1. This forum section is a read-only archive which contains old newsgroup posts. If you wish to post a query, please do so in one of our main forum sections (here). This way you will get a faster, better response from the members on Motherboard Point.

SGI finally on its last legs?

Discussion in 'Intel' started by YKhan, Sep 9, 2005.

  1. YKhan

    YKhan Guest

    "Ernst & Young has advised the Company, however, that its audit report
    is likely to contain an explanatory paragraph with respect to the
    Company's ability to continue as a going concern."

    Silicon Graphics, Inc. Delays 10-K Filing: Financial News - Yahoo!
    Finance
    http://biz.yahoo.com/prnews/050908/sfth177.html?.v=1

    I know it's been said over and over again for the past 15 years that
    Silicon Graphics was about to fold, but with it's near complete
    reliance on Itanium as its processor of choice, it's really been
    cornered into a niche of a niche market.

    YOusuf Khan
     
    YKhan, Sep 9, 2005
    #1
    1. Advertisements

  2. YKhan

    Dan Koren Guest


    More to the point -- it cornered itself.


    dk
     
    Dan Koren, Sep 9, 2005
    #2
    1. Advertisements

  3. YKhan

    John Savard Guest

    If you're making supercomputers, and you can't make your own chips like
    Cray or NEC, the Itanium _is_ the best choice, since it offers the most
    performance *per core*. Also, it has population count - and _byte_
    matrix multiply.

    Of course, people whose applications are highly parallelizable can use
    chips that offer more performance per dollar, and still get their
    answers as quickly.

    John Savard
    http://home.ecn.ab.ca/~jsavard/index.html
    http://www.quadibloc.com/index.html
    _________________________________________
    Usenet Zone Free Binaries Usenet Server
    More than 140,000 groups
    Unlimited download
    http://www.usenetzone.com to open account
     
    John Savard, Sep 10, 2005
    #3
  4. YKhan

    YKhan Guest

    Cray's future is now with Opteron processors, not its own chips.

    Yousuf Khan
     
    YKhan, Sep 10, 2005
    #4
  5. YKhan

    vince Guest

    I think the following 4 ratios are each more important than per core
    performance:

    1) price/performance
    2) performance/socket
    3) performance/watt
    4) performacne/rack

    Opteron wins all 4 of these. I bet we see Opteron growing fast on the
    top500.org now that they have dual-core chips. You want to bet on
    Itanium?

    -- Vince
     
    vince, Sep 11, 2005
    #5
  6. YKhan> I know it's been said over and over again for the past 15 years
    that
    YKhan> Silicon Graphics was about to fold, but with it's near complete
    YKhan> reliance on Itanium as its processor of choice, it's really been
    YKhan> cornered into a niche of a niche market.

    Dan> More to the point -- it cornered itself.

    Forest Baskett was the guy who made the call to go with Itanium. Intel
    made him (and later, many other engineers) sign NDAs which said,
    essentially, that once they'd looked at the Itanium they couldn't work
    on CPUs for SGI ever again. So FB signed the thing, took a look, and
    then couldn't have a really critical conversation with anyone who
    actually knew how to build a CPU. There were a whole lot of people in
    the company that could have professionally taken apart Intel's claims
    for that thing, but the nature of the business arrangements forced SGI
    upper management to make the decision with no technical input.

    You can't fault Ed McCracken for decisions made with no technical
    input: he's a sales guy, it's not his job to know that stuff. But you
    can fault him for getting himself into a position where he had to make
    big decisions with essentially no technical input -- i.e. just Forest
    Baskett's take.
     
    Iain McClatchie, Sep 11, 2005
    #6
  7. YKhan

    John Savard Guest

    You want to bet that nine women can have a baby in one month?

    Yes, Itanium will soon be history if it remains as badly overpriced as
    it is. If Intel is capable of making a multicore Itanium with any yield
    at all, they will also be able, soon, to make single-core Itaniums with
    much higher yield and lower prices.

    Performance per core is important - or, rather, *latency* is important -
    because a great many important applications that can benefit from
    supercomputer power are not fully parallelizable.

    The processors used in the Cray X-1 and the NEC SX-6, for example, are
    significantly more powerful than even the Itanium, however. Although
    they don't get redesigned for process improvements as frequently.

    Multicore, as opposed to multi-chip, offers only one technical benefit -
    faster access to a shared cache. It isn't that often that an application
    can be parallelized, and yet the separate threads still access the same
    memory. The big benefit of multicore seems to come from Microsoft
    licensing policies.

    The Itanium isn't quite a supercomputer-on-a-chip, but we're getting to
    the point where enough transistors are available for that.

    John Savard
    http://home.ecn.ab.ca/~jsavard/index.html
    http://www.quadibloc.com/index.html
    _________________________________________
    Usenet Zone Free Binaries Usenet Server
    More than 140,000 groups
    Unlimited download
    http://www.usenetzone.com to open account
     
    John Savard, Sep 11, 2005
    #7
  8. YKhan

    vince Guest

    Supercomputers today use lots of processors and applications are
    designed for this.
    Itanium instructions take about 2 times the memory of AMD64
    instructions,
    and 2 times the memory bandwidth, and 2 times the instruction cache,
    and 2 times the instruction cache bandwidth. Can't imagine Itanium
    ever
    matching AMD price/performance.
    The Opteron performance per core is not that far behind (even ahead on
    integer).

    Latency makes me think of Pathscale.com's Infinipath. This is a low
    latency
    interface card that lets Opterons use cheap infiniband switches. They
    say
    1.32 Microseconds MPI Latency (8-byte, 1/2 round trip). Seems like
    this
    should do well. What do you think?
    I think the main benefit is better use of transistors. If you have 1
    billion
    transistors and you make one core and cache, while I use 1 billion
    transistors and make 2 cores and 2 caches, I ought to be able to beat
    you
    if I am any good and our benchmark is reasonably parallel.
    I have had some fun emails with them. They say that multi-chip-modules

    count as multiple processors for licensing. They said each die counts
    as a
    processor. I then asked if the soon to be released Paxville from
    Intel, with
    2 die (and I gave them a link to a picture), counted as 2 processors.
    They
    said they were reviewing their policy. If they come up with some
    definition
    where Paxville's 2 die don't make it a multi-chip-module, I will be
    amused
    to see it.
    Any of the current chips is supercomputer level performance from some
    10 or 15 years back. And remember, if Itanium is close to being a
    Supercomputer, then Opteron must already be there.
     
    vince, Sep 11, 2005
    #8
  9. YKhan

    Dan Koren Guest


    I would not try to pin the blame on any one person.

    In the big picture it is completely irrelevant who
    made the decision and how. In fairness one must also
    consider the dilemma these people are/were facing:
    computer vendors with less than $15B/year revenue
    can no longer afford to do processor development.



    dk
     
    Dan Koren, Sep 11, 2005
    #9
  10. YKhan

    Del Cecchi Guest

    I can fault some doof for signing a NDA with such terms. Where were the
    lawyers that should have advised him? Although he should have been wise
    enough not to recognize the problem on his own.

    My guess is that he had decided to transition to Intel for business
    reasons, and the notion that Itanium would be a technical flop never
    occurred to him. Besides, what choice did he have?
     
    Del Cecchi, Sep 12, 2005
    #10
  11. The term applications is ambiguous here. If you mean the problems, they
    are not designed, they exist, and may or may not be amenable to parallel
    solutions (more pedanticaly solutions using parallel computation). If
    you mean the application code, which I assume you do, sometime you can
    do parallel computations and sometimes the problem is linear.

    The big benefit of the Cray2 was that it was fast for linear vector
    computations, and could solve problems which were inherently linear.
     
    Bill Davidsen, Sep 12, 2005
    #11
  12. YKhan

    Alex Johnson Guest

    Choose your words more carefully. Itanium was not a technical flop.
    Well, the original was, but SGI did not abandon MIPS until Itanium II
    was declared the world's most powerful microprocessor for numerically
    intensive applications (ie, scientific computing, SGI's primary client
    base) after knocking the previous title holder, Power 4, from that spot
    in mid-02.

    Alex
     
    Alex Johnson, Sep 13, 2005
    #12
  13. John Savard wrote:

    [SNIP]
    At present, in Opteron land, according to www.spec.org, you would
    need ~1.3 women to have a floating point baby in 9 months. OTOH
    you would need just ~0.9 women to have an integer baby in 9 months.

    That doesn't seem like a big gap to me, a proportion of that gap
    could well be down to cache size disparity (1Mbyte vs 9Mbyte).
    That is not a massively hard thing for AMD to fix, but it is an
    expensive fix to fabricate.
    It has to be expensive, 3 to 9Mbytes of cache is a huge chunk of
    silicon. It would be interesting to see what a K8 core could do
    with 8Mbyte of L2 cache.

    Cheers,
    Rupert
     
    Rupert Pigott, Sep 13, 2005
    #13
  14. YKhan

    Del Cecchi Guest

    You are right, I should have said "MIGHT BE a technical flop", (in that
    it wouldn't meet the needs of a wide customer set) or "MIGHT BE a
    (plain) flop", meaning mostly in a business sense. I guess the jury is
    still out on the issue of Itanium's long term future. At least in my
    view, not that I have any special knowledge.
     
    Del Cecchi, Sep 13, 2005
    #14
  15. YKhan

    prep Guest

    Not so by a long shot. Someone recently replaced 833MHz EV6s with 6
    month ago fastest mostest AMDs. FP throughput is about 1/2 what it was
    at best. Note that any EV6 is not the real bleeding edge anymore.
    --
    Paul Repacholi 1 Crescent Rd.,
    +61 (08) 9257-1001 Kalamunda.
    West Australia 6076
    comp.os.vms,- The Older, Grumpier Slashdot
    Raw, Cooked or Well-done, it's all half baked.
    EPIC, The Architecture of the future, always has been, always will be.
     
    prep, Sep 13, 2005
    #15
  16. YKhan

    vince Guest

    That doesn't seem like a big gap to me,

    Opterons just got about 10% faster in the last couple days. Sun/AMD
    announced some 120 watt Opterons that are like 10% faster. So the
    floating point gap is noise at this point. Opteron's integer lead is
    bigger though.

    The other thing is that when you put 4 Opterons together you get very
    nearly 4 times the performance with no glue chips. This is not so with
    Itanium. So the Opteron price/performance lead is even better once we
    are talking about real systems and not just 1 CPU.
     
    vince, Sep 13, 2005
    #16
  17. YKhan

    Dan Koren Guest


    Certainly not.

    It still is the frying edge though! ;-)


    dk
     
    Dan Koren, Sep 13, 2005
    #17
  18. wrote:

    [SNIP]
    Is that simply a matter of tuning, poor compilers or what ? Sounds
    like a pathological case being tickled.

    Cheers,
    Rupert
     
    Rupert Pigott, Sep 13, 2005
    #18
  19. YKhan

    Hugh Fisher Guest

    MIPS might not have been officially abandoned by SGI, but it was
    definitely pining for the fjords. And there didn't seem to be
    any plan B at SGI if the Itanium didn't ship on time, so the
    MIPS product line was almost stationary for several years.

    From a business sense I don't think Itanium can be considered
    a flop at all. It wiped out PA-RISC, MIPS, and Alpha as
    competitors in desktop/server computing, which Intel might
    consider a worthwhile investment even if Itanium had been
    cancelled altogether in 2000.

    cheers,
    Hugh Fisher
     
    Hugh Fisher, Sep 14, 2005
    #19
  20. YKhan

    Grumble Guest

    Most codes will fit in the Icache. Why do you mention memory bandwidth
    alongside Icache bandwidth?

    What you said is mostly true, however, keep in mind that IA-64 is a
    three-operand ISA while IA-32 is a two-operand ISA.

    IA-64: C = A + B
    IA-32: A = A + B

    If you still need the old value of A, you'll have to copy it to another
    register.
     
    Grumble, Sep 14, 2005
    #20
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.