back to article Next Superdome CPU chips amble into HPE

There have been reports that Intel has already started shipping its latest, Kittson version, Itanium processor chip. The current Itanium processors are the 9500 series, code-named Poulson, and introduced in 2012. There are four versions: 9520, 9540, 9550 and 9560. HPE uses Itanium chips in its high-end Superdome 2 servers, …

  1. cray74

    With Due Respect to Eric Idle...

    Itanium: I'm not dead.

    Scrap Collector: What?

    Pundits: Nothing. [hands the collector his money] There's your nine pence.

    Itanium: I'm not dead!

    Scrap Collector: 'Ere, he says he's not dead.

    Pundits: Yes he is.

    Itanium: I'm not.

    Scrap Collector: He isn't.

    Pundits: Well, he will be soon, he's very ill.

    Itanium: I'm getting better.

    Pundits: No you're not, you'll be stone dead in a moment.

    Scrap Collector: Well, I can't take him like that. It's against regulations.

    Itanium: I don't want to go on the cart.

    Pundits: Oh, don't be such a baby.

    Scrap Collector: I can't take him.

    Itanium: I feel fine.

    Pundits: Oh, do me a favor.

    Scrap Collector: I can't.

    Pundits: Well, can you hang around for a couple of minutes? He won't be long.

    1. Anonymous Coward
      Anonymous Coward

      Re: With Due Respect to Eric Idle...

      Itanium: Tis but a scratch!

      Everyone Else: A scratch? Your arm's off!

    2. Allison Park

      Re: With Due Respect to Eric Idle...

      IBM shot the first two bullets. They dropped Itanium from their Intel line then dropped it from their software support. The only thing that kept it alive was HP-UX and HP has said that will not get ported to Xeon.

      Oracle doubled the price on Poulson so only an idiot would "upgrade" to Poulson with the same performance per core and 2x the Oracle licenses. With Xeon getting Fab before desktops expect Itanium to be fabbed for one day and Intel will sell the chips until 2025.

  2. Anonymous Coward
    Anonymous Coward

    I don't think that Oracle decided to drop devloping SW for HP-UX and Itanium on the basis of low sales, it was the most popular non x86 platform for them. They dropped it because they'd just bought Sun and wanted to kill off the competition for their new found HW division and HP-UX and the Itanium were basically a platform on which Oracle stood. The low sales bit came after Oracle had said they were going to stick the knife in.

    If you wanted to run an environment that was too big for x86 (at the time) and didn't work well on a RAC then HP-UX on the SD was the way to go.

    One minute Oracle and HP were best buddies and the next moment they weren't.

    1. Anonymous Coward
      Anonymous Coward

      IBM Power had to be larger than Itanium and HP-UX for Oracle at the end. They would probably like to drop IBM Power as well though, probably fear some DB2 defects.

  3. aussie-alan

    Opteron killed Itanium

    I was an architect for moving Sabre's pricing and shopping applications from an 8-way mainframe cluster to open systems and we started with HP-UX and Itanium (in 2003). Two interesting things happened. First, when the new Itanium chips came out, Linux was ready and HP-UX was not, so we went to Linux. Not much later, a couple of people were arguing for x86 and a colleague and I proposed trying the new Opteron servers from HP. They loaned us a couple of beta-test machines and we put them into the production cluster and tried them with live traffic - we saw twice the perfomance at half the price.

    Intel had to match AMD and extend x86 to 64-bit, which ate the market for Itanium. So, indirectly, AMD's Opteron helped kill Itanium

    Alan

    1. Bronek Kozicki

      Re: Opteron killed Itanium

      Yes, it is actually rather ironic that AMD design had played crucial role to open the floodgates of Xeon chips into the data centres. Shame they rarely get recognition for this.

    2. HPCJohn

      Re: Opteron killed Itanium

      Alan, you have it.

      We should not forget that AMD came out with the amd64 architecture first - the hint is in the name!

      I remember Linux talks by the developer for amd64 architecture Linux in the UK (arggh! Can't remember his name).

      I put in the first Opteron 64 bit HPC cluster in the UK at Manchester Uni (at least I think it was).

      But also don't forget that Itanium was a very good processor for scientific applications.

      And indeed SGI is mentioned - their Altix shared memory machines of course used Itanium, and I managed several terabyte sized Altixen later on.

      I remember if a blade failed the SGI Field Engineer (hello Andy !) had to phone the USA to get a security code to join up a new blade. Can't have them damn communists buying blades on EBay and building their own gosh dammit supercomputer. No Siree.

      1. leon clarke

        Re: Opteron killed Itanium

        I seem to remember that the original Iantium plan was that it'd eventually totally replace the 32bit-only x86, even on desktop once desktop needed 64 bit. But AMD came along and screwed up that plan. After that, Iantium never really made sense as the desktop customers weren't funding all the development as they were supposed to.

    3. Erik4872

      Re: Opteron killed Itanium

      Very good observation...I had a very similar experience working on migrating another critical airline system from a Unisys mainframe onto Itanium around the 2003-2004 timeframe. Itanium on Linux was the only game in town if you needed 64-bit support (for high memory use applications) and didn't want to lock yourself in to IBM or Sun.

      It's really interesting to see this whole thing slowly unfold over a decade or two. I'm assuming HPE is going to either port HP-UX and the Superdome to Xeon hardware or kill it entirely. However, it's not clear they can do this in one shot -- there are three OS platforms running on Itanium now (HP-UX, OpenVMS and NonStop.) Systems running on these OSes are pretty long in the tooth and usually critical to a company or government agency's business...

      It would be cool to see these legacy OSes survive on x86-64, but HPE isn't exactly thrilled about supporting them, so I assume they're going to slowly wither and die.

      1. Chris King

        Re: Opteron killed Itanium

        VMS is already on it's way to x86-64, but not as a HP product - https://vmssoftware.com/

        Non-Stop supposedly has Xeon-based hardware available, but I haven't seen any roadmaps for a couple of years.

        HP-UX 11i v3 will be supported until at least 2025, earlier 11i's are now "PVS without SE", so HP will "support" you (but you don't get any new patches/updates) until 2019 - no idea what happened to HP-UX on x86, but I suspect it was taken round the barn and shot a while ago.

        1. STZ

          Re: Opteron killed Itanium

          "Non-Stop supposedly has Xeon-based hardware available, but I haven't seen any roadmaps for a couple of years."

          HPE's NonStop Division has successfully migrated the NonStop OS to x86. They call this new server line NonStop X and are delivering those systems since quite some time now, meanwhile having moved into the second generation of NonStop X. Roadmaps are available but seldom shown, as the typical HPE person does not know much about those systems ...

      2. Anonymous Coward
        Anonymous Coward

        Re: Opteron killed Itanium

        Itanium on Linux was the only game in town if you needed 64-bit support (for high memory use applications) and didn't want to lock yourself in to IBM or Sun.

        Not really. Alpha 21064 was available from 1994 onwards, and had a flat 64-bit logical address space with 43 bits of the virtual address space allowed to be non-zero. It already supported 34-bit physical address space, so it was possible to go beyond 4Gb if you could afford it (very few people could, of course). 21164, which followed in 1995 had a 40-bit physical address support; 21264 increased that to 44 bits; 21364 increased in again to 48 bits. This was in 2003, and things really went downhill rapidly for Alpha after that point, for reasons which had nothing to do with its technical merits :-(

        You were also able to actually get some rather large memory configurations in the Compaq AlphaServer line in that time frame. Unlike Itanium, Alphas also tended to get closer to the peak flops rating on the real-life code: Itanium really suffered when the compiler wasn't able to guess the operation latencies correctly and keep the functional units busy, perhaps because some of the data had to come from the main memory.

        So no, Itanium wasn't the only game in town by a very long stretch.

        1. Anonymous Coward
          Anonymous Coward

          Re: Opteron killed Itanium

          > Not really. Alpha 21064 was available from 1994

          But Alpha was already dead as a product by then. Compaq weren't interested in paying for chip development and were busy porting VMS onto Itanium (interestingly on HP N class systems, which were only sold to the public with PA chips, but had Merced buses internally, and this was way before the HP Compaq merger).

          The old DEC semiconductor business was sold to Intel (part of a court case) and a chuck of the Alpha design team didn't like it. It's those Alpha engineers that were responsible for the AMD64, so they got their revenge in the end.

    4. Anonymous Coward
      Anonymous Coward

      Re: Opteron killed Itanium

      > So, indirectly, AMD's Opteron helped kill Itanium

      Oh believe me there was nothing indirect about it.

      Intel cocked up the first generation of Itanium what shipped was too little and too late. AMD then shipped their 64bit x86 chips and the Itanium workstation market died over night. HP had traditionally kept old models on the price list for ages to cope with the long purchasing cycles for government customers. With the Itainium workstations they pulled the plug on the Friday evening and by Monday morning you couldn't get one.

      From there on Intel's dream of retiring x86 and moving the world to IA64 was dead and HP, who'd started the ball rolling was the only major player left.

      In the end it proved easier to speed up other HW and harder to produce Itanium compilers than they'd expected, but those are minor issues. The AMD64 killed off Itanium on the day it was released by being backward compatible.

  4. Anonymous Coward
    Anonymous Coward

    missing #4

    However, multi-processor, multi-core Xeon developments, the unwillingness of mainframe users to migrate to POWER/SPARC/Itanium server alternatives, and the rise of public cloud computing are three things that have weakened the POWER/SPARC/Itanium server market.

    You forgot the winning combination of the pretty bad (and highly variable) absolute performance and premium price.

    1. Bronek Kozicki

      Re: missing #4

      And #5 availability of compilers

  5. Colin Bull 1
    Devil

    Put that right for you

    Our understanding is that Itanium was developed by Intel to compete with IBM's POWER and Oracle/Sun's SPARC processors AND DEC/Compaq Alpha.

    Itanic could not compete with Alpha so Intel bunged HP a few sweeteners to take over Compaq and drop Alpha. The itanic has been sinking ever since

    1. kain preacher

      Re: Put that right for you

      Um the Itanium start off as an HP thing and they asked intel for help.

      1. Anonymous Coward
        Anonymous Coward

        Re: Put that right for you

        > Um the Itanium start off as an HP thing and they asked intel for help.

        It was kind of a mutual ask. Intel thought that Risc was going to kill x86, it was doing from a perf point of view and doing it with a much lower transistor count. HP didn't have the fab and were interested in off loading the costs. Intel were also chasing other business with HP at the time, so it seemed like it was going to be to everyone's advantage.

        Oh also don't forget that Intel didn't have the whole x86 business to themselves, they'd foolishly signed contracts which meant that other people (such as AMD) could make x86 chips. Intel wanted to move away from x86 to something where they didn't have such immediate competition.

        1. kain preacher

          Re: Put that right for you

          "they'd foolishly signed contracts which meant that other people (such as AMD) could make x86 chips"

          They did not have a choice. When they signed the deal to make the CPU IBM's new computer IBM demanded second source in case Intel could not meet demand.

          1. Roo
            Windows

            Re: Put that right for you

            "They did not have a choice."

            Indeed, second sourcing requirements were not unusual back in the day - particularly for "Defense" projects.

  6. HPCJohn

    Alan, you have it.

    We should not forget that AMD came out with the amd64 architecture first - the hint is in the name!

    I remember Linux talks by the developer for amd64 architecture Linux in the UK (arggh! Can't remember his name).

    I put in the first Opteron 64 bit HPC cluster in the UK at Manchester Uni (at least I think it was).

    But also don't forget that Itanium was a very good processor for scientific applications.

    And indeed SGI is mentioned - their Altix shared memory machines of course used Itanium, and I managed several terabyte sized Altixen later on.

    I remember if a blade failed the SGI Field Engineer (hello Andy !) had to phone the USA to get a security code to join up a new blade. Can't have them damn communists buying blades on EBay and building their own gosh dammit supercomputer. No Siree.

  7. Steve Todd

    The fact that Itanium relied on a smart compiler

    to statically do instruction scheduling, that x86 figured out how to do dynamically in hardware didn't help. Newer generations of processor thus required a recompile to work at full speed on later CPUs.

    It was always too little, too late, and not enough better than the much cheaper x86/x64 ranges to draw much attention.

    1. Roo
      Windows

      Re: The fact that Itanium relied on a smart compiler

      There are other drawbacks to EPIC in addition to recompilation (other architectures suffered from this at various points in their life too). Here's a couple.

      * Dynamic scheduling in hardware has more information available to it - and it can of course it can be tuned to fit the hardware to a much larger degree (instruction grouping, uOps etc).

      * EPIC pushes you towards massive register files and huge buses in the core, which means longer slower wires, more transistors switching state on a cycle, which all means much slower clock rates in comparison to RISC designs on any given process at a given level of power consumption.

      Digital Equipment Corp (amongst others) explored VLIW long before EPIC when silicon was even more expensive and decided to go the RISC route with Alpha. The evidence shows that fabrication was not given enough respect by the guys who came up with the EPIC ISA (that was clear very early on in EPIC's life).

      It's an interesting beast though and it wasn't a total disaster, so fair play to the folks who made it work.

  8. Korev Silver badge
    Pint

    Good discussion

    I've learnt a lot from this, have a virtual pint

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like