directly based on the tasks a typical designer performs on a 
      typical day.
    
      Now we just need to sell lots of them!!  

    Yes, the problem is, a few million PC Magazine readers were just
    convinced otherwise.  A letter-to-the-editor at PC Mag with a
    pointer to the Pro/E Mag's official Bench97 results couldn't hurt...
    worst case it just doesn't get published.
    K

    Given the market breakdown mentioned previously, there is sufficient
    fragmentation to suggest that we HAVE to have Alpha be well supported
    on more than just the number 1 application. The workstation notesfile
    is littered with sales calls to catia (Dassault/IBM) sites which have
    crashed and burned. Although the revenue share for Autocad is 1/3 of
    Pro/E it is still not an insignificant share of the market.
    
    I agree with Russ, we should be persuing every avenue to being
    successful in this chosen marketplace. If we focus merely on being
    successful with a chosen product, we will potentially miss out on a lot
    of success.
    
    According to .17, Pro/E has the largest share of the market, but its
    only 24.6% of the CAD/CAM market. Many CAD shops are highly parochial
    in their choice of tools. I think we are remiss if we dont seek to
    optimise the potential gains in a market segment where we feel we can
    succeed.
    
    - John
    
    
    
    
    
    

    Note that the Digital Personal Workstation 433a in the article is our
    *entry* level machine!  without an optional Cache!

    No one in my group or any group around here is suggesting to just let
    AutoCAD drop off the plate.  We are pushing to keep Alpha supported on
    AutoCAD.  
    
    If the truth be known the difference in performance between a
    Petium based system and a Alpha based system running AutoCAD is not
    that great, you have to remember that AutoCAD is designed to work on
    any PC with 16MB of memory, while Pro/E is a high end cad tool.
    
    				Bart

>>    If the truth be known the difference in performance between a
>>    Petium based system and a Alpha based system running AutoCAD is not
>>    that great, you have to remember that AutoCAD is designed to work on
>>    any PC with 16MB of memory, while Pro/E is a high end cad tool.
    
	This leads to the question: "Is the paltry difference between
	Intel and Alpha for Autocad due to the fact that Autocad/Alpha
	is a recompile with #define ALPHA use only keep it working at 
	all?"

	My suspicion is that most applications that go through a mere
	recompile will probably not achieve the gains that an Alpha
	specific redesign might - byte/word boundary problems and the
	other architectural optimizations can kick us in the teeth when
	only a recompile is done.

	Anyone care to shed more light or debunk this thought?

	- norm

    Well I wasn't shedding too many tears over the loss of Autocad for this
    very reason. Pro/E's heritage is UNIX and presumably portable and flat
    memory model and such. AutoCAD has years of tuning to get the most out
    of the x86/DOS/windows memory weirdness - which almost certainly will
    lead to lots of byte accesses which most ALphas out there won't be too
    good on.
    
    regards
    john

    re .8
    
    The AS100 also had half the memory and less than half the number of
    spindles. That won't have helped with latency and paging.
    
    One note mentions "...for systems up to 1000 users" and "for
    systems that used more than one disk for information storage". The
    graphs and tables displayed make no mention of systems with these
    conditions, so I expect that what is here is a summary of a much larger
    document.
    
    It also mentions the use of RAID controllers and that the AS4000 used
    Adaptec 2940 w/uw. I don't believe these are raid controllers, so did
    they substitute NT raid on this system? This would cost something in
    CPU performance.
    
    [I know, you are just taking a shot based on what was available...and
    it certainly is interesting to see a pair of I6200's in the ballpark
    with a 400Mhz Alpha.]
    r
    
                         

    re .35
    Well, ok, there's more shots available... ;-) (albeit at least
    a year old now, this shows Alpha server CPU utilization no better
    than Pentium-based servers with the classic file and print server
    workload).  K  
    
    The following is a summary of results from a series of file and print
    performance characterizations...
    
Client populations listed are at an average of 80% CPU utilization on the
server.

                           Simultaneous
   Platform                Active Clients
   --------                --------------

AlphaServer  400 4/233        325 Clients
AlphaServer 1000 4/266        350 Clients
AlphaServer 2100 5/250        675 Clients
  (single CPU)
AlphaServer 2100 5/250       1300 Clients
  (dual CPU)
Prioris HX 5133 DP            400 Clients
  (single CPU)
Prioris HX 5133 DP            650 Clients
  (dual CPU)
Prioris ZX 5133 MP           1000 Clients
  (single CPU)
Prioris ZX 5133 MP           1700 Clients
  (dual CPU)


Configs
-------
All: Windows NT Server V3.51, Build 1057, SP3-SP4


AlphaServer 400 4/233

Memory:  160 MB memory
Disk:    SCSI
         MAXTOR 7345-SCSI - System disk
         RZ28 - File Service
         RZ26 - Applications (for Remote tests only)
              - Mail files
Network: Ethernet, 1-2 segments, DE435's


AlphaServer 1000 4/266

Memory:  512 MB memory
Disk:    SCSI
         RZ28 - System
         RZ28 - File Service
         RZ26 - Applications (for Remote tests)
         RZ28 - Mail files
Network: Ethernet, 1-3 segments, DE435's


AlphaServer 2100 5/250

Memory:  256 MB memory
Disk:    SCSI
         RZ28 - System
         RZ28 - File Service  (also tested RAID, up to 6 x disks)
         RZ26 - Applications (for Remote tests), print spool files
         RZ28 - Print spool files
         RZ28 - Mail files
Network: Ethernet, multiple segments, DE435's, also tested DEC FDDI PCI


Prioris HX 5133DP

Memory:  64 MB memory
Disk:    SCSI
         RZ28 - System
         RZ28 - File Service
         RZ26 - Applications (for Remote tests)
         RZ28 - Mail files
Network: Ethernet, 1-3 segments, DE435's


Prioris ZX 5133 MP

Memory:  256 MB memory
Disk:    SCSI
         RZ28 - System
         RZ28 - File Service RAID, 6 x disks
         RZ26 - Applications (for Remote tests), print spool files
         RZ28 - Print spool files
         RZ28 - Mail files
Network: Multiple Ethernet segs to DEC FDDI PCI



Client Configuration:

Software: Microsoft Windows V3.1
          Microsoft Office V4.2
          Microsoft Word V6.0a
          Microsoft Excel V5.0a
          Microsoft Powerpoint V4.0
          LOTUS cc:Mail V2.10

Hardware: 486 class PC's (up to 200 of these)
          - 16 MB of memory
          - 1 Local disk
          - Network device - DE205 (EWRKS driver)
                             DEPCA (DEPCA driver) or
                             DEC EISA (DEPCA driver)

    
    
    I don't get it.  PC Magazine says that the Alpha is the fastest 
    by wide margins, but we feel compelled to prove the magazine wrong?
    
    What's wrong with taking a bow and selling something?
    
    Kevin
    
    

Sell something?  You're obviously working for the wrong company!

				Steve

    re .36:
    
        I recall seeing a presentation by Dick Sites showing work he'd done
        that revealed huge inefficiencies at the instruction level in the
    	way WinNT (I think it was v3.5x) ran on Alpha as compared to Intel.
    	Some of it was due to Alpha not having byte-level granularity...
    	I wonder if the efficiency has gone up with NT 4?
    
           erik g.
    

    
    Boy, we get good press and so many people want to shout "but, it's
    not as good as it appears !!"
    
    I've said it before, but we are our own worst enemy ..
    
    p.s. keep in mind that EV6 is supposed to have more byte oriented
    capabilities, so it should be interesting when it appears later
    this fall..(or sooner?)
    
    Bottom line is that NT is a 32bit OS running on a 64bit hw platform. To
    date, NT has been optimized for a 32 bit hw (Intel) because the demand
    was not there for high end 64bit stuff. However, if anyone has seen
    recent stories about how frustrated Gates is with continually being
    positioned as desktop and departmental only, then you will quickly
    realize that this is going to change.
    
    Microsoft knows that if they wait until 1999 for 64bit and really
    high end cpu capabilities, then they might as well retire about 1/2 
    of their high end plans right now.
    
    Microsoft wants 64bit NT more than DIGITAL right now. All of their
    high end prod's (SQL, TP, Steelhead etc) are not going to compete
    with the likes of Oracle, Sybase etc unless they get NT reved up
    a whole bunch ... As an example, Gates in Information Week this
    week is quoted as stating he wants SQL to do 1 billion transactions
    per day in the very near future ..
    
    Only problem Microsoft has is that they don't have many folks that
    can comprehend what 64bit really means in high end. This is why
    Oracle should do really well in NT space for next little while ..
    
    Now, if we could only convince the Msoft types to tweak NT in a few
    discreet places..(more than just VLM) then we AND Microsoft will see 
    better results.
    
    My $.02 anyway,
    
    / Kerry

    Yes some proper code tweaking does help ... witness the Lotus Notes
    results where the fastest Alphas DO show an advantage over the fastest 
    Intel processors.
    
    

    Re: .41
    Re: Byte granularity.
    
    Does not the current EV 56 have byte level instructions?
    
    Bill
    

Yes, The EV56 (shrunk 21164, sometimes called 21164A, but not by
Digital Semiconductor) supports byte & word memory reference instructions.
This chip has been shipping since at least last fall (northern hemi).
The PCA56 (21164PC) chip will also support byte & word instructions, along
with MVI (multimedia) instructions.  This is the cost-reduced EV5-based
processor that's priced at $295-$495 for speeds of 400-533 MHz sampling
now with volume this summer (northern hemi).

So, any system running faster than 350 MHz contains a processor that
provides byte & word memory reference instructions.  For example:
  PWS 433a and faster		  (Miata)
  AlphaStation 500/400 and faster (Bret)
  AlphaServer 2x00/375 and faster (Gamma)
  AlphaServer 4x00/400 and faster (Rawhide)
  AlphaServer 8x00/440 and faster (Turbolaser)

Some of the above systems don't support byte & word access to I/O
space.  I believe all systems built with EV6 processors will support
byte & word access to I/O, which might make it easier to port drivers
originally written for x86.

    OK. My experience is more along the lines of VAX's where when a new CPU
    came out you had a release of VMS with new sysloann files and whatever
    else was required.
    
    Well, here's a $64 question. By adding CPU's with byte instructions,
    will ...
    a) code (OS or App) have to be re-compiled
    b) the HAL understand that this CPU can do these things
    c) a minimum rev of NT (3.51 + sr x)
    d) minimum rev of application
    ...or is this automagical somehow?
    
    Rob Wall
    
    [Not trying to start a technical rathole here, reply off line if you
    like. [email protected]]
    
    
    

    
    The instructions are *new* so, except for I/O using sparse space,
    applications do not *have* to change.  Applications can be recompiled
    to take advantage of the new instructions using the architecture
    switch.  In addition, "fat" binaries are being investigated to allow a
    single image that is "fixed up" at runtime to allow "better"
    performance when the instructions are available (but not perhaps as
    optimal as a image optimized for the instructions).
    
    The EV56 platforms so far have provided sparse I/O space, allowing them
    to be compatable with the EV5 platforms.  On EV6, with one exception
    that I know of, sparse space will no longer be implemented on
    platforms, potentially requiring device driver changes, especially
    graphics drivers which tend to directly touch I/O space instead of
    using O/S callbacks.  The HAL or HAL-like layers in the O/S must
    support the flat I/O space, but this should be little different than a
    new platform.
    
    The O/S "could" be optimized for the new instructions.   I don't know
    what the NT, and UNIX plans are, but I do not think VMS will provide
    byte-optimized kernel code... although there may be a potential to do
    specific images.
    
    

    Backward compatibility is probably the main inhibitor to takeing advantage
    of newer instructions. How long weren't people compiling windows software
    to 386 compatibility? Only practical way would be inherent (no worry)
    compatibility, either with emulation on machines where it may be missing
    or smart executables and/or run-time optimizers. Maybe its time for 
    FX!64? :-)

    >Per

    Metatools has ported Bryce 2 to the Alpha.
    
    But they must be wrong about Alpha being faster too.
    
    
    								-mr. bill

    I just saw the cover of PC magazine this month:
    
    See the Fastest Workstations in the World reviewed it barked and 
    begged...
    
    Then what was on the cover an SGI and an HP PC/Workstation...
    
    What does this say by default...
    
    How much would it have cost to get our Workstation on the Cover...
    
    Why don't we have marketing people who's only job is to make sure
    that we get front cover/inside coverage of our stuff in these RAGS...
    
    You can lead a corporation to town but you can make it market...
    
    JMHO
    
    John W.
    
    

|    Then what was on the cover an SGI and an HP PC/Workstation...
    
|    How much would it have cost to get our Workstation on the Cover...
    
    We'd probably have to be one of the two market leaders.  Like SGI and
    HP.  
    
    We're not.  
    
    I think you overestimate the power of marketing.
    

    On the other hand, when the magazine cover reads "See the fastest
    workstations in the world," it would be nice if they showed them.
    DIGITAL's market position and advertising aside, it's frustrating to
    see things like that.

    re: -.2
    
    I think you underestimate the power of marketing and advertizing...
    
    Take PC magazine... Have a 12 month advertizing contract for 
    several pages each month... Think they'll put a non-advertizing
    competitor on the cover when your box is close or best in the 
    running?
    
    Yes advertizing and marketing is expensive.. 
    Yes we need to dedicate people to do this. Our competitors have such
    marketing teams with their goals being the amount and type of 
    coverage portraying their products by the RAGs and by the Analysts.
    
    Of course you also have to go where the HQ of the RAG and not send 
    internal E-mail to the Digital Network or post flyers in all the liquor 
    stores and daycare centers in the GMA...
    
    Of course that's just my opinion.. I could be wrong...
    
    John W.