AMD unleashes 12-core Opteron processors

The high end of the Opteron product family has just gotten a little bit bigger. AMD has announced the availability of its 6000-series Opteron platform, which combines 8- and 12-core processors with a brand-new G34 socket, DDR3 support, newer I/O technologies, and many other improvements.

At the heart of today’s launch is Magny-Cours, a new processor design made up of two 346-mm² dies totaling 1.808 billion transistors and sharing a single CPU package. AMD targets 6000-series Opterons at two- and four-socket servers, and the firm says Opteron 6000-based systems from vendors like HP, Dell, Acer, Cray, and SGI are on the way. Here are the various Magny-Cours flavors you can expect to see in the wild:

Processor Cores Clock speed L3 cache HT3 speed ACP Price
Opteron 6176 SE 12 2.3 GHz 12 MB 6.4 GT/s 105 W $1,386
Opteron 6174 12 2.2 GHz 12 MB 6.4 GT/s 80 W $1,165
Opteron 6172 12 2.1 GHz 12 MB 6.4 GT/s 80 W $989
Opteron 6168 12 1.9 GHz 12 MB 6.4 GT/s 80 W $744
Opteron 6136 8 2.4 GHz 12 MB 6.4 GT/s 80 W $744
Opteron 6134 8 2.3 GHz 12 MB 6.4 GT/s 80 W $523
Opteron 6128 8 2.0 GHz 12 MB 6.4 GT/s 80 W $266
Opteron 6164 HE 12 1.7 GHz 12 MB 6.4 GT/s 65 W $744
Opteron 6128 HE 8 2.0 GHz 12 MB 6.4 GT/s 65 W $523
Opteron 6124 HE 8 1.8 GHz 12 MB 6.4 GT/s 65 W $455

Each one of these new CPUs has four DDR3 memory channels and can support up to 12 memory modules. In fact, the Opteron 6000 series inaugurates a new class of 2P system that can host up to 96GB of memory and features a total of eight memory channels. AMD believes such systems are especially well suited for some key applications, including virtualization, databases, and HPC installations.

In an interesting twist, all Opteron 6000 CPUs should work in either two- or four-socket configurations—so no more price differences between 2P- and 4P-capable variants of the same CPUs (or, in AMD’s words, the end of the "4P tax"). AMD anticipates that its customers, the larger server vendors, will offer relatively inexpensive 4P Opteron machines based on 6000-series processors, in addition to 2P systems with comparatively large amounts of memory. By pursuing these new frontiers in 2P and 4P capability and pricing, AMD hopes to capture a sizable portion of the server market in a competitive environment made rather difficult by the Intel’s formidable 32-nm Xeons.

AMD has measured an 88% integer performance boost and a 119% floating-point performance increase with the 6000 series compared to previous-gen Istanbul six-core Opterons.  CPU-to-CPU communication speeds have increased by 33% thanks to a faster 6.4 GT/s HyperTransport rate. The new Direct Connect Architecture 2.0 also reduces the maximum number of hops between sockets in a 4P configuration from two to one. Full AMD-V virtualization capabilities come as standard, as well.

Just because Magny-Cours gathers 12 cores on a single package doesn’t mean it’s a power guzzler. As you can see in the table above, 12-core Opteron 6100 processors can be had with ACP ratings as low as 65W. In addition, AMD has implemented deeper C1E sleep states; CoolSpeed, a new feature that lets server admins specify temperature limits within the data center; Advanced Platform Management Link, which enables remote monitoring and control of power states and cooling; and support for low-voltage, 1.35V DDR3 memory modules.

The Opteron 6000 series will make use of AMD’s 5600-series chipset, which delivers I/O virtualization, HyperTransport 3.0, and PCI Express 2.0 connectivity.

AMD intends to keep the G34 platform alive for some time. The announcement specifies that next-generation Opterons based on the Bulldozer architecture will work in the same socket with the same chipset as 6000-series Opterons. (Bulldozer is due next year.)

Before then, in the second quarter, we should see the other half of AMD’s revamped Opteron offering: the Opteron 4000 series. Code-named Lisbon, those chips will essentially have the same silicon as Magny-Cours, just with a single die per package (so either six or eight cores). Expect them in the second quarter of this year as part of a new C32 platform aimed at one- and two-socket servers.

Comments closed
    • Auril4
    • 10 years ago

    Minus all the ethics violations.

    • ub3r
    • 10 years ago

    Looking more like an intel chip…. Nice..

      • Auril4
      • 10 years ago

      Minus the ethics violations that Intel is so famous for.

    • jinjuku
    • 10 years ago

    Any idea on when Mainboards will hit? I have some aging servers that I would love to consolidate.

    I can’t believe I can run a 12 core box with a 500 watt PSU. Mind bending.

    • potatochobit
    • 10 years ago

    what does HE mean

      • Game_boy
      • 10 years ago

      Lower power band than standard – in this case 65W ACP.

      • dpaus
      • 10 years ago

      What HE said…. (as in, l[

      • ssidbroadcast
      • 10 years ago

      High Explosive. (if you’re playing X-Com)

        • jdaven
        • 10 years ago

        I love that game. I actually downloaded the entire collection on steam for like $5. Unfortunately, only the first game was any good.

    • Forge
    • 10 years ago

    I’m starting to wonder if a pair of 8 core Opterons might not be a better value than my Core i7. Do want x264 encoding benchmarks!

    • slaimus
    • 10 years ago

    Does anyone know why server platforms do not have integrated graphics? It seems like it would make sense for them to include it with sideport memory rather than having a super low-end graphics the Rage Pro or Matrox G200e that most servers seem to have. The only reason I can possibly justify is that servers need all the PCI-E lanes they can get, so they put video on legacy PCI.

      • Scrotos
      • 10 years ago

      Do you mean as a dedicated PCI card versus something built into the motherboard?

      Because all the rack-mountable servers I have now have integrated video, no external PCI card needed.

      Or are you talking about the video hardware hanging off the PCI bus versus the PCIe bus? If so, I’m not sure how that distinction turns into “integrated” versus “not integrated” when you still have the video hardware as a chip on the motherboard. It seems stupid to dedicate space for video when that PCI-X or PCIe slot would be better filled with a RAID or NIC card.

      • Chrispy_
      • 10 years ago

      I’m pretty sure servers use ancient PCI graphics chips for legacy reasons. Most of them spend less than four hours hooked up to a KVM before sending their GUI goodness down a network cable via either remote desktop, VMWare or similar…..

      A lot of server platforms run on Linux these days and for that you just want compatibility, reliability and consistency. If the ATI rage works and does everything you need to install your platform, then why bother changing it?

      • reactorfuel
      • 10 years ago

      If you’re ever using video on a server, something’s gone horribly wrong, or you’re setting it up for the first time. Why not use something cheap and time-tested, when performance doesn’t matter one bit?

      • chunkymonster
      • 10 years ago

      You don’t need more than basic graphics on a server board…it’s not like your to play a round of Crysis on an ESX server…

        • wira020
        • 10 years ago

        That should be the reason to have integrated graphic i think…

      • danazar
      • 10 years ago

      Mature stable drivers. You don’t want a server crashing every couple weeks because the video driver hasn’t been tested and retested for years to not cause ANY problems EVER. Modern graphics cores, including IGP graphics, use newer drivers that haven’t seen the kind of polish and refinement that old graphics chips have. Rage Pro and Matrox 200-series drivers have been made truly stable for most server platforms for years now, and given that stability is far more important than video performance in a server, this is what server users *[

    • Saribro
    • 10 years ago

    Wouldn’t the C32 chips be 4-6 cores instead of 6-8?

      • Game_boy
      • 10 years ago

      Yes. That should be changed in the article.

    • derFunkenstein
    • 10 years ago

    l[

      • Meadows
      • 10 years ago

      Doing the movie playback duties of the entire street.

    • dpaus
    • 10 years ago

    12 cores in 65W is pretty damn impressive, even if they are only 1.7 GHz. Hell, 12 x 2.3 GHz cores at only 105W is impressive.

    No mention of Thuban’s “Turbo Core” capability, which surprises me a bit.

      • Game_boy
      • 10 years ago

      Server chips would not benefit from Turbo, as it makes the performance unpredictable (bad for commercial applications) and increases power usage more than the performance gain.

      AMD’s John Fruehe says he gets far more requests to /[< reduce<]/ the max clock of server processors and memory, to save power. Essentially, when a clockspeed boost or core count increase would give the same performance boost, on the server it's better to have more cores as power use goes up less. On the desktop it would be clockspeed since most apps use few threads, hence Turbo.

        • Sahrin
        • 10 years ago

        There’s effectively no difference between turbo boost and what you’re describing. If they can scale the freuqency up based on load, they can scale it down for the same reasons. (PowerNow already does this, actually).

          • Game_boy
          • 10 years ago

          OK. Dynamic frequency yes (with Powernow), operating opportunistically at higher speeds no, because you might as well use the TDP headroom to add more cores or reduce the TDP value. Any server workload is going to stress most of the cores, not a small number, so there will be few opportunities to run a few cores at high speeds and the rest idle.

          Anyway, go ask AMD’s John Fruehe about it and he’ll tell you the same thing. And it’s not for lack of Turbo ability, because Thuban has it so they could do it if required. But customers don’t want it.

            • OneArmedScissor
            • 10 years ago

            What’s funny is that the Nehalem EX chips have turbo boost, even though those are much closer to what the new Opterons are than the 6-core Xeons. They’re clearly intended for nothing but no holds barred HPCs with the deepest pockets funding them.

            We’re going to have to stop calling these things “server CPUs.” They all have extremely specific target applications now.

            • Krogoth
            • 10 years ago

            Turbo-cooling is disabled though. HPC environment can’t risk getting instability, data corruption and bad results from an overclock. It costs them more money than the productivity boost that an overclock yields.

            • swaaye
            • 10 years ago

            Turbo Boost is a validated clock increase. It’s not overclocking as that term implies that you’re running beyond specification.

            It’s certainly not the usual enthusiast-style unstable wreck that can game for a half hour before BSODing. 🙂

    • Voldenuit
    • 10 years ago

    Johan quite clearly spelt it out that the suitability of the Opteron vs the Xeon (and vice versa) depended heavily on the application and workload profile. So there is no one “winner”, unless it’s the customer.

    • bjm
    • 10 years ago

    Finally some confirmation that Bulldozer will work on the same socket. At least on the server platform that is.

    • Game_boy
    • 10 years ago

    Does anyone know why Anandtech’s review benches the 6174 ($1165) against the X5670 ($1440) rather than the X5660 ($1219)?

    It lead to the reviewer declaring a victory for the Xeon in Cinebench:

    §[<http://it.anandtech.com/IT/showdoc.aspx?i=3784&p=6<]§ when if they had used the 5660, its true price competitor, the Opteron would have won?

      • cygnus1
      • 10 years ago

      Damn good question

      • djgandy
      • 10 years ago

      Way to cherry pick a single benchmark 🙂

      The 5660 would beat it too by the looks of it.

      Also $150 is a tiny cost in a system like this, so it’s generally the best vs the best.

      These sort of servers are likely 8k+ by the time they’re loaded up.

        • Game_boy
        • 10 years ago

        I don’t think it’s cherry picking, because in the other benchmarks, the Xeon’s lead would be less convincing (or the Opteron’s more) as well. I accept that’s the only benchmark where the whole outcome would change, but that’s just to illustrate it.

        And the 5660 would lose, because its clockspeed is 95.6% of the 5670’s so if we assume perfect scaling then it would score 14.67. Even without perfect scaling that’s still lower than the Opteron.

        At the very least I don’t see how he can call that bench for the Xeon when the Opteron isn’t even up against that in price.

          • djgandy
          • 10 years ago

          Read the article you posted. VMWare etc…

        • just brew it!
        • 10 years ago

        $150 is also tiny compared to the cost of electricity over the life of the system. So performance/watt is probably the measure that people will be looking at.

      • bjm
      • 10 years ago

      Maybe he didn’t have one?

        • Game_boy
        • 10 years ago

        Yes, but he said “the Xeon is the CPU to get”. Well, it isn’t, because the Opteron would win against its Xeon price competition.

        If he’d said, “we don’t have a 5660, so scale all our 5670 results down by 4-5% or so” somewhere, and then in that benchmark said it was an effective tie because the Opteron was cheaper by the same % that it was slower, that would have been fine.

        It’s like benching a Phenom II X4 945 against an i5 750 and calling the i5 “the CPU to get” with no other comment. Because the 945 has its own competition you should be comparing it to, not the i5.

          • OneArmedScissor
          • 10 years ago

          This is pretty much always a problem with websites who review only what they’re handed by the manufacturer.

          They can only go with what they’ve got, and anything else they say tends to be speculation they never bother to back up.

          Damned if you do, damned if you don’t.

      • flip-mode
      • 10 years ago

      He tells you why:

      q[< The 6176 looks a bit ridiculous as it delivers only 4% more performance at 30% higher power and 20% higher prices. The real reason behind this CPU is to battle another tanker, the Nehalem EX that Intel is going to launch tomorrow. The TDP and clockspeeds of that huge chip are very similar. If your application scales poorly and you don't care about power consumption, the X5677 is your champion; it is probably the fastest chip on the market for applications with low thread counts. The most interesting parts that AMD offers are the dodeca-core 6174 (2.2GHz), the octal-core 6136 (2.4GHz) and the octal-core low power 6128 (2.0GHz). The 6174 targets those with well scaling multi-threaded applications such as huge databases and virtualized loads. The 8-core 6136 might even be better as most schedulers find it easier to distribute threads and process over a power of 2 cores. Lots of applications also don't scale beyond 16 cores and the chip comes with a 200MHz clockspeed bonus and a very reasonable price.<]q If you read that and look how things line up on his table it makes pretty perfrct sense.

        • Game_boy
        • 10 years ago

        I wasn’t questioning why he used the second-highest Opteron. That’s fine. But he compared it, in the review body, to the X5670 and made little mention of the fact the 5670 was more expensive and that you should scale down its benches to a 5660 to actually decide between Xeons and Opterons.

        His initial table had the correct matchups. But in the Cinebench scores I linked, he /[

          • Sahrin
          • 10 years ago

          And then disabled commenting so no one could call him on it.

          • flip-mode
          • 10 years ago

          The 5670 is $200 more than the 5660, $300 more than the Opteron. In the server world, that’s like paying $0.15 for cheese on your burger. Sure, some people decline on principle, but it’s pocket change.

      • Mr Bill
      • 10 years ago

      I’d say 15.35 versus 14.81 was not a clear hands down call
      §[<http://it.anandtech.com/IT/showdoc.aspx?i=3784&p=6<]§ unless you believe his underlying assumption that most renders only use a couple of cores where core clock speed matters more. Also, the 6174 outperforms all comers on the Linux version of Blender 2.5. The dual Xeon X5670 did get a bit faster between windows and Linux versions they just did not get as large a boost as the dual Magny-Cours 6174.

        • outcast
        • 10 years ago

        Turn the page…don’t just read the cover…

          • Mr Bill
          • 10 years ago

          I don’t mind admitting, I read it all; but that does not mean I understood all the subtleties. “Data mining applications clearly benefit from having “real” instead of “logical” cores.” and “We estimate that the new Opteron 6174 is about 20% slower than the Xeon 5670 in virtualized servers with very high VM counts.” Without knowing how the applications use cores its hard reconcile those two statements.
          Hopefully Damages review will draw out the root causes of these benchmarks.

            • JumpingJack
            • 10 years ago

            I think it is fair to be critical of Johan’s review in this case for a few reasons, none of them his fault.

            – it is a bit short, more data and actual power measurements would have been better.
            – The SAP is estimated and not an actual measure on the MC
            – The rendering benchmarks are interesting, but not actual server class benchmarks.

            I think a lot of this has to do with Johan’s frustration that he only had HW for about a week then had to post a half-assed review. For a launch of a CPU that for all intents and purpose closes the performance gap in 2P to almost nothing, there were very few headlines and a notable lack of data.

            • TravelMug
            • 10 years ago

            There’s also the fact that the CPU reviewed was all over the news a few weeks back that it’s already shipping to customers and partners, yet as of today I don’t see it anywhere in the online configurator on the Dell website. As opposed to the Beckton based 4, 6 and 8 core Xeon 6500 and 7500 series which are available as an option for the PowerEdge R810 even though the CPU has not been officially released yet (=as of this hour when I’m writing the post).

            So maybe they could have sent him a CPU a bit earlier if it was shipping for weeks 🙂

            • djgandy
            • 10 years ago

            Yes power graphs would have been nice!

            • Mr Bill
            • 10 years ago

            MS over at Lost Circuits seems to get the most reliable power numbers. Perhaps he will get hold of one.

    • BoBzeBuilder
    • 10 years ago

    If I’m not mistaken, these are just based on the Phenom architecture with more core.
    Meh, I can wait till Bulldozer gets here.

      • Game_boy
      • 10 years ago

      Does it matter? Since it’s priced competitively against Intel, with similar power consumption, then why not buy it?

      Very few server buyers need the absolute top bin and the high cost/power consumption that comes along with it. Unless you’re in that top 5% bracket, the Opteron seems like a fair choice.

      If you mean as a desktop chip, then of course this isn’t going to be suitable since desktop workloads have few threads. You’re better off buying an i7 or the 3.2GHz Thuban when it arrives.

      • khands
      • 10 years ago

      Everything’s been based off of some iteration of the K10 for a while now, Bulldozer is (IIRC) going to be the first real overhaul in a while.

        • ronch
        • 10 years ago

        And the K10 is based on the K8, which was released in 2003. And the K8 is an evolutionized K7, released in 1999. In fact, we all know that if you look at the architecture of K7, K8 and K10, they’re almost the same.

        It’s about time AMD gives this architecture a rest. Still, it’s amazing how far AMD can push the original K7 architecture.

          • just brew it!
          • 10 years ago

          Yes, the K7-K8-K10 lineage has proved itself to be very adaptable. AMD doesn’t have the resources to start from scratch every few years. If they’d gone down that road, they’d have filed Chapter 11 by now.

          • SoulSlave
          • 10 years ago

          A proof of how freaking great that architecture was…

            • swaaye
            • 10 years ago

            Not as great as the P6 architecture from 1994 or so that is the basis of Nehalem essentially. 😀

            But you can go back to K5/K6 for some AMD (and NexGen) ingenuity. But Athlon was quite a big change. Those Alpha guys they brought in to save the day, and all.

            • axeman
            • 10 years ago

            Also proves what a failure netburst was since it got abandoned for an “new” chip based upon its predecessor. P6 -> Pentium M -> Yonah -> Core2. Although to be fair the guys that developed netburst faced a lot of pressure to increase clock speed no matter what the tradeoffs were. And now current processors have been tweaked to the point we’re pushing up against 4Ghz again.

            • swaaye
            • 10 years ago

            Apparently they just had no clue that semiconductor physics were going to make sure their architecture had no future.

      • ronch
      • 10 years ago

      Yes, that’s what it is. Two Istanbul cores. Still, if your software environment can take advantage of it and you really feel the need for speed now, there’s no sense in waiting for Bulldozer. Besides, when Bulldozer comes out it’s not recommended to grab one right away. It’ll be expensive and who knows if it will still have some bug lurking.

      I’m confident this CPU will be enough for the most demanding tasks. And the ACP figures are very pleasing.

      And yes, I also can’t wait for Bulldozer to come out, but I’ll hold out till 2012 before I get it. Meanwhile, my Phenom II X3 720 should be ok.

      • Rakhmaninov3
      • 10 years ago

      Ironically, the word “Bulldozer” sounds intimidating but real bulldozers move pretty slowly. Somewhat of a departure from quite a few previous AMD chips that were named after fast/athletic things like horses.

    • mdkathon
    • 10 years ago

    Is there any speculation on the Opteron 6128 and use in a non-server environment? It seems like a pretty sweet price for an 8 core processor. I’m not up to date on pricing/availability for ATX compatible boards for these CPUs, or if there is anyone that has been quoted into looking into producing boards.

    • khands
    • 10 years ago

    Woo! They’re here!

Pin It on Pinterest

Share This