Slashdot Mirror

← Back to Stories (view on slashdot.org)

Fujitsu Announces 16-core SPARC64 IXfx (and the Supercomputer It Powers)

Posted by timothy on from the soul-of-a-new-machine dept.

First time accepted submitter A12m0v writes with a link to Fujitsu's announcement of its next generation of supercomputer, from which he pastes: "PRIMEHPC FX10 runs on the newly-developed SPARC64 IXfx processors, which offer a very significant boost in performance over the SPARC64 VIIIfx processor on which they are based and which power the K computer. Each processor has 16 cores and achieves world-class standalone performance levels of 236.5 gigaflops and performance per watt of over 2 gigaflops." Not that K is any slouch.

45 of 68 comments (clear)

  1. Oracle? by Anonymous Coward · · Score: 2, Interesting

    Sun was using Fujitsu SPARC64 processors on some of their servers before the buyout. Are there any news on whether Oracle is going to develop new servers with SPARC64 IXfx processors?

    1. Re:Oracle? by Anonymous Coward · · Score: 2, Interesting

      Rumor from engineer friends is no. They will resell some Fujitsu servers co-branded with Oracle and that's it.
      The longer term roadmap is even more grim for SPARC. I still have a SunBlade on my desk that works like brand new, but sadly is about as useful as the SGI under the desk :/

    2. Re:Oracle? by TheRaven64 · · Score: 5, Interesting

      I don't think Oracle is interested in this market. They wanted the subset of Sun hardware that is good for databases and web apps (i.e. the Tx line), but they aren't really interested in being in the general-purpose server market. They want to be able to control the entire stack from the hardware to the applications, and everything in the middle. There are two reasons for this. The obvious one is that it lets them really tune for performance out of the box. The second, and more important, is that it lets them offer support contracts for the entire machine. If anything goes wrong with it, hardware, operating system, database, or in the business apps, you won't get your in-house IT staff to fix it, you'll just call Oracle. These contracts can be really expensive, and still seem like a good deal compared to keeping a few admins on staff.

      --
      I am TheRaven on Soylent News
    3. Re:Oracle? by Anonymous Coward · · Score: 2, Informative

      Not true. Oracle will continue to jointly develop the SPARC64 with Fujitsu. The servers produced today, dubbed the M-series are in fact sold and produced under Oracle logistics -- not Fujitsu. Fujitsu is a reseller of this technology, so if you buy a server from them, you're having it delivered from Oracle. Its Fujitsu who is the reseller. Further updates to the M-series will include moving toward the LDoms technology where there is an expected convergence with the M-series and T-series sometime 2015 according to the road map. I fail to see how this is a "grim" roadmap.

      The Fujitsu computer that is the FX10 has been developed outside the Oracle/Fujitsu framework and this is one of the reasons that it doesn't run Solaris. The Oracle agreement does not cover the FX10 framework. The VIIIfx CPU is being fitted in Oracle chassis and boards today and expected to be available shortly. Though there are some difficulties with this chip with respect to power consumption and cooling, but nothing insurmountable. From the little that I know, the IXfx CPU is an extension of the VIIIfx and the jury is still out on whether this chip will indeed be compatible with the VIIIfx chassis. Suffice to say that its generic name in contrast to previous SPARC64 chips, it should be possible to integrate it. However looking at the design specs of the FX10 computer it seems as though a lot of the platform is near die or on die. Thus it could be a special chip which won't be suitable for general purpose computers.

    4. Re:Oracle? by sleigher · · Score: 5, Informative

      Right, except have you ever called Oracle support? Had a sev 1 recently. We have 24x7x4 hr call back. Did I get a call back after 4 hours? no. 6 hours? no. 8 hours? no. It was 3 days later we actually got a manager to have a meeting with us to figure out why we can't get support for our hardware. Go ahead, get rid of your staff. Depend on Oracle to help you. That's a great idea. I assure you, you will not regret it.

      --
      All points of time and space are connected.
    5. Re:Oracle? by TheRaven64 · · Score: 1

      Nope, never called them, and never heard anything good from people who have - I'm not recommending buying Oracle, just outlining their thinking with regard to the Sun acquisition.

      --
      I am TheRaven on Soylent News
    6. Re:Oracle? by Anonymous Coward · · Score: 1

      That is not correct -- the M series is entirely a Fujitsu design. I know because I was at one time privvy to the roadmaps.

    7. Re:Oracle? by rubycodez · · Score: 1

      liar, Oracle had nothing to do with the Sparc 64 VII, VII+, VIIIfxe "venus" development. Those are purely Fujitsu's design, Oracle only sell servers such as M using it, they have neither the resources nor brains to design such a thing.

    8. Re:Oracle? by Doctor+Memory · · Score: 1

      Yeah, it's a shame. A buddy of mine just got rid of his Ultra3 workstation because it was too slow to be usable (doing software development running NB7.1, GlassFish and PostgreSQL). My old Ultra20 is still usable, but it's not really "snappy" (and not a SPARC system, despite the name).

      --
      Just junk food for thought...
    9. Re:Oracle? by the+linux+geek · · Score: 1

      The VIIIfx supposedly doesn't support SMP, so I'm skeptical as to how interested in it Oracle is.

    10. Re:Oracle? by assantisz · · Score: 1

      Not my experience at all. We never had any trouble with Oracle support. Sure, the online portal sucks donkey balls but getting them to reply to us within the contracted time has never been a problem. If it happened to you, you should make a stink with your account representative.

  2. Re:sigh... by Anonymous Coward · · Score: 1

    Current Solaris license fees would bankrupt you if it did!

  3. Relevent specifications by EdZ · · Score: 1

    I'm disappointed at this development. 'Icsfics' is a lot less fun to say than 'veeeeefix'.

  4. Re:sigh... by Anonymous Coward · · Score: 1

    It cannot and will not ever run Solaris. The simple reason is that Fujitsu has developed this platform outside the Oracle agreement and thus they cannot run Solaris on it.

  5. I wish someone would put sparc on the desktop by Viol8 · · Score: 1

    But for a reasonable price , not the silly money that Sun wanted for their desktop systems. I'd love to have an up to date solaris box to develop on.

    1. Re:I wish someone would put sparc on the desktop by inflex · · Score: 1

      Never thought that their "lower end" gear was too expensive, like their Opteron workstations, but that said, I'm probably biased because I was with the Sun Developer program for my software - so the hardware was likely strongly discounted.

    2. Re:I wish someone would put sparc on the desktop by fuzzyfuzzyfungus · · Score: 1

      Out of curiosity, are there any remaining applications for solaris where hardware considerations(presumably local access to the video card; but there might be others) would drive customer purchases of a desktop sparc, rather than whatever the Dell rep is trying to get rid of and an SSH or Sun Ray Software and a sparc server?

    3. Re:I wish someone would put sparc on the desktop by TheRaven64 · · Score: 3, Insightful

      I have one of Sun's cheap workstations - the Blade 100, which was 100% cheap commodity crap plus an UltraSPARC CPU. The problem is that they don't have anything like the economies of scale required to make cheap chips. If Intel is selling 100 chips for every one that Sun is selling (which is quite optimistic for Sun / Oracle), then the unit cost of the SPARC is going to be a lot bigger, even if the two chips are the same size and made on the same process, just because all of the one-off costs (including R&D) are spread over a much smaller number of chips. A cheap Sun workstation is still likely to be a few hundred dollars more than an equivalent x86 system, even if all of the components other than the CPU are the same, and that limits them to people who really need to be able to develop on SPARC...

      --
      I am TheRaven on Soylent News
    4. Re:I wish someone would put sparc on the desktop by cpghost · · Score: 1
      If you're happy with UltraSPARC IIIi's, try some Sun Blade 1500 (1 ultrasparc) or SunBlade 2500 (2 ultrasparcs) desktop deals on eBay. I've got a couple of those very fine machines for less than $100 each, and I'm running Debian Squeeze, FreeBSD, and Solaris on them just fine.

      But you're right. I wish we could buy SPARC IV- or later based desktops anytime soon. At this point in time, only FUJITSU would probably consider producing some, if we manage to convince them by showing enough interest. Oracle is way out of touch with the developers' community, they're a lost cause.

      --
      cpghost at Cordula's Web.
    5. Re:I wish someone would put sparc on the desktop by cpghost · · Score: 1

      What's wrong with OpenSolaris on an x86 system?

      1. OpenSolaris, and 2. x86, maybe?

      But seriously now: try to beat a 16 cores SPARC T3 if you have anything heavily multithreaded and/or compute intensive. Even the most recent x86 CPUs are still lagging behind... and GPU acceleration prevalent in x86-world, though fine, takes you only so far.

      --
      cpghost at Cordula's Web.
    6. Re:I wish someone would put sparc on the desktop by Viol8 · · Score: 1

      Its no longer supported and wasn't as good as sparc solaris anyway. Plus I'd like something other than x86 for a change.

    7. Re:I wish someone would put sparc on the desktop by Viol8 · · Score: 1

      "A cheap Sun workstation is still likely to be a few hundred dollars more than an equivalent x86 system"

      A couple of hundred I could handle. What annoyed me was back in the day a desktop sparc would be at least 50% more than the equvalent x86 and when you'd added in the optional (!) keyboard, monitor etc it could easily be twice as much. For high end kit you'd be looking at 3 or 4x the price.

      As for ebay , there are some bargains on there but I need up to date hardware and OS really, to match what I use at work.

    8. Re:I wish someone would put sparc on the desktop by rubycodez · · Score: 1

      from the opensolaris web site: OpenSolaris 2009.06 was the last OpenSolaris binary release published by Sun Microsystems and is no longer available for download.

      unlike BSD, opensolaris is dead. abandoned graveyard dead.

    9. Re:I wish someone would put sparc on the desktop by Doctor+Memory · · Score: 1

      Maybe not. I bought one of their Ultra20 workstations new for around US$800 and that was without any discounts. More expensive than building an equivalent box from parts, but the parts are top notch (Tyan motherboard, nVidia NV280 graphics card). My only complaint was having to track down a funky "SPUD" bracket to mount a second hard drive.

      --
      Just junk food for thought...
    10. Re:I wish someone would put sparc on the desktop by the+linux+geek · · Score: 1

      The T3 is crap for anything compute-intensive. The whole point was "throughput computing," not processing power, and even at that it's pretty poor. Look at the speccpu benchmarks sometime.

    11. Re:I wish someone would put sparc on the desktop by Darinbob · · Score: 1

      "Back in the day" there weren't many x86 workstations that could do what a Sparc or other Unix workstation could do. Sun even sold some x86 machines which didn't do too well. Of course the workstations sold for much more than a lowly PC. There weren't x86 workstations at the time, that cpu was intended for very low end commodity home and business machines. It took a lot of Intel resources to beef things up, and I doubt they would have done it without competition from x86 clones.

    12. Re:I wish someone would put sparc on the desktop by unixisc · · Score: 1

      Sun used to offer SunBlade workstations, and some priced @ $7000. I agree that was steep, but standard for its time. Also there was a company called Integrix that made SparcStations, as well as another called Tatung. They're either gone, or changed.

      As far as the RISC based computers go, 2 of the top CPUs - Alpha & PA-RISC are dead, Sparc is more alive due to FJ than Oracle, MIPS is alive but nobody uses them in computers anymore - only in routers, and it seems to have lost its games console to POWER variants (I think POWER now owns the games market). Itanium is overpriced garbage, and so one is stuck w/ the x64s from Intel & AMD. I'm not counting ARM here. Let's see how OpenRISC evolves.

      I myself prefer RISC computers. I do wish someone makes an OpenSparc CPU based on UltraSparc (whatever its number is by now) and then an unixstation based on it running either BSD or Linux, and see OpenIndiana ported there. It would be a good machine to devlop on. As for the other question above about OpenSolaris on x86, that's so passe.

    13. Re:I wish someone would put sparc on the desktop by unixisc · · Score: 1

      Yeah, but it was forked just in time before Oracle pulled the plug. You might want to check out OpenIndiana

  6. Pricing would be interesting! by gentryx · · Score: 2

    Fujitsu is fishing in the same waters as IBM does with their BlueGene machines: both lines are designed to deliver 20 PFLOPS and both are traditional systems in the sense that you don't have accelerators like GPUs, which are still awkward to program for the average physicist. Thus, to potential buyers the TCO would be interesting. From what I've heard BlueGene/Q is twice as power efficient as the Sparc VIIIfx design, but those were just 8-cores, not 16-cores.

    So, assuming comparable total power consumptions and a affordable price tag, Fujitsu could snatch several deals from big blue, perhaps even the recently failed Blue Waters, although my money is on Cray for that machine.

    --
    Computer simulation made easy -- LibGeoDecomp
    1. Re:Pricing would be interesting! by Rockoon · · Score: 1

      Indeed, 236 GFLOPS on a single CPU is quite impressive. Thats over twice the performance of an OC 2600K (4.6ghz.)

      The 8-core VIIIFX chip pushed 128 GFLOPS with 760 million transistors on 45nm .. this IXFX is I guess marketed as a drop-in replacement chip so it must be using a smaller process size to fit in the same socket.

      --
      "His name was James Damore."
    2. Re:Pricing would be interesting! by Macman408 · · Score: 1

      A high-end GPU will do 5 to 20 times as many FLOPS, at 3-6x the efficiency. Quite possibly at a lower cost, too, given that they're relatively high-volume parts, compared to a SPARC. And before anybody complains about "well yeah, but GPUs can't do X", well, that's what they get for using FLOPS for a benchmark. Unless what they're interested in is FLOPS, they really shouldn't use them as a benchmark!

    3. Re:Pricing would be interesting! by joib · · Score: 1

      Oh? So how come the VIIIFX based "K computer" then, apart from being the current #1 in performance, also beats the GPGPU clusters (with the latest Nvidia Fermi cards) in flops/watt on the latest top500 list: http://top500.org/list/2011/06/100 ? And heck, that's on linpack, which should be the pretty much optimal workload for a GPU.

    4. Re:Pricing would be interesting! by Macman408 · · Score: 1

      What you're looking for is the Green500 list, where the K computer is #6; behind an ATI GPU+Intel i5 cluster and two NVIDIA Fermi+Intel Xeon clusters (two IBM Blue Gene Q prototypes sit at the top). The first three are fairly small (100+ on the top500), but the NVIDIA systems sit at #5 and #54 on the top500 as well, so it doesn't appear to be a scalability issue. I have no knowledge of the design tradeoffs of the individual systems, but I'd say that it's fairly impressive that both the top500 and the Green500 have so many GPUs in the top 10, given that they're both CPU-dominated lists.

    5. Re:Pricing would be interesting! by gentryx · · Score: 1

      I'm doing my PhD in HPC. From my perspective GPUs do indeed offer a lot of GFLOPS but it's often impossible to max them out. Especially for stencil codes (read: virtually all physical simulation codes) this is hardly possible because of the low operational intensity of stencils. CPUs achieve much higher efficiencies here because they can do cache blocking. The caches on e.g. Fermi are much too small to do that well. So no: in this case a GPU won't necessarily yield you a higher performance or efficiency

      As for the pricing: in scientific computing we typically care for double precision. For that you have to by Nvidia Tesla cards, which cost several thousand dollars, each. I don't know the prices of the Sparc chips, but I doubt they'll be higher.

      --
      Computer simulation made easy -- LibGeoDecomp
    6. Re:Pricing would be interesting! by joib · · Score: 1

      What you're looking for is the Green500 list [green500.org]

      Indeed, but the site was down when I wrote my previous reply so I had to resort to the top500 list and calculating flops/watt for the few top entries manually. :)

      In any case, as one can see from the list, the best GPU machine manages to beat the K machines by a factor of 1.66, a far cry from the factor of 3-6 you originally claimed. And most GPU machines fall behind the K.

      I think the sparc viiifx is quite impressive, it gets very good flops/watt without being a particularly exotic design. Basically it's just a standard OoO CPU with a couple extra FP units and lots of registers clocking at a little lower frequency than usual. No long vectors with scatter/gather memory ops, no GPU's, no low power very slow embedded CPU's like the Blue Genes etc.

      I have no knowledge of the design tradeoffs of the individual systems, but I'd say that it's fairly impressive that both the top500 and the Green500 have so many GPUs in the top 10, given that they're both CPU-dominated lists.

      Large GPGPU clusters are still a relatively new phenomenon, give it a few years and I suspect you'll see a lot more of them.

    7. Re:Pricing would be interesting! by Macman408 · · Score: 1

      In any case, as one can see from the list, the best GPU machine manages to beat the K machines by a factor of 1.66, a far cry from the factor of 3-6 you originally claimed. And most GPU machines fall behind the K.

      I was going by the per-chip benchmarks of both the SPARC and of GPUs, since that was what was at hand.

      I think most of these GPU clusters have very high-powered CPUs (eg Xeons) in them. I wonder what the perf per watt would be if you replaced the Xeon with, say, a low-power embedded chip, like an ARM (or an Atom, if you must). Granted, that further limits the types of computations that your supercomputer can do, but it might get you a lot closer to that 3-6x factor for the chip alone.

      Large GPGPU clusters are still a relatively new phenomenon, give it a few years and I suspect you'll see a lot more of them.

      No doubt. When I was doing my master's research, they were just starting to look interesting. I was working in reconfigurable hardware, but had one or two peers playing around with GPUs. That's one of those things that people keep looking at, and occasionally use for something, but has never really taken off like GPGPU has.

  7. Re:Breach of Contract by sleigher · · Score: 2

    I would love to. My point is I warned the company to stop buying Sun hardware when the acquisition went through. I knew it was gonna be like this. They are giving us dedicated back line support people we can call directly. I am sure that is sustainable by Oracle as well...

    --
    All points of time and space are connected.
  8. do they sell them at bestbuy? by Cyko_01 · · Score: 2

    facebook is so slow these days!

    1. Re:do they sell them at bestbuy? by Neil+Boekend · · Score: 1

      Yeah, and it's also small with all those toolbars. I need a new PC to have a larger browsing segment on IE again.

      --
      Well, I might have a way, but it only works on a semi spherical planet in a vacuum.
  9. Re:sigh... by sunderland56 · · Score: 1

    It cannot and will not ever run Solaris.

    You sound like a lawyer. I think you're on the wrong web site; this is slashdot - we're geeks here.

    Sure, they probably won't be able to sell it commercially with Solaris - but I bet that Solaris runs just fine. Might need a device driver or two, but there's a fully supported DDK to make writing them easy. I think that Fujitsu engineers would be slacking if they *didn't* have a prototype unit running Solaris in their lab.

    // posted from my x86 laptop running solaris

  10. Re:please, Fujitsu, please by Brane2 · · Score: 1

    I second that.

    I'm from penguin crowd, but nevertheless it would be really nice to work on some decent, many-cores non-x86-crap design...

    Ideally, on dual socket board...

  11. 'tofu' interconnect, a 6D Mesh/Toroid? by adinb · · Score: 1

    maybe someone with a more recent HPC CS degree can break down this interconnect/routing architecture to me? I loved playing on 3D toroidal meshes, especially on SIMD MasPars (communication penalty was 1 instruction cycle & we could select which processors would execute an instruction...from what I remember).

    Are there any advantages to this topology for certain classes of problems (the MasPars were awesome for matrix math & image processing) ? Or is the sole advantage in routing speed/interconnect traversal?

    --
    Moderation is for Monks!
  12. Re:please, Fujitsu, please by galanom · · Score: 1

    Dual socket, inexpensive, and in ITX motherboard?

  13. Re:sigh... by unixisc · · Score: 1

    Do Solaris licenses per CPU mean per CPUs, or per the actual number of computers that would be running it? B'cos 8000+ CPUs in one computer counts as only... one computer. Or do they charge per #seats?

    Also, given how OpenIndianna, Illumos and Nexenta are forks of Solaris, how do they not support Sparc? What other market is there - do they seriously expect people to use them on their Xeon or Opteron servers?

  14. Re:please, Fujitsu, please by Brane2 · · Score: 1

    Affordable. Something like Tyan's dual socket Opteron boards. â300-400.