back to article Blueprints revealed: Oracle crams Sparc M7 and InfiniBand into cheaper 'Sonoma' chips

Oracle revealed on Monday the details of its new bargain-basement Sparc processor code-named Sonoma. The blueprints were shown at this year's Hot Chips semiconductor conference in Cupertino, California – and we've got a copy of the slides. Sonoma is billed as a "low-cost Sparc processor for enterprise workloads," that you're …

  1. Anonymous Coward
    Anonymous Coward

    I always wonder how companies can make money off such low volume chips. But it is true that the market supports thousands of different microcontroller chips at the other end of the scale. I don't understand the market economics of it. Maybe the manufacturing cost is like $1 or $2 but they are sticking $1000 price tags on the things.

    1. P. Lee Silver badge

      >Maybe the manufacturing cost is like $1 or $2 but they are sticking $1000 price tags on the things.

      Do they still even sell chips or even sparc hosts?

      It's been ages since I used a sparc but they were never about high-volume throughput, they were about being rock solid. That attribute has been devalued as systems scaled up and everyone has much wider redundancy and failover solutions, than "it hasn't gone down in living memory - if it does, we'll restart it, but its rather unlikely."

    2. Phil 4

      Who said anything about low volume? Oracle has over 400,000 customers running its SW and over 310,000 are running Oracle DB. And millions are running Oracle MySQL and other Oracle open source SW. So even if Sonoma only addressed Oracle SW workloads, that’s a pretty large volume. And with Oracle's rapidly growing Cloud IaaS business, there seems to be some incredible opportunities for a hyperconverged/integrated processor that will be aggressively priced.

      1. Roo
        Windows

        Oracle already ship SPARCs and they already ship SPARC software. How many of those 400K customers actually run SPARC binaries ?

        Taking my last 5 "bluechip" employers as an example precisely 5/5 of them ran Oracle, 0/5 of them ran Oracle on SPARC (2 of them migrated away from SPARC before I got there). Admittedly that's a very small sample - but I am be curious to know if you have seen a different picture.

        1. Phil 4

          From what Ive heard, roughly 20% of Oracle customers are running on SPARC so theres a huge 80% opportunity for Sonoma (and SPARC in general) to go after that business beside the larger private/hybrid/public Cloud IaaS. With Itanium all but dead and IBM Power going after HPC, seems the only alternative to x86 is SPARC for the enterprise. And with security becoming a more critical factor than even cost or performance, Sonoma and its big brother SPARC M7's Application Data Integrity technology, they may just have the technology to break the strong hold.

          1. Anonymous Coward
            Anonymous Coward

            If you were choosing a CPU platform, would you choose:

            a) ARM. Low power, low cost and getting stronger

            b) x86. A little more costly than ARM, but gives you a broad range of power/performance models to match your requirements

            c) I'm stuck with a requirement for X that only runs on platform Y. I can't easily migrate. Currently pleading with vendor to provide a new platform to support our requirements beyond 5 years....

            d) SPARC. It has ummmmmm

            The years of x86 vs SPARC/Itanium/POWER are over and the x86 vs ARM battle is looming

            1. PlinkerTind

              x86 small workloads

              "...If you were choosing a CPU platform, would you choose:

              d) SPARC. It has ummmmmm

              The years of x86 vs SPARC/Itanium/POWER are over..."

              It is true that x86 works fine today. If you have smallish workloads. An Intel Xeon E7v3 can cope with some nice workloads. Small that is. If you need to tackle the largest workloads, there are no other choice than using Unix such as POWER or SPARC.

              The largest x86 business server for Enterprise has until recently been 8-sockets. Just recently there was a 16-socket x86 server released. 16-socket x86 servers are immature and scale badly. And guess what? 64-socket SPARC server beats 8/16-socket x86 servers.

              For instance, the best SAP score is 320.000 saps for x86 with the brand new Xeon E7v3. The top record? It's SPARC with 844.000 saps, it's SPARC all the way. There is no way an x86 can come much higher than 300.000 saps. SGI UV2000 with 10.000s of cores can not do it, as it is a cluster. And scale-out clusters can not run business software. You need a scale-up server with as many as 32-sockets or 64-sockets. Decades ago there was an 144-socket SPARC server.

              So, yes, if you only have smallish workloads, x86 is fine and you dont need Unix. If you have extreme demands, only Unix will do. The biggest POWER8 server, the E880, has only 16-sockets and 16TB RAM, so maybe it will be considered midsize. The largest SPARC servers have 64-sockets and 32TB (and soon 64TB), and are from Fujitsu, the mighty M10-4S. So the only high end business servers with >32-sockets out there, are SPARC. POWER8 maxes out at 16-sockets. Intel Xeon maxes out at 16-sockets as well (but the new 16-socket x86 likely has awfully bad performance as neither Windows nor Linux has ever scaled this high earlier).

              1. Anonymous Coward
                Anonymous Coward

                Re: x86 small workloads

                So, yes, if you only have smallish workloads, x86 is fine and you dont need Unix. If you have extreme demands, only Unix will do.

                Very true. Many of the commentards here may think they have large workloads, but compared to the top guys in the Fortune 500 they're insignificant. Until you really need the top-end IBM POWER or Oracle/Fujitsu SPARC systems you have no idea of just what those boxes can do.

              2. Anonymous Coward
                Anonymous Coward

                Re: x86 small workloads

                >> It is true that x86 works fine today. If you have smallish workloads. etc.

                All this is highly interesting analysis .... in a world where workloads cannot execute (optimally or at all) without a vertically scaled 32 socket system. One might find such a world, somewhere in the multiverse. Or said 32S system might be called the Glooper v2 and installed in the Royal Bank of Ankh-Morpork.

                1. PlinkerTind

                  Re: x86 small workloads

                  "......All this is highly interesting analysis .... in a world where workloads cannot execute (optimally or at all) without a vertically scaled 32 socket system. One might find such a world, somewhere in the multiverse. Or said 32S system might be called the Glooper v2 and installed in the Royal Bank of Ankh-Morpork....."

                  Typically business entréprise workloads such as databases can only be run on a single fat scale-up server such as huge Unix boxes. It is extremely difficult to do heavy transaction processing on a cluster, syncing data integrity across all nodes, rollback, etc. No one has done it. That is the reason you must pay $$$$ for Oracle database on a huge Unix instead of byuing a cheap clustered database on x86 that can beat oracle. Clustered databases don't exist, too difficult to create.

                  Sure, there are Datawarehouses that are clustered but they are only used for reading data and analyzing so they don't have to save data and guarantee data across many nodes. They are not used to store and edit data. There are clustered databases such as SAP Hana but these are typically RAM databases only used for reading and analyzing, ie used as Datawarehouse. You don't store data in ram. SAP Hana also has a database for storing data, transactions, but guess what? It only runs on scale up 16 socket server called UV300H. No cluster. The SGI UV2000 cluster runs a database TimesTen at USPS, but it is only used for analyses, detect fraud. It has a real database for storing data too.

                  In short, there are no clustered databases used to store transactions. They are all used for reading and analyzing data. The only type of servers for databases storing transactions are scale up servers, ideally with 32-sockets or more.

                  1. Anonymous Coward
                    Anonymous Coward

                    Re: x86 small workloads

                    In short, there are no clustered databases used to store transactions. They are all used for reading and analyzing data.

                    Citation please, because that doesn't sound correct at all. ;)

                    1. Anonymous Coward
                      Anonymous Coward

                      Re: x86 small workloads

                      "In short, there are no clustered databases used to store transactions. They are all used for reading and analyzing data."

                      Hmmm, ever heard of RAC?

                      Everywhere else in the world, outside oracle hw marketing dungeons, if you have very huge workload for Oracle DB you think how to set up RAC best, only Oracle would be so rich to be able to pay for those mythical 32+ socket machines and still justify that, economically and feature-wise. I wonder how many such machines they sold - 5? 3? Or maybe just zero.

                      1. PlinkerTind

                        Re: x86 small workloads

                        ".....Hmmm, ever heard of RAC?...."

                        Oracle RAC is mainly for uptime. Everything is mirrored on two identical servers. Sure, you can also distribute different database tables among the mirrored servers - which is what Oracle calls "partitioning" but you can not have a cluster of nodes handling the same database table rows. That would be very inefficient to sync data among nodes with roll back etc

                        1. Down not across Silver badge

                          Re: x86 small workloads

                          Oracle RAC is mainly for uptime. Everything is mirrored on two identical servers. Sure, you can also distribute different database tables among the mirrored servers - which is what Oracle calls "partitioning" but you can not have a cluster of nodes handling the same database table rows. That would be very inefficient to sync data among nodes with roll back etc

                          I hate to say it, but you obviously don't have the slightest clue about Oracle RAC and how it works.

          2. Matt Bryant Silver badge
            Facepalm

            Re: Phil 4

            "....so theres a huge 80% opportunity for Sonoma (and SPARC in general) to go after that business..." Yeah, the Sunshiners have been pushing that line since UltraSPANKed, it's just every year that "opportunity" gets bigger as more and more customers ditch SPARC for x64. ROFL!

        2. Anonymous Coward
          Anonymous Coward

          "Taking my last 5 "bluechip" employers as an example precisely 5/5 of them ran Oracle, 0/5 of them ran Oracle on SPARC (2 of them migrated away from SPARC before I got there). Admittedly that's a very small sample - but I am be curious to know if you have seen a different picture."

          I don't know about large scale enterprise, but Solaris/SPARC servers are becoming increasing uncommon in the carrier market (which HP seams own, at the moment). Back in the day, when I started out in Telecoms, providing the server side of our solutions, on anything but Sun boxes, was pretty much unthinkable. Now it's all HP servers hosting (mostly Linux) VMs. I miss the simplicity/maintainability of the SPARC servers, but modern HP servers aren't that far behind.

      2. Matt Bryant Silver badge
        Facepalm

        Re: Phil 4

        "....Oracle has over 400,000 customers running its SW...." On Intel, and not on Slowaris.

        1. s2bu

          Re: Phil 4

          Last I checked Matt, Solaris runs fine on Intel. I wouldn't call it slow either.

          1. boatsman

            Re: Phil 4 == last time I checked

            solaris on x86 was pretty much very dead, by design, decisions, policy and strategy of Oracle itself.

            1. iOS6 user

              Re: Phil 4 == last time I checked

              > solaris on x86 was pretty much very dead, by design, decisions, policy and strategy of Oracle itself

              You made my day :)

              In few weeks will be released Solaris 11.3. I would be really glad to se any Linux distribution with comparable number of new features (most of them are not available on any Linux).

              On Horizon is Solaris 12.

              1. PlinkerTind

                Re: Phil 4 == last time I checked

                Solaris vs Linux on same hardware, typically shows that Solaris has superior performance.

          2. Phil 4

            Re: Phil 4

            Actually, Solaris is ranked the fastest OS in the world according to many public benchmarks including TPC-C, SPECjEnterprise2010, SAP SD 2-Tier http://www.tpc.org/tpcc/results/tpcc_perf_results.asp https://www.spec.org/jEnterprise2010/results/jEnterprise2010.html http://global.sap.com/solutions/benchmark/sd2tier.epx

      3. Anonymous Coward
        Anonymous Coward

        Re: Phil 4

        And millions are running Oracle MySQL and other Oracle open source SW.

        "Oracle" and "Open Source Software" in the same (serious) sentence is kind of sad. Most of the Oracle "Open Source" users are very much not using Oracle software by choice, instead they were using a non-Oracle product which Oracle bought the primary authors of then proceeded to fuck-up.

        PostgreSQL, MariaDB, and other "Oracle MySQL" alternatives have a lot of new users thanks to Oracle's "excellent handling" of things.

        That being said, there are a few Oracle projects that aren't all bad. BtrFS looks like it could be successful given time.

        1. Phil 4

          Re: Phil 4

          I disagree with your statement. Since Oracle acquired MySQL, its popularity has actually risen. And Oracle DB is going stronger as well-by choice.

          Your statements are based on outdated perceptions. If you look at all Database ranking, http://db-engines.com/en/ranking, Oracle DB and Oracle MySQL are dominating the rankings and MySQL has actually increased in last 6 months. Sure, other alternatives are out there, and some have increased quite dramatically, but nothing comes even close in rankings.

          1. Anonymous Coward
            Anonymous Coward

            Re: Phil 4

            Just to clarify, do you work for Oracle or an Oracle partner or something?

            You seem very involved in promoting Oracle.

        2. future research

          Re: Phil 4

          "That being said, there are a few Oracle projects that aren't all bad. BtrFS looks like it could be successful given time."

          If oracle really cared about linux (and not just copied RedHat distro) They would release the updates they made to ZFS since owning it under the GPL licence.

          1. Down not across Silver badge

            Re: Phil 4

            If oracle really cared about linux (and not just copied RedHat distro) They would release the updates they made to ZFS since owning it under the GPL licence.

            Maybe. As luck would have it FreeBSD has very well working ZFS implementation. Best of both worlds as you don't have to suffer some annoying Linuxisms either.

      4. Anonymous Coward
        Anonymous Coward

        Re: Phil 4

        And with Oracle's rapidly growing Cloud IaaS business ...

        Someone's been drinking the kool-aid... :p

      5. boatsman

        404 --- most of oracle's software

        does NOT run on sparc.

        yes, one can do it. but most shops do not.

        Anyway, these chips are not about volume. it is about many of sockets, hundreds of cores in a single system.

        typically what you do not see in x86 based systems, and the ones that do, do not scale.

      6. Anonymous Coward
        Anonymous Coward

        means nothing

        These numbers mean completely nothing, I'm working as oracle SW consultant (for 3rd party, not for oracle itself), I'm working with tens of different oracle customers around whole Europe per year and the only sparc machines I've seen running oracle since at least 5 years are very old machines, purchased still from Sun. Between thousands of servers I see per year I have seen only 1 sparc supercluster bought from Oracle 2 years ago by one of the customers - and they weren't very happy from this purchase so it's unlikely they will ever buy another one. Currently X86 has at least 90% of this market, remaining 10% (but high end 10%) is Power, others are non existing anymore. I don't know who's buying these sparc machines, at least here in Europe no one. The only place where they can use them is in their own cloud - though I really doubt, Oracle is using practically only x86 and linux too, they are not stupid. I strongly suspect it's paperware.

        1. This post has been deleted by its author

    3. Daniel von Asmuth Bronze badge
      Headmaster

      What market?

      The first thing to notice is that it takes some cool millions to design a processor. Most companies now act like ARM and design processors, but do not produce them. Building a fab will cost many billions. but some fab owners will produce third-party chips.

      You don't get to buy SPARC processors on your local hight street. The conclusion: all that matters is the price and performance (and reliability) of a complete Oracle system.

      1. Anonymous Coward
        Anonymous Coward

        Re: What market?

        I would argue that it costs billions to develop a high performance CPU rather than millions. If you aren't going to sell enough software/hardware/services to justify the investment, when do you stop investing?

        This isn't about high street sales. I suspect that Larry just wants to last one generation longer than Itanium so he can laugh at HP.

    4. Anonymous Coward
      Anonymous Coward

      Easy

      By charging more. They literally need only 1 customer. If that customer is willing to order enough chips and pay enough money, they'll keep doing it. For all of Oracle's faults - losing money isn't one of them.

  2. Ragequit
    Joke

    Just don't...

    backward engineer their crypto instructions looking for vulnerabilities. That's against their EULA. oO

  3. Anonymous Coward
    Coat

    ...but... can it run Crysis?

  4. John Savard Silver badge

    What I Think Is the Big Question

    When they say "low cost", is that low compared to mainstream Intel Core i7 chips - or low compared to the cost of the chips in IBM's latest z13 legacy mainframes - which is what Oracle, as a database company, is directly competing with?

    1. Steve Davies 3 Silver badge

      Re: What I Think Is the Big Question

      well Duh! As long as it costs as little as $1 less than a z13 Mainframe then Larry will be happy.

      The real question should be

      "How many arms and legs is this gonna cost me."

      {if we want to keep complaint with EL-Reg units of meansurment that is}

      1. Roo
        Devil

        Re: What I Think Is the Big Question

        My guess is that an M7 might be about 1/4 of a kidney. x1000 prices won't happen because they won't yield enough in a month to collect that many together in one place. ;)

    2. LDS Silver badge

      Re: What I Think Is the Big Question

      At least, compare it with Intel chips designed for servers, not desktop/laptops...

      1. Roo
        Windows

        Re: What I Think Is the Big Question

        "At least, compare it with Intel chips designed for servers, not desktop/laptops..."

        Kinda tricky to do CPU compare the last time I tried that from the published benchmarks. Case in point Oracle only publish SPECrate figures so I have no idea how well a modern SPARC will run the billions of lines of single threaded code out there.

    3. Down not across Silver badge

      Re: What I Think Is the Big Question

      When they say "low cost", is that low compared to mainstream Intel Core i7 chips -

      i7? You should at the very least compare with a Xeon.

  5. Anonymous Coward
    Anonymous Coward

    O-cannibalism?

    It was always a question for me about Sun Micro, how they can sell x86 and SPARCs in the same time?

    In case of the Big-O, it looks much worst - Unbreakable Linux, Solaris, x86, beef-up SPARCs, exa-clusters on x86 and in the near future - some big Sonoma clusters... Keeping in mind, the Big-O is a software company with sales margins lager than any hardware company even can imagine in their dreams. What is the secret? Why they carrying on a low profit hardware portion of the business? If the answer is they selling appliances, then what OS the appliance is running (Linux or Solaris) and what is the actual hardware will be irrelevant for the user except the price tag and the support contact. Somehow, former Sun hardware portion of the Big-O still coexists with rest of variety of software only groups and ejects every year something new.

  6. -tim

    Cheaper chips?

    Why do I think this won't lead to reasonably priced systems?

    We bought a bunch of X1 and V100 sun boxes about a decade and a half ago when they were at the $1,000 price point. The only reason we are dumping them now is we can't buy disks for them since their PATA controller chip has a bug with disks over 120G. The SPARC IIi that is in those made with modern techniques and a SAS/SATA/PCIe in the X100 box would be great for appliances and so far we are using more power trying to virtualize them than the old stuff too. 15 years ago a $1,000 SPARC box would outrun a $1,000 x86 box for most loads. Today a $20,000 SPARC box holds its own aginst a $6,000 x86 box.

  7. Michael Wojcik Silver badge

    Oh good, this again

    It also supports unsafe algorithms such as MD5, SHA-1, and DES, weirdly.

    Clearly written by someone with a very limited understanding of cryptography, particularly as it's used in the real world.

    MD5 remains suitable for many purposes. And it's still used for many purposes, including some where it's not so suitable, modulo the threat models of the participants. Accelerating it on the chip is very cheap once you're accelerating pretty much anything else, so why not?

    The same goes, but even more so, for SHA-1. SHA-1 is no longer suitable for signing X.509 certificates used to validate the identity of anything of significant interest to attackers willing to devote some resources, true; but it's fine for many other things. And certificates signed using SHA-1 aren't going to suddenly disappear.

    DES, in the form of EDE 3DES, is just fine for most purposes where the value of the data doesn't exceed the cost to break a 112-bit symmetric key.

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

Biting the hand that feeds IT © 1998–2019