Bulldozer goes pro with Opteron 4200, 6200 processors

AMD didn’t wait too long to unleash the workstation- and server-bound versions of its new Bulldozer microarchitecture. The company has just announced the “immediate launch and availability” of its 32-nm Opteron 4200 and 6200 chips, which are succeeding the 4100 and 6100 series in the company’s existing Socket C32 and G34 platforms, respectively. Where desktop implementations of Bulldozer max out at eight cores and dual memory channels, some of these new server offerings deliver twice as many of each—all in the same power envelopes as previous-gen models.

Here’s the new Socket G34 lineup. As a reminder, the Socket G34 platform enables two or four CPU sockets, quad DDR3 memory channels per socket, and up to 12 DIMMs per socket:

Model Cores Clock speed Peak Turbo speed L3 cache TDP Price
6282 SE 16 2.6GHz 3.3GHz 16MB 140W $1019
6276 16 2.3GHz 3.2GHz 16MB 115W $788
6274 16 2.2GHz 3.1GHz 16MB 115W $639
6272 16 2.1GHz 3.0GHz 16MB 115W $523
6238 12 2.6GHz 3.2GHz 16MB 115W $455
6234 12 2.4GHz 3.0GHz 16MB 115W $377
6220 8 3.0GHz 3.6GHz 16MB 115W $523
6212 8 2.6GHz 3.2GHz 16MB 115W $266
6204 4 3.3GHz N/A 16MB 115W $455
6262 HE 16 1.6GHz 2.9GHz 16MB 85W $523

…and here’s the C32 family. AMD’s Socket C32 platform enables one- or two-socket configs, with two memory channels and as many as four DIMMs per socket:

Model Cores Clock speed Peak Turbo speed L3 cache TDP Price
4284 8 3.0GHz 3.7GHz 8MB 95W $316
4280 8 2.8GHz 3.5GHz 8MB 95W $255
4238 6 3.3GHz 3.7GHz 8MB 95W $255
4234 6 3.1GHz 3.5GHz 8MB 95W $174
4226 6 2.7GHz 3.1GHz 8MB 95W $125
4274 HE 8 2.5GHz 3.5GHz 8MB 65W $377
4228 HE 6 2.8GHz 3.6GHz 8MB 65W $255
4256 EE 8 1.6GHz 2.8GHz 8MB 35W $377

All of the chips above support DDR3-1600 memory speeds, for what it’s worth. Be sure to check out our Bulldozer review for the skinny on AMD’s new microarchitecture.

In addition to the new chips, AMD says it has “announced the expansion of its 2012 roadmap.” That expansion involves the Opteron 3000 series, a new Bulldozer-based lineup aimed at the same AM3+ socket type used by desktop FX-series processors (though, of course, server-friendly features like ECC memory support will be on the menu). The Opteron 3000 family will debut in the first half of next year, and it will be aimed at “the ultra-dense, ultra-low power 1P Web Hosting/Web Serving and Microserver markets.”

Comments closed
    • brucect
    • 8 years ago

    Finally AMD BD show up against Intel Xeons and shows its not another BS.
    AMD will be fine and gain some market share from Intel. TR is there any benches?

      • drbaltazar
      • 8 years ago

      i can see these proc getting a nice feature as for those saying they re clocked lower not by much lol.also 16 core and just using 8 (1 per module)so as not to split ressource would make lot of sence would still be higher then the latest e serie from intel for about same price!cant wait for benchmark to pop up and test the 16 core at 8 core enabled.i dont care about 16 core we all know where buldozer stand when they slit ressource (window get all confused)maybe they will have something in the 7xxx serie graphic to help on that front like nividia does for they corp gpu released yesterday

        • sschaem
        • 8 years ago

        The ‘splitting’ resource argument is only valid for the FPU. For logic/integer workload not using the 16 core would be silly.
        What AMD bring to the table is super cheap 64 core (integer workload, and possibly code thats FMA heavy) per blade solution under 500watt, with loads of memory.

        With 128 blades per rack, AMD now offers a 8192 processor density for ‘cheap’.
        Bulldozer actually shine in term of power efficiency at lower voltages over previous Opteron models.

        AMD expectation is to manufacture LOADS of Opteron 6200…

        So far only a few website have investigated bulldozer architecture to see what its really made off.
        The key is in power efficiency at lower voltage.

        Fx-8150 3.6ghz 8 core / 125TDP : .2304
        6276 2.3ghz 16 core / 115TDP : .32

        On paper we can see that the Opteron 6276 is 40% more power efficient then the FX-8150 at max usage.
        But I expect the difference to be even more dramatic with real testing.

          • NeelyCam
          • 8 years ago

          Interlagos isn’t even a [i<]native[/i<] 16-core chip - it's two 8-"core" chips bolted together. In fact, I would say that Interlagos actually has two quad-core chips bolted together. Intel has had a true, [i<]native[/i<] 10-core chip out for a long time now. AMD is so behind its competition that it's not even funny anymore.

            • clone
            • 8 years ago

            why is Dell selling them…. the fools, the silly fools?

            [url<]http://www.crn.com/news/data-center/231902878/latest-dell-poweredge-server-doubles-performance-behind-amds-16-core-opteron-chip.htm?cid=nl_alert[/url<]

            • sschaem
            • 8 years ago

            And AMD priced their chip accordingly.

            Xeon E7-8870 10core: $4500 , 130w, release in June 2011
            Opteron 6282 16core: $1100, 115w, release in Nov 2011

            [url<]http://www.spec.org/cpu2006/results/res2011q4/cpu2006-20111020-18712.html[/url<] [url<]http://www.spec.org/cpu2006/results/res2011q3/cpu2006-20110729-17779.html[/url<] Specint 2006: 1040 vs 1080 ... Not that big of a delta for a chip that is 4 time more costly and has higher TDP. But for floating point, yes, the Xeon is almost 60% faster.

            • ish718
            • 8 years ago

            I don’t think Redhat OS has hypervisor support for bulldozer yet…

    • brucect
    • 8 years ago

    double post

    • Bensam123
    • 8 years ago

    AMD could salvage their desktop variants by bringing these parts down there… With all these offerings it almost looks like AMD shot itself in the foot for the sake of product segmentation and the tiny little server segment they still somehow hold onto.

      • Bensam123
      • 8 years ago

      12-16 cores (6-8 real ones) might actually be able to compete with SB-E depending on what you’re doing. It’s a shame they chose to segment them off based on sockets.

        • sircharles32
        • 8 years ago

        Clock speed is the issue.
        In order not to require liquid nitrogen, they have to be clocked WAY down from what Zambezi is.
        With the clocks so low (2.1 – 2.3 GHz), they’d get crushed in desktop (non heavily threaded) workloads.
        So, I don’t think having more FPUs would help in the typically lightly threaded desktop world, especially at such low clocks.

        Zambezi either needs to be clocked through the ceiling, or be redesigned so that it has much better IPC with lower clocks.

        Throwing in more processing units, will not help it on the desktop.

    • odizzido
    • 8 years ago

    A chip with SE at the end at the highest pricepoint? That looks really strange to me since I’ve always seen an SE product as the Senile Edition.

      • just brew it!
      • 8 years ago

      At least it isn’t “SX”, as in the 386SX and 486SX (which were crippled versions of their “DX” bretheren). I guess the “SX” stood for “SuX”!

    • ronch
    • 8 years ago

    I hope server-oriented benchmarks are coming up. If BD makes a good showing there, it would at least improve perception of it. People may forgive AMD and BD’s weaknesses as a desktop processor if they realize BD’s design is sound for server-oriented stuff.

      • [+Duracell-]
      • 8 years ago

      Anandtech is working on server benches with the 6276, I believe.

        • ronch
        • 8 years ago

        It’s up now. BD doesn’t break any new barriers, but at least it’s decent.

    • ish718
    • 8 years ago

    [quote<]AMD has worked to ensure optimization and/or support on many commonly used server operating systems. Linux 2.6.37, Windows Server 2008 R2 SP1, Xen 41, Ubuntu 11.04, and VMware vSphere 5.0 already have OS hypervisor support for Bulldozer, while others such as RedHat Enterprise Linux 6.2 and Windows 8 Server are currently in development. ~Dailytech [/quote<] Why AMD didn't do this for windows 7?

      • JMccovery
      • 8 years ago

      Maybe because Windows 7 is not a Server OS?

      It is possible that if there will be a SP2 coming for Windows 7, that this code could be included.

        • StuG
        • 8 years ago

        Windows 2008 R2 is based off W7, so don’t be so sure about that.

        • ish718
        • 8 years ago

        So that would AMD’s excuse?
        AMD: since windows 7 is not a server OS, all you guys who bought the desktop bulldozers are screwed, for now atleast.

        Yeah, that makes a lot of sense…

    • marvelous
    • 8 years ago

    It seems AMD is all about more cores since it can’t quite catch up to the latest Intel processors clock for clock.

    • Xenolith
    • 8 years ago

    Are the cores gimped like the desktop counterpart. In other words, are there only 8 FPUs in the 16 core versions?

      • Game_boy
      • 8 years ago

      We’ve been through this. Phenom II had 1 128-bit FPU per core. Bulldozer does too. Unless you are running AVX code there is no gimping.

        • Xenolith
        • 8 years ago

        I’m trying to catch-up, so relax. I just did some reading and it looks like the FPU that is in each module is two 128-bit FPU modules combined into one 256-bit FPU.

        I am interested in building my own personal supercomputer for modeling. It would be heavy floating point work. Cray picked Interlagos, so there must be something good going on here.

          • Deanjo
          • 8 years ago

          [quote<] Cray picked Interlagos, so there must be something good going on here.[/quote<] Yup they sure did, it was relatively cheap to use to feed the tesla setups where most the heavy workload occurs.

          • Game_boy
          • 8 years ago

          Because supercomputer contracts are so widely publicised (as you just proved, thinking that Cray using it means it’s good), sometimes the deals are heavily discounted (as if it was marketing spending by the CPU maker).

          • WillBach
          • 8 years ago

          There are a few cases where AMD Bulldozer makes sense. Namely, if you’ll be:
          [list<] [*<]using [url=http://en.wikipedia.org/wiki/FMA_instruction_set<]FMA[/url<] [/*<][*<]making your own chipsets [/*<][*<]writing your own coherency library [/*<][*<]using old AMD sockets [/*<][*<]using very particular workloads that stress RAM read bandwidth [/*<][*<]need a very low absolute price[/*<] [/list<] Do you know exactly what kind of applications you'll be running?

        • ish718
        • 8 years ago

        Yeah, except the 128 bit FPU in Phenom 2 doesn’t have to share resources with another symmetrical 128 bit FPU as in Bulldozer…

          • BobbinThreadbare
          • 8 years ago

          All units on one die have to share some resources.

            • ish718
            • 8 years ago

            You missed the point. *Refer to Bulldozer and Phenom II architecture diagram*

            Bulldozer’s Flex FP is not identical to Phenom II’s FPU…

        • sschaem
        • 8 years ago

        Bulldozer share one FPU across 2 core, and that FPU comes with two 128bit floating point execution units TOTAL.
        So when two threads are running on a module, both thread need to share this single FPU.

        The result is that Bulldozer can only sustain *one 128bit* SIMD fpu instruction per cycle per core.
        While Intel can sustain 3x 256bit AVX operation per cycle per core.

        Without code that heavily rely on FMA, Bulldozer FPU theoretical throughput is even much lower then Phenom.

      • ish718
      • 8 years ago

      There are two shared symmetrical 128bit FPUs in each module, they share the scheduler and other resources. Unless the application is highly threaded and optimized for bulldozer, a 8 core will perform like a 4 core and so on.
      There is no desktop application that is optimized for bulldozer’s Flex FPU, that’s why it performs like sh** compared to Sandy Bridge in applications that make extensive use of the FPU.

    • provoko
    • 8 years ago

    Where are the 16 core benchmarks? Is it safe to just double the 8 core benchies?

    Edit:
    Thanks for the answers, sorry that I smell, next time I’ll take a shower before posting. Haha.

      • derFunkenstein
      • 8 years ago

      If you hadn’t noticed the 16-core chips are clocked significantly lower. So no, it’s not safe.

      • Chrispy_
      • 8 years ago

      No. it’s never double in my experience, even when comparing cores running at identical clocks.

      1.8x improvement is an optimistic average, where optimistic means that the majority of what you’ll be running is extrememely well threaded. Realistically, that’s not very realistic 😉

      • chuckula
      • 8 years ago

      [quote<]Is it safe to just double the 8 core benchies?[/quote<] Not even close. You have to remember the great differences between a single 8-core FX chip and the 16 core MCM modules. First, the MCM modules even at 140 Watt TDP have turbo clocks that are well below even the base clock of the FX-8150. Second, in a server application there will be far less turbo going on due to the fact that you should be stressing most of the cores most of the time in a server workload that will show good performance results on this chip. Third, the MCM introduces some degree of overhead when there needs to be communication between cores in different dies in the module, or if there needs to be a memory transfer from one bank of RAM connected to one die to a core in the other die. Having said that, the benchmarks that people care about in the server world are not the normal benchmarks you see on TR/anandtech/etc. These chips would be atrocious at playing video games, but they'd destroy a high-end FX-8150 at web server and database applications. The trick will be in seeing how well they scale and how well they do against next-generation Xeons.

      • Chrispy_
      • 8 years ago

      I have no idea why poeple have downrated your [i<]questions[/i<]. What is wrong with you people? Are questions no longer welcome on these comment sections? Will I be downrated for [i<]asking questions?[/i<] Go, crazy social-experiment, go!

        • provoko
        • 8 years ago

        Thanks for pointing that out, I guess I smell, haha.

        • destroy.all.monsters
        • 8 years ago

        He provoko-ed them into it!

    • bcronce
    • 8 years ago

    Hope to see benchmarks soon and I hope they fixed at least some of whatever was wrong with the first BD chips… /crossesfingers

      • OneArmedScissor
      • 8 years ago

      They’re identical silicon. What they “fixed” is not overvolting them to run the core and L3 cache clocks wildly out of sync for marketing bullet points.

        • bcronce
        • 8 years ago

        If the regular BDs were over-voltaged, then hopefully these ones have decent power draws, since power increases by the square of the voltage.

        I really want AMD to do well.

    • Dposcorp
    • 8 years ago

    6276 16 2.3GHz 3.2GHz 16MB 115W $788
    6274 16 2.2GHz 3.1GHz 16MB 115W $639
    6272 16 2.1GHz 3.0GHz 16MB 115W $523
    6238 12 2.6GHz 3.2GHz 16MB 115W $455
    6234 12 2.4GHz 3.0GHz 16MB 115W $377

    Those arent bad prices for what you get; look at the starting prices:
    12 cores for $377, 16 cores $523.

    Good boards are found around $400, like this ASUS KGPE-D16 SSI EEB 3.61 Server Motherboard Dual Socket G34 AMD SR5690.
    [url<]http://www.newegg.com/Product/Product.aspx?Item=N82E16813131643[/url<] 64GB Ram support, multiple Gigabit and PCI-E X16 connections, SAS onboard. If I still did a lot of photo and video editing, Id be tempted, even though Intel has the better IPC; this aint bad.

    • slaimus
    • 8 years ago

    The oddity is the 6204 which is a 4 core MCM with no turbo that costs the same as the highest priced 12 core MCM.

    Maybe a special order item for memcache servers? Even there you would get the much cheaper 8 core instead.

    Weird

      • khands
      • 8 years ago

      Yeah, I don’t get that one at all.

      • WillBach
      • 8 years ago

      Could it have only one core activated per module? That, with the higher clock speed, could make it a good choice for some workloads. The socket and the MCM design mean that each non-shared module would have it’s own memory controller, if they really did disable half of the integer units.

        • OneArmedScissor
        • 8 years ago

        Their chart doesn’t say, but it’s 2 x 2MB L2, so it’s just two modules, one on each chip. They must have had something specific in mind related to L3/memory access speed.

      • Arclight
      • 8 years ago

      Weird for sure and the specs look good imo, i mean 4 cores clocked at 3.3 GHz, 16 MB L3, 115W…but then again this is Bulldozer so performance is lackluster regardless of the spex.

    • NeelyCam
    • 8 years ago

    “Immediate launch and availability”? I sure couldn’t find them available on Newegg (while SB-E was in stock)…

    To me this sounds like AMD felt they needed to do a reactionary paper launch to battle Intel’s real one.

      • chuckula
      • 8 years ago

      Yeah… it is a little interesting how this announcement is “coincidentally” made on the same day that SB-E became available….

        • Mourmain
        • 8 years ago

        The only thing that’s underhanded about it is that it’s a paper launch. Otherwise, it’s a good thing to let your customers know what you’ll be able to offer them in the future…

      • Goty
      • 8 years ago

      Retail availability != OEM availability.

        • DancinJack
        • 8 years ago

        This.

        Also, why is anyone surprised this announcement came on the same day as SB-E? This is how things work.

      • Yeats
      • 8 years ago

      My vendor has some, and other vendors have certain models – like the 6234 – in stock and ready to ship.

      • BaronMatrix
      • 8 years ago

      Look again. They have 4 on Newegg.

        • derFunkenstein
        • 8 years ago

        This is a day later. Hindsight is 20/20 and all that.

        • NeelyCam
        • 8 years ago

        Link? Can’t find them…

          • boomshine
          • 8 years ago

          you’re blind dude, how come you could post here? kidding aside, HP already have models with opteron 6200 series, ill be ordering 3 2p servers later 🙂

          • kc77
          • 8 years ago

          [url<]http://www.shopblt.com/cgi-bin/shop/shop.cgi?action=enter&thispage=011003000506_BLE8722P.shtml&order_id=!ORDERID![/url<]

            • NeelyCam
            • 8 years ago

            That’s not Newegg

      • clone
      • 8 years ago

      AMD is filling OEM orders and cares next to nothing about the occasional sale of a server chip on Newegg perhaps?

      I don’t believe Newegg is the world, I could be wrong but I’d have to assume the market for these cpu’s is 1st to government’s and 2nd to Sun / Oracle / IBM and 3rd to OEM’s and 4th to Newegg.

      • sschaem
      • 8 years ago

      Yesterday a few online shop had a few of the 6200 model in stock for immediate shipping.
      Earlier today newegg joined that list.

      If you want to buy a bunch you can…

    • chuckula
    • 8 years ago

    OK: Quick breakdown of this information (wouldn’t it have been nice if AMD had sent out samples so TR and other review sites could test these chips?):

    1. The 4-series chips: Take your desktop Bulldozer, downclock it to the (pretty low) clock speeds and increase the price and BAM you’ve got a 4 series chip with the accompanying performance hit. Don’t expect any miracles.

    2. The 6-series chips: These are more interesting since you now have up to 16 cores in a 2-chip MCM module. Expect the 8 and 4 core chips to perform *worse* than equivalent 4 series chips because of the added complexity of the MCM (these chips just exist for socket compatibility). The 16 core chips, however, are actually interesting. At some server tasks these chips should do quite nicely simply because of the core count. The TDP is a little scary, but if you can get away with a single socket 16 core setup vs. a 2 socket setup, then you could actually end up saving some power.

    It’s pretty obvious that AMD didn’t send out review samples of these chips for a reason. Having said that, the 16 core 6200 series chips have some potential in a pretty good number of server applications.

    Edit: I see the AMD fansquad is out in force today. Amusing how I’ve been downmodded even though I said that AMD’s 16 core chips could be useful in a bunch of server tasks. Apparently any post that doesn’t conclude by calling John Fruehe an Intel fanboi is going to get downmodded by the AMD squad these days.

      • khands
      • 8 years ago

      The 4256 EE looks interesting from a power standpoint, I hope some enthusiastic OC’er gets one and sees what can be done with it.

      • Game_boy
      • 8 years ago

      You buy server chips for the increased support and guarantees, not because they are physically better than the desktop chips. The clocks can’t be rated as high because AMD has to support them for 24/7 100% load.

      Server customers wouldn’t buy desktop chips even if they are better value because they need that guarantee.

Pin It on Pinterest

Share This