Site posts new Zambezi processor specs

The names and model numbers of AMD’s upcoming Zambezi processors might be kind of an open secret—the first leaks can be traced back to mid-March. Actual specifications are another matter, though, and the guys at DonanimHaber may be the first out there with the scoop.

According to their information, the lowest-end Zambezi processor will have four cores, a 3.6GHz base clock speed, and a 3.8GHz Turbo speed. There will also be a 3.3/3.9GHz six-core model, a 2.8/3.7GHz entry-level eight-core model, and a 3.6/4.2GHz eight-core top of the line. (The site also mentions one quad-core chip and one six-core chip whose clock speeds haven’t yet been set.)

Thermal envelopes look like they’ll stay largely at 95W except on the eight-core offerings, which will fit into larger 125W TDPs. Total cache sizes, meanwhile, look like they’ll reflect core counts—4MB for four cores, 6MB for six, and 8MB for eight.

Comments closed
    • j1o2h3n4
    • 8 years ago

    I really hope Zambezi can outperform i7-2600k at its price point. Otherwise, without competition intel would be gods and we buyers more like beggers without a choice. Microsoft has Apple, we need AMD to stand to INTEL.

    • rephlex
    • 8 years ago

    Hmm, why is the FX-8120 model listed with a TDP of both 95 and 125 watts?

      • FuturePastNow
      • 8 years ago

      Presumably because it’ll have two different versions. Several Phenom II X4 processors were available in both 95W and 125W flavors. Generally the 95W versions go to OEMs and the 125W chips go in retail boxes.

    • HisDivineOrder
    • 8 years ago

    Nice speeds. Better than I thought they’d wind up at. That’s good news for AMD and good news for consumers because if AMD is even in the ballpark, Intel will actually have to compete again.

    Still, I’d not consider anything until I see Trinity next year.

    • ronch
    • 8 years ago

    With the 8-cores supposedly delivering performance comparable to the i7-2600, the i5-2500 looks quite compelling, considering it’s not that far behind and has a 95W TDP.

    I really hope BD gives Intel a big slapping, but seriously, it has to pull off some nifty power saving tricks to match upcoming Ivy Bridge 22nm models, being built on a new, immature 32nm process node. I wonder which CPU I’ll eventually end up replacing my Phenom II with.. Ivy or BD. As usual, Performance/Watt/Dollar rules.

      • Corrado
      • 8 years ago

      Well we don’t know what Ivy Bridge will run like. Remember when the Prescott’s came out and no one could believe how hot they ran? They were the first 90nm CPU’s but somehow ran MUCH hotter than the equivalent Northwood @ 130nm?

        • Skrying
        • 8 years ago

        Except Intel has already put out a number of platforms that specifically rely on Ivy Bridge’s decreased heat output and power consumption.

      • xeridea
      • 8 years ago

      Not having to replace your motherboard for every new CPU is a huge perf/cost bonus. Intel has so many sockets I don’t try to keep up, they follow no easy naming scheme, and can have different sockets even with the same CPU class.

        • jensend
        • 8 years ago

        AMD certainly has been much more accommodating to those who wish to have an upgrade path than Intel, but we don’t know yet how that will play out with the next generation of processors. Rumors I’m seeing right now claim that both Zambezi (Bulldozer) and Lynx (Llano) are the end of the line for their respective sockets (AM3+, FM1), which would be disappointing. I’d really been hoping to get a Llano system soon and then get a Trinity upgrade a year and a half or so down the road.

          • shank15217
          • 8 years ago

          Wow you really are unrealistic, Zambezi is a whole new CPU architecture that still runs on the older AM3 boards and Llano has a upgrade patch called Trinity, what else did you want?

            • jensend
            • 8 years ago

            What on earth are you calling unrealistic?

            Zambezi won’t run on most “older AM3 boards”; the main idea with the AM3+ socket is that AM3+ boards can run with old K10.5 cpus.

            Lots of people will buy completely new systems for Bulldozer; buying a 9xx chipset boards with a K10.5 now and switching to Bulldozer in a month or two doesn’t seem to me to be terribly appealing, though if you desperately need something now I guess it helps. But people might have liked to be able to purchase a Zambezi system this year knowing that the enhanced-Bulldozer (Komodo?) processors would be a drop-in upgrade a year or two down the road.

            Instead we have pseudo-compatibility for the K10.5->Zambezi leap and (according to the rumors) no socket compatibility for the more minor Zambezi->enhanced-Bulldozer step.

            That you think the existence of Trinity is a rebuttal to my post suggests that you didn’t bother to read what I wrote, i.e. that the rumors suggest Trinity won’t fit in the AM1 socket. If Trinity fits in the FM1 socket I’ll be quite pleased.

        • ronch
        • 8 years ago

        That certainly is a boon for AMD users. I imagine buying an FX processor and using it with my current mobo, then buying a new mobo to support faster DDR3 speeds later on. Or buying an AM3+ mobo first then an FX processor later. Either way, costs are deferred.

    • OneArmedScissor
    • 8 years ago

    That’s combined L2 cache, not total. I could see them not having the full 8MB L3 for each one, though, despite what that chart says.

    • kamikaziechameleon
    • 8 years ago

    With those high thermal envelopes and clock speeds that leaves me wondering what I could milk by overclocking.

      • derFunkenstein
      • 8 years ago

      Relatively speaking, these are low. 95W for a 6-core and 125W for an 8-core are lower than Phenom II. Even a 95W quad with L3 cache is down from standard Phenom II X4 variants.

    • Zoomastigophora
    • 8 years ago

    When we say cores in reference to BD, are we talking about modules or hardware threads?

      • DancinJack
      • 8 years ago

      Please, not this again.

      [url<]https://techreport.com/articles.x/19514[/url<] A module contains two cores. e:spelling

        • bittermann
        • 8 years ago

        So the 41xx series has 2 modules with 4 total cores? Just trying to get it straight….

          • DancinJack
          • 8 years ago

          Correct. The product names actually specify how many cores they have with the first number. 2 cores per module.

      • OneArmedScissor
      • 8 years ago

      Just cores, like it says. There are eight integer cores.

      If you want to be really technical, you can’t lump the integer and floating point units together as “hardware threads.” They don’t work the same way.

      Modules aren’t terribly important on our end of things beyond how it may or may not affect power gating. That remains to be seen, probably until there’s a mobile version of Bulldozer, which is a ways off. AMD probably would have done better to never mention modules. It’s just confusing.

        • DancinJack
        • 8 years ago

        [quote<]AMD probably would have done better to never mention modules. It's just confusing.[/quote<] +1

          • xeridea
          • 8 years ago

          Modules need to be mentioned, they are a major switch in architecture, like HT, but actually worth something. For people who don’t care, they just say cores, and if you are reading up on them, they aren’t rocket science.

          1 module = 2 integer cores + shared FPU. (which is a + because it is a 256-bit FPU vs standard 128-bit, so software that can take advantage can use it for say 8 32-bit operations per clock, but can always be used as 2 128-bit FPUs (1 by each core in module). L2 cache is shared per module.

          The reason it is important is that it is actually worth doing to pack more cores. Intels HT is basically worthless, because it has a 40% performance hit per thread, so it gets you 20% total boost in best case, and slows down thread processing. If not being used, there is still a small performance hit. It is wasted die space and complexity. So basically HT is marketing hype so they can say they have 8 threads, and review sites n such will say 4 cores (8 threads), like the 8 threads actually means something. Intel will probably have it forever as marketing hype, just like they refused to care about anything but having highest clock speed, even though AMD was crushing them in performance for years with lower clocks. Bulldozer modules are said to have only a 7-8% hit from shared resources, but can actually be faster per thread if not all being used, due to the 256-bit FPU.

            • OneArmedScissor
            • 8 years ago

            The problem is that it looks like a simple marketing term, like hyper threading, but it was meant to be used internally to describe a very complex circuit.

            Enthusiasts, technically oriented as they may claim to be, want pretty bar graphs that are as cut and dry as possible, not highly technical details of circuit designs, so that is what the websites deliver. Unfortunately, unlike hyperthreading, you can’t explain modules that way. They’re not a feature you can switch on and off and show a picture of.

            And thus, we end up with all this BS going around about modules being “1.5 cores” and “hardware hyperthreaded cores,” while the people saying such things never bothered to look into how a lot of the shared circuitry is significantly beefed up and really doesn’t work that way.

            • sschaem
            • 8 years ago

            ?? HT is actually very beneficial in maximizing IPC and transistor efficiency.
            Thinking Intel is a marketing company and not a technology company is a grave mistake…

            HT is here to reduce stalls. Something AMD processor cant do.
            When does a core stall.. very often, and that kills your IPC efficiency.
            Stall happen when waiting for loads, when the store pipeline is full, when a unit is in use, etc…

            You cant issue a division every cycle and sustain throughput. so if a thread does a bunch of division it will iddle,
            with HT the core will continue computing on the other thread.

            And the extra cost is small as it only need small resources duplication.
            This mainly grow the register file, and thats a tiny , tiny part of the chip.

            And in a way AMD is using this for their floating point execution logic.

            • OneArmedScissor
            • 8 years ago

            There you go with the out of context, blanket IPC thing again…

            The funny thing is that on a desktop, if you turn HT off, it speeds some real world things up, while having it on helps almost nothing. The scenarios it’s closer to universally beneficial for are server oriented and rarely applicable to a desktop.

            It’s hard to call that an IPC improvement. It’s not even an IPC improvement in highly threaded applications, only [i<]specific[/i<] highly threaded applications. That makes it purely an application specific feature - much like QuickSync, except that's for video encoding, one of very few things HT benefits! Doh! But it sure does push it to the top of certain synthetic benchmark charts, while running an extra $100 for a desktop quad-core that's generally neck and neck in the real world with its i5 counterpart, lower clock, smaller cache, and all. Not about marketing? Okely dokely, neighbor!

            • sschaem
            • 8 years ago

            cinebench 24% speedup
            x264 27% speedup

            Even if AMD can approach Intel IPC/efficiency per transistor, it will need 5 core to compete with a 4 core HT.

            If you are testing hyperthreading on workload that is not compute bound, but memory bound, HT wont help.
            HT is to boost the IPC, and its highly successful in compute bound apps… the one that need more core, no more memory bandwidth.

            Application that dont benefit from HT most likely wont benefit from more core… bulldozer wont help.

            You can try to spin it the way you want.. HT is a valuable feature that AMD doesnt have, and give Intel the advantage in squeezing the most performance per transistor.
            Look at Atom with and without HT… huge perf gain. Zacate could be 20% faster with this feature.

            • xeridea
            • 8 years ago

            Zagate could be 20% faster if they didn’t have completely worthless graphics like Intel does. Zagate is on par or slightly faster than outdated Atoms and have faster graphics for same power level. If AMD wanted they could give you crappy graphics and boost clock speed, but they go for the more balanced approach. Not that lots of people play games on their netbooks/ultraportables, but it helps a lot with video playback, accelerated web page rendering, and other common tasks. Explain how this gives Atom more performance per transistor?

            • UberGerbil
            • 8 years ago

            [quote<]Look at Atom with and without HT... huge perf gain. Zacate could be 20% faster with this feature.[/quote<]Atom is an in-order processor, and thus benefits enormously more from SMT hiding memory latency and pipeline stalls. Suggesting an out-of-order processor like Zacate would benefit from SMT in the same way betrays either a basic lack of understanding of the technology or a willingness to overlook the facts in order to make a point.

            • sschaem
            • 8 years ago

            Is sandy bridge an In-order processor in your world?
            If not, how do you then explain with all your deep understanding that without HT its 25% slower in CPU intensive task ?

            And Atom get beyond 20% speedup with HT because its in order…

            Really, believing an out of order processor cant benefit from HT is total ignorance of the FACTS, let alone processor architecture.

            • UberGerbil
            • 8 years ago

            I don’t believe an OoE processor would not benefit, and I didn’t say anything of the sort. I said that using Atom’s improvement from SMT as the basis for making the case for its benefit in a processor like Zecate is misrepresentation at best. If you want to argue from Sandy Bridge’s numbers with and without SMT, go right ahead; I wouldn’t have complained. But for some reason you dragged Atom into this, and that’s a spurious argument at best.

        • derFunkenstein
        • 8 years ago

        edit: eh, I’m not wording this the way I want to. I’ll try again later.

      • ronch
      • 8 years ago

      Um, hasn’t there been enough discussion about that already?

      Integer Clusters, my boy. Two of those are in one BD module.

        • bittermann
        • 8 years ago

        Not really for the average joe here… 🙂

        What I’d really like to know is in Win7 task manager am I going to see 2 or 4 cores under the performance tab, with lets say the 41xx series…and are those actually 2/4 real cores or not?

          • BobbinThreadbare
          • 8 years ago

          I think you should see one core per integer unit, so the OS can assign threads to them.

          • jensend
          • 8 years ago

          “average joe” then needs to go back and read what’s already been written; it’s quite clear. A module really is two physical CPU cores, and will show up to software as such; the floating point unit and its scheduler are the only major things shared between the two cores, and since the FPU can do 128-bit or smaller ops (x87, mmx, sse1-4) for both cores simultaneously, the fact that there are shared resources is only going to be apparent when you’re running a lot of 256-bit floating point ops (AVX, introduced with Sandy Bridge).

          Throughout the K10 era, AMD has been doing OK at matching Intel’s floating point performance. When Intel introduced 256-bit AVX ops, AMD saw that by sharing the hardware to implement these ops between two cores they could dedicate a lot more silicon to improving their integer performance, branch prediction and scheduling, etc., which hadn’t been keeping up with Intel’s.

          Intel’s Hyperthreading is really one physical core since the only execution resources which are duplicated are those which store the thread state (data and control registers), so two threads always have to compete for the integer and FP execution units.

          The benefits of Hyperthreading (or symmetric multithreading, which is the non-Intel-branding name for the same idea) come when one thread has a pipeline stall; if this is frequent enough and switching to the other thread can be done quickly enough, you can get better IPC from hyperthreading. The benefits are very very dependent on the tradeoffs you’re making in designing the pipeline and schedulers; it’s not a panacea. If it were a magic +25% performance wand like some people here seem to think, Intel definitely would have used it on Penryn and AMD would have been using it by now; it just wouldn’t have had much of a benefit with the P6/Core or K8/K10 pipelines and schedulers.

          • ronch
          • 8 years ago

          You’ll see in Win7 task manager exactly the number of cores that AMD says the product has.

      • NeronetFi
      • 8 years ago

      I am under the impression there are 2 cores per module. So a Quad Core would be 4 physical cores paired in to 2 Modules.

Pin It on Pinterest

Share This