Consolidated Business & Enterprise Computing Rant Thread

Discussion in 'Business & Enterprise Computing' started by elvis, Jul 1, 2008.

  1. itsmydamnation

    itsmydamnation Member

    Joined:
    Apr 30, 2003
    Messages:
    10,687
    Location:
    Canberra
    no IA64 was shit is shit and always will be shit!

    To paraphrase the prophet Linus Torvell , anyone who thinks they can get general performance improvements by pushing complexity from hardware into compliers and software is an idiot. IA64 is VLIW , and VLIW has no OOOE has terrible performance on branches, requires static scheduling and just all around sucks , there is a reason even in GPGPU NV is effectively SIMT VEC2 and GCN/RDNA is SIMDx4 + scalar.

    The whole x64 is shit cuz reasons are largely BS , x64 has a power/complexity cost at decode stage because of variable length instructions and thats it. All the random arse old x87 etc all get handled via microcode rom. For Servers there is no compelling reason to switch from x64 to ARMv8 in multi threaded scalar code because @ max socket performance they are equal , x64 is well ahead in single/light thread performance thanks to better turbo/DVFS and x64 is much better at SIMD.

    The bigger issue for ARM is given all the company that do ARM at high revenue have abandoned ARM in the server space ( qualcomm, samsung, broadcom) how are ARM server entrants going to survive the decade they need to get to sufficent market share , the hyper scalars will just roll there own so there is no market there.
     
  2. OP
    OP
    elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    44,036
    Location:
    Brisbane
    Blah blah all high end compute will be done on specialist GPUs/FPGAs (or even quantum computers) with hypervisors babysat by power efficient boring CPUs by 2030 or I buy you a nice steak dinner.

    x86 dies for high end compute by the end of the decade, relegated only to enterprise shitware like Mainframe is today.

    I'm betting big on Nvidia to utterly dominate HPC.

    Screenshot_20210415-155429.png
     
    Last edited: Apr 15, 2021
    RaZ likes this.
  3. cvidler

    cvidler Member

    Joined:
    Jun 29, 2001
    Messages:
    14,964
    Location:
    Canberra
    show us where IA64 touched you.

    it was not meant to replace the consumer space, it was meant to take over server compute, where you can spend the resources on the code to get it optimised.

    and to think compiler complexity/optimisations are bad, well, there's compiler optimisations/complexity for every different type of x86 there is today (and there's dozens of them), every release, there's another special instruction here and there, AMD do it one way, Intel does it another. compiler optimisations. the end result is unless you compile your own code, you get some generic compiler output targeted at the lowest common denominator, which probably means you're running code optimised for broadwell.
     
  4. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,413
    Location:
    Brisbane
    C'mon, maybe in a edge cases, but let's be honest, shitty inefficient, unoptimised code makes the world run
     
    bcann and Rass like this.
  5. cvidler

    cvidler Member

    Joined:
    Jun 29, 2001
    Messages:
    14,964
    Location:
    Canberra
    and shitty unoptimised code gets us processors full of shitty unoptimised optimisations to run it and we get <insert current long list of processor vulnerability hotness>.

    I'm certainly biased on the topic, my job/career for over a decade has been around performance optimisation and making developers cry when I can easily (and with automated tools 99% of the time) point out huge benefits to be made.
     
  6. m0n4g3

    m0n4g3 Member

    Joined:
    Aug 5, 2009
    Messages:
    3,728
    Location:
    Perth, WA
    While I generally agree with what you are saying... I hope it's not nvidia that dominate that marketspace. They already are shady, slimey assholes they don't need more reasons to be shadier, slimier assholes.... :p
     
    karnophage likes this.
  7. OP
    OP
    elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    44,036
    Location:
    Brisbane
    Zero disagreement from me. And that's, in part, why I'm betting on them.

    Nobody nice makes it far in the tech industry.
     
    Fred Nurk, mooboyj and Rass like this.
  8. Fred Nurk

    Fred Nurk Member

    Joined:
    Apr 5, 2002
    Messages:
    2,244
    Location:
    Cairns QLD
    Haha. One Rich Arsehole...
     
    elvis likes this.
  9. GumbyNoTalent

    GumbyNoTalent Member

    Joined:
    Jan 8, 2003
    Messages:
    9,827
    Location:
    Briz Vegas
  10. QuakeDude

    QuakeDude ooooh weeee ooooh

    Joined:
    Aug 4, 2004
    Messages:
    8,545
    Location:
    Melbourne
    Anyone know a GOOD Wyse Management Suite expert, preferably Melbourne based? Shit is driving me crazy. We know enough to hack our way through, but not enough to troubleshoot things properly. Dell won't provide someone for T&M as they want a proper specification with deliverables... apparently "make it work" isn't acceptable lol
     
  11. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    14,504
    Make it work is Vague.
    We use WMS to manage a bunch of 5070's and 5040's.

    I liked the Teradici management Console (for PCOIP) better, but they fucked it with a batshiat insane licensing model.

    What's the issue you're having with WMS? It's been a while since we set ours up, but its very much set and forget.
     
  12. QuakeDude

    QuakeDude ooooh weeee ooooh

    Joined:
    Aug 4, 2004
    Messages:
    8,545
    Location:
    Melbourne
    I know.. probably doesn't help that I used that phrase to summarise a bunch of smaller isssues that were all contributing to an overall frustration with the product.

    I've managed to work out a few of our issues, but the one key frustration is that when I unbox a thin client and setting it up at site after being imaged centrally, it doesn't check in. It says its registered in WMS, but WMS doesn't see it. If I go into the WMS client, unregister then reregister it it shows up in the main console. Its probably something I did wrong during the sysprep phase as I think the thin I was working off as the master was registered to WMS at that point.
     
  13. Luke212

    Luke212 Member

    Joined:
    Feb 26, 2003
    Messages:
    10,074
    Location:
    Sydney
    HPC is already GPU dominated so whats this end of decade thing? Its only your stuff requiring high RAM thats not GPU. And its only high ram because they are lazy.

    and the GPGPU is a misnomer. there is nothing general about a gpu. its highly specialised. intel and amd are the ones making GPUs.

    the whole thing revolves around given parallelism (nvidia) vs discovered parallelism (intel) and this is problem driven, not technology driven.
     
    Last edited: Apr 20, 2021
  14. looktall

    looktall Working Class Doughnut

    Joined:
    Sep 17, 2001
    Messages:
    26,591
    how are you managing registration?
    Are you using groups for different sites and a group token pushed from DHCP scope options?
     
  15. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    14,504
    We don't have any site specific requirements, so I've not had cause to try multiple sites. I'll have a look when I'm back in the office next week,
    we use DNS, SRV records to hand out WMS bits and bobs, with the _WMS_GROUPTOKEN record set to the main group.
    If we have bespoke needs, at the moment, we just move the 2 or 3 terminals we need, into a subgroup for testing.
     
  16. Cape_Horn

    Cape_Horn Member

    Joined:
    Dec 23, 2001
    Messages:
    2,470
    Location:
    Shooting Baker
    Umm...
    "Its only high ram because they are lazy"

    Do you not have large datasets and/or tight SLA's?
     
    elvis likes this.
  17. EvilGenius

    EvilGenius Member

    Joined:
    Apr 26, 2005
    Messages:
    10,791
    Location:
    elsewhere
    With zero latency wifi fed by 10gb token ring networks there's no reason to hold your data sets in ram.
     
    GumbyNoTalent and elvis like this.
  18. power

    power Member

    Joined:
    Apr 20, 2002
    Messages:
    66,236
    Location:
    brisbane
    trivial to implement, i did in my spare time 5 years ago.
     
    elvis likes this.
  19. cvidler

    cvidler Member

    Joined:
    Jun 29, 2001
    Messages:
    14,964
    Location:
    Canberra
    you should commercialise it, it'd be a gold mine in the HFT market.
     
  20. power

    power Member

    Joined:
    Apr 20, 2002
    Messages:
    66,236
    Location:
    brisbane
    why should I anyone can do it, i don't feel like it's worth my time.
     

Share This Page

Advertisement: