AMD begins shipping Brazos, announces Bulldozer-based APUs

Hot on the heels of that Brazos hands-on demo we were treated to last week, AMD shared some fresh, juicy details about its upcoming products at this morning’s Financial Analyst Day presentation.

First, while we’re on the subject of Brazos, AMD said the first Zacate and Ontario accelerated processing units (APUs) began shipping from its assembly facility in Singapore early this morning. The first retail systems based on them will be out "in the January time frame." AMD also put an end to the rumors about delays to its Llano APU, saying volume production will begin in the second quarter of 2011 and shipments will kick off by the middle of the year. So much for those Taiwanese sources, huh? Finally, it was revealed that the first processors based on AMD’s new, high-end Bulldozer architecture are on track to ship for revenue next summer.

Not content to tease the new products with words and PowerPoint slides, AMD treated attendees to live demonstrations of Brazos, Llano, and (Bulldozer-based) Zambezi systems. The demos have, happily, made it up onto YouTube:

AMD’s Rick Bergman went on to show some of those roadmap slides we all love so much, starting with some desktop goodness:

We already know about Zambezi, Llano, Ontario, and Zacate, but the big news today was AMD’s planned 2012 offerings. At the very high end, Komodo will feature eight second-generation Bulldozer cores (or "up to 10" cores, according to the press release), while Trinity will bring Bulldozer to the mainstream as part of a DirectX 11-class APU. Both Komodo and Trinity will be 32-nm parts, just like Zambezi and Llano. At the low end, Krishna will succeed Ontario and Zacate with up to four second-gen Bobcat cores manufactured using a 28-nm fabrication process—presumably of the bulk variety, either from TSMC or GlobalFoundries.

Trinity and Krishna will also make appearances on the mobile side of things, although there, single-core configurations will remain available at the low end. Note the "tablets" segment in the left column, which Ontario and Zacate are slightly poking out into, as well. Later in the Analyst Day presentation, AMD executives hinted that Brazos-based slates will show up early next year alongside the netbooks, consumer ultraportables, and nettops we were already expecting.

Finally, in the buttoned-down world of servers, AMD is prepping offerings with 20 second-generation Bulldozer cores. GlobalFoundries’ 32-nm SOI technology will remain the fab process du jour, too. It looks like AMD’s love affair with APUs won’t manifest itself in this realm, although of course, server folks looking for a little stream computing love can always pony up for some FireStream cards.

Comments closed
    • sweatshopking
    • 9 years ago

    I’m not going to say anything trollish. People need to realize that APU’s are the future. So seriously. for the benefit of all, lets keep this thread appropriate, on point, and objective.

    • Duck
    • 9 years ago

    omg even the guy in the vid says Llano wrong. 2 Ls make a Y sound. It’s spanish. Eee-yarno. Bit like that. A spanish accent helps.

      • JumpingJack
      • 9 years ago

      You haven’t spent much time in Texas have you…. AMD has a major presence in Austin, and Llano is no doubt a reference to Llano, TX and it is indeed pronounced Lan-oh.

      While you are indeed correct, double ll’s in spanish produce the ‘y’ phonetic, if you were to stand in town square and sing how much you love or hate y-an-o you will get very many a strange stare focused your way. Better yet, if you got argumentative with a Lan-oh -an of the wrong demeanor, you may just get the snot beat out of you.

        • stdRaichu
        • 9 years ago

        Sounds even weirder to those of us who grew up in Wales, where double-L has a sort of “Cthl” sound. Llano would be pronounced Thlar-no.

    • burntham77
    • 9 years ago

    A six-core, 32nm Bulldozer chip with a reasonable wattage rating in 2011… I smell new system build!

    • sschaem
    • 9 years ago

    I will be very curious to see benchmark results between a Zacate E-350 and a i3 330UM since both are rated at 18w
    If we have bulk chip pricing from AMD, please add any Intel chip priced at the E-350 price range (+-5%)

    Also Please include web browser benchmarks this time around…
    And if possible include a common multitasking scenario.
    Like run the benchmark suite, with and without running an app like wimamp streaming AAC from the web over wifi

    Thank you Santa

      • jdaven
      • 9 years ago

      Most likely the Zacate CPU performance will be way lower and the Zacate GPU performance will be higher compared to that 18w Core i3. No surprises.

      However, this is not a valid comparison. Zacate is tailored to be low cost. The final price will probably be below $100. That 18w Core i3 I believe is almost $300.

      The real comparison is between Atom dual core and maybe slightly cheaper 45 nm CULV Core 2 Duo’s but these are still expensive as well.

      edit:
      Everyone keeps forgetting price. Price is king!

        • sweatshopking
        • 9 years ago

        You’re still alive!!! where have you been!!

    • kamikaziechameleon
    • 9 years ago

    So bulldozer is gonna be a new socket right? I’m just about to upgrade and was wondering about a AM3 socket life span.

      • Deanjo
      • 9 years ago

      AM3+ or AM3R2 socket. It remains to be seen if they are going to be able to be used in older AM3 boards. It might be a case of a BIOS upgrade for some boards. AMD does have the 900 series chipset on the roadmap so it might be required to get full functionality out of the Bulldozer cores.

      More info here:

      §[<http://www.xbitlabs.com/news/mainboards/display/20100924152855_AMD_to_Introduce_Bulldozer_Compatible_Core_Logic_Sets_in_Q2_2011.html<]§

        • stmok
        • 9 years ago

        Its not the chipset being the factor. The problem is the electrical requirements of the newer design. AMD had little choice (engineering-wise).

        It was either make Bulldozer not backward compatible.
        OR
        Introduce another 6 to 12 month delay and make a backward compatible version.

        This latter choice wasn’t an option, because customers from various markets were tired of hearing about Bulldozer, but not seeing any product being released.

        AMD also did a little research and found very few people actually open up their desktop computers to upgrade their processors. (The majority upgrade their whole computer.)

        So they decided on a compromise:
        * Socket AM3+ CPUs aren’t backward compatible with existing mobos.
        * Socket AM3+ mobos are able to accept Socket AM3 processors.

        Basically…Forward compatible, but not backward compatible. This will allow you to carry your existing AM3 processor into 2011. (AM3+ mobos will be released before Zambezi arrives.)

        In 2011, it doesn’t matter if you choose AMD or Intel…You will need a new motherboard.

          • kamikaziechameleon
          • 9 years ago

          So do you think that it will support DDR3 Tripple channel?

            • stmok
            • 9 years ago

            Nope. Its dual-channel.

            • kamikaziechameleon
            • 9 years ago

            Well that is stupid as long as you make a new socket just make the jump, otherwise they will have to in 12 months time anyways.

          • smilingcrow
          • 9 years ago

          “AMD also did a little research and found very few people actually open up their desktop computers to upgrade their processors. (The majority upgrade their whole computer.)”

          Translation. AMD’s marketing droids who have been pushing backwards compatibility as a big feature (in lieu of much else to push at times) have now switched track 180%. Bill Hicks was so right about people who work in marketing. §[<http://sennoma.net/main/edits/Hicks.html<]§

            • yuhong
            • 9 years ago

            Anyone remember the iMac G5 vs iMac G5 iSight about how easy it is to open?

    • amirol
    • 9 years ago

    Dear DEER AMD Go a head ..

      • dpaus
      • 9 years ago

      Oh, good – they were waiting for you to make up your mind…

    • Hattig
    • 9 years ago

    Bulldozer’s architecture makes complete sense. Separate int cores. Shared FPU because FPU instructions on average are far less frequent than int instructions. FPU still enhanced though, can can be used as 2×128-bit FMAC FPUs or 1×256-bit FPU. Overall performance of a module versus a single core equivalent is 1.8 times faster, but in far less die space than a full dual-core implementation. This compares very well with HyperThreading.

    If you’re doing bazillions of FPU instructions (the one situation that the Bulldozer arrangement might not be optimal) you should be looking at OpenCL / GPGPU anyway.

      • sschaem
      • 9 years ago

      on target!

      AMD see OpenCL (or directcompute) running much better on the GPU side then on a beefed up 256bit SIMD CPU.

      SIMD was (still is for probably another 24month) a crutch.
      It was a low hanging fruit to beef up computing, but now SIMD dispatched by a CPU scheduler is hindering performance.

      I think Apple and AMD (and at the forfront nvidia) got it right, and Intel is either way to early (larrabbe) or way to late (lake of opencl drivers)

      AMD never been so well positioned with their designs for windows notebook/netbook

      But nvidia wont be easy to displace for Android tablets…

    • Edgar_Wibeau
    • 9 years ago

    (deleted)

    Damn, wrong place, should have been an answer to thread below 🙁

    see §[<https://techreport.com/ja.zz?id=517876<]§

    • Vaughn
    • 9 years ago

    too many for you to count!

    • bdwilcox
    • 9 years ago

    Man, who knew Mel Gibson could wakeboard so well?

    • Deanjo
    • 9 years ago

    Hope like hell you will be able to disable those graphics cores easily in 2012. I don’t need an IGP so why the hell would I want an under performing graphics solution adding more heat to my CPU.

      • OneArmedScissor
      • 9 years ago

      32nm CPUs are absolute flamethrowers. Your computer will melt even when the entire GPU is powergated.

        • NeelyCam
        • 9 years ago

        l[<"GF 32nm SOI based CPUs are absolute flamethrowers. Your computer will melt even when the entire GPU is powergated."<]l Fixed that for ya.

      • stmok
      • 9 years ago

      The point of GPU-based IGP is really about future GPGPU roles.

      You want to dump highly parallelized loads to a GPU-based solution where possible. Integrating GPU into a CPU allows one to then have the discrete video card doing 3D work, while the IGP’s cores is used for OpenCL/DirectCompute roles.

      The Desktop slides for 2012 clearly shows only Trinity (Budget/Mainstream) and Krishna (SFF) will have IGPs by the r[<*[

        • Deanjo
        • 9 years ago

        r[

          • Voldenuit
          • 9 years ago

          Hm… interesting. Does that mean that it’ll require yet another new socket (since AM3+ has no video out)?

          Or does it mean that it has the GPU cores but they’re only usable for compute purposes, and that you will still need a discrete GPU to drive a display?

            • Deanjo
            • 9 years ago

            How can AM3+ have no video out? The boards would still have to have video out as it would be the only way to get the video from the APU.

          • Game_boy
          • 9 years ago

          No, the slide actually means it can be paired with a discrete DX11 GPU. Which is obvious. It would say APU if it was on-die. The “AMD Analyst Day Decoder” PDF on AMD’s website explicitly says discrete.

          • stmok
          • 9 years ago

          l[http://blogs.amd.com/work/fadcodenames/<]§ /[<*[<“Komodo” Market: Server and Performance Desktops What is it? “Komodo” is AMD’s next generation _[

    • tejas84
    • 9 years ago

    Poor AMD.

    I feel sorry for them

      • jalex3
      • 9 years ago

      fail troll is fail? you feel sorry for amd having a great laptop solution and a sweet 8 core?

      • burntham77
      • 9 years ago

      Your feelings for AMD mirror your feelings for yourself.

    • cqcumber
    • 9 years ago

    valencia, interlagos, sepang, man, im sure whoever came up the code name is a Formula 1 fan.
    all those names r world famous racing circuits.

      • shank15217
      • 9 years ago

      Intel named cores after horses and rivers and lakes, maybe they are nature fans or more likely they want a consistent naming scheme for all their projects that is easy to remember.

      • sweatshopking
      • 9 years ago

      LOL. yes, yes they are. You’re a quick one aren’t ya 😉

    • plasticplate
    • 9 years ago

    Wonder how SSK is going to troll this topic…….Maybe start an Intel/AMD war?…..

      • jdaven
      • 9 years ago

      I’ll help get things started.

      AMD is goin’ whoop up on Intel with those dozers.

        • NeelyCam
        • 9 years ago

        You’re lost and confused.

        1) No benchmark data on BD. So far it’s all a couple of random samplers.
        2) GF 32nm SOI has yield issues.
        3) What’s the price on those bulldozers? SOI is expensive, you know.
        4) SandyBridge is coming out very soon, while BD volumes are still far away.
        5) BD can’t possibly compete with IvyBridge, and if the 32nm SOI delays get any worst, IB is BD’s direct competitor.
        6) BD doesn’t even have an IGP. SB does.

        This is game over before it even begins.

          • stmok
          • 9 years ago

          l[<2) GF 32nm SOI has yield issues.<]l Nope. The yield issues is because of Llano's complex GPU+CPU design. The problems have been resolved at the cost of a 6 month delay. (Supposed to be released early 2011. This has been pushed to mid-2011.) Bulldozer is fine with 32nm SOI. Prototypes meet AMD's yield requirements. Engineering Samples will be released in December to selected partners for qualification and testing of motherboard designs. l[<6) BD doesn't even have an IGP. SB does.<]l FAIL in understanding the various markets. ...There's three versions of Sandy Bridge. (1) Budget to Mainstream => LGA1156 will be replaced by LGA1155. (2) Performance desktop and low-end Dual-CPU socket systems. => LGA1366 will be replaced by LGA1356. (3) High-end Dual-CPU and Quad-CPU socket systems. => LGA1567 will be replaced by LGA2011. Only (1) has IGP. (2) and (3) do NOT! Bulldozer is in the same league as (2) and (3). NOT (1). Llano will be in the same league as (1).

            • MadManOriginal
            • 9 years ago

            Same general components, yes, same league, absolutely not. Llano is basically Phenom II-class with an IGP and on the CPU side that was just about competitive with Core 2. SB’s graphics are much improved even over the already on-par i3/i5 graphics and will probably be roughly competitive with AMD 80 shader parts but on the CPU side SB will crush Llano.

            • stmok
            • 9 years ago

            Did you read the context of whole response?
            => Same league in terms of the *[

            • MadManOriginal
            • 9 years ago

            Sheesh, don’t get your panties in a bunch because you chose poor wording by writing ‘same league’ rather than ‘target market’ the first time.

            • Voldenuit
            • 9 years ago

            Why are you comparing SB’s IGP to Bobcat’s 80-shader IGP? Llano will have somewhere between 320 to 480 shaders in its first iteration, and don’t forget that they are programmable (unlike SB’s).

            I expect SB and Llano to be trading blows in the real world (better CPU performance for one, and better GPU performance and flexibility for the other), and it should only get better for AMD in 2012 when Llano inherits the Bulldozer architecture and when (or perhaps I should say *if*) more software supports heterogenous computing.

            • MadManOriginal
            • 9 years ago

            Yes, I was thinking of Bobcat, for some reason I find AMD’s product codenames harder to keep straight than Intel’s, but I could have sworn I remember something about Llano’s IGP being 80 shaders too. In any case even these ‘good’ IGPs are just good slow stuff although for casual or older gaming they are good. While there are some good HPC uses I’ve given up on GPGPU in the near term, say 3-5 years, as something to do the only ‘slow’ things left for most consumers like media encoding. I don’t care about encoding video in low quality to use on a portable device and in general the encode quality for GPGPU is still poorer. Otoh if there is a ULV SB (~18W) it would be fair to compare it to Bobcat.

          • jdaven
          • 9 years ago

          Okay I’ll bite.

          1) True but what do you expect from an unreleased product. This is the only time (before a product launch) where expectations can be irrationally high.

          2) This is just rumor unless you can kindly provide a link from an official GF/AMD rep saying otherwise. BTW, did you see the line about AMD squashing the rumors that Llano is delayed.

          3) Well AMD has been using SOI in all products for about 6 years now and they are consistently less expensive than their Intel counterpart. This hurts their margins maybe but price will be competitive just like now. No reason to think otherwise.

          4) Yes, SB is coming before BD.

          5) The roadmaps show BD coming in 2011 and BD2 coming in 2012 which is the scheduled arrival of IB. So I’m not sure what your point here is. I can say that IB can be delayed too without any proof.

          6) BD has an IGP in 2012. Llano has one in 2011. And yes SB will have one in the beginning of 2011 just like the current 32 nm Core iX’s. What’s your point?

            • NeelyCam
            • 9 years ago

            2) I’m too lazy to look for links, but some three months ago in the AMD earnings conference call, global foundry defect rate reduction was raised as an issue. Or something like that. And that was the reason for the original Llano delay (and pulling in Ontario/Zacate)

            3) The price has always been based on performance/price of the competition… and AMD hasn’t had the performance to keep the price high. If BD really is going to be as good as fanbois expect, it will certainly be priced higher as well. Forgot what happened with the FX line?

            4) AMD has not been delivering to their roadmaps for a few years. Meanwhile, Intel has been ticktocking with consistency for, what, three years now? There is zero reason to think that IvyBridge would be delayed, while there’s plenty of recent history to suggest that AMD might not deliver on time.

            5) Semantics. BD “The 2012 Edition” – also referred to as BD2 – is not the BD I was talking about. AMD will be without a bulldozer+IGP combination for over a year, while Intel has had some for almost a year… and SB (=more appropriate competitor to BD than 2-core Westmere) will have one for over a year before BD.

            Meanwhile, wake me up when Llano will finally be shipping in true volume. I predict a quasi-paper launch much like 5870’s/5870’s of Q4/2009.

            • Voldenuit
            • 9 years ago

            q[<5) Semantics. BD "The 2012 Edition" - also referred to as BD2 - is not the BD I was talking about. AMD will be without a bulldozer+IGP combination for over a year, while Intel has had some for almost a year... and SB (=more appropriate competitor to BD than 2-core Westmere) will have one for over a year before BD.<]q Sandy Bridge EN (the high end desktop part) will not have an IGP either. It really doesn't matter much that these high end parts (including BD) don't have IGPs, as users who buy them will be buying high end discrete GPUs (or GPGPUs) as well.

            • jdaven
            • 9 years ago

            BTW, I’ve always wanted to bring something up with you about manufacturing process advancement.

            Just because a company comes out with a new process doesn’t mean the whole industry transitions from day one. It can take YEARS before the 50% threshold is reached.

            Case in point, Newegg allows you to filter your search based on process. Here is the current breakdown for desktop processors:

            65 nm 5
            45 nm 74
            32 nm 13

            The current process is 45 nm not 32 nm on the desktop. 32 nm is just the most advanced process but not the process currently used on the majority of desktop products. Global shipments for 32 nm will not exceed 50% for probably another year (almost 2 years after introduction).

            The real comparison between AMD and Intel is how fast each company ramps up the shipments of a certain process. If AMD ships greater than 50% of its processors on 32 nm within this year, it will be almost two years ahead of how fast it takes Intel to reach this point.

            I would argue that AMD has the advantage of being more agile in its technology transitions than Intel.

            • NeelyCam
            • 9 years ago

            Interesting comments.

            First, looking at the # of products at particular nodes out there doesn’t necessarily correspond to the actual shipping volume of each node. We all know that Intel has 80% or so of the whole market. AMD obviously has no volume on 32nm, but if a bit over half of Intel’s volume is 32nm, that alone could mark the “industry” transition to 32nm.

            However, Intel does ship 45nm parts because they still sell well, and extracting more profits from those 45nm investments makes good money for Intel. In a way, I buy your argument that AMD can be more agile in switching to a new technology, because they are not (directly) on the hook for paying for the process R&D.

            But I’m thinking that if there ever was a moment when AMD&GF/Whoever threatened Intel’s process/product/cost advantage by switching to a next generation earlier than expected, Intel would be perfectly capable of quickly ramping up their latest technology volume and kill the previous-gen stuff if they really wanted to. To them it’s a profit-maximizing problem more than anything else.

            I would argue that getting the same performance from the next-gen process is going to cost only a little bit less than what it did in the previous-gen… The big benefit is getting the performance lead because of the next gen (so you can charge premiums). But scrapping perfectly capable 45nm fabs just because you can make 32nm cheaper in the long run doesn’t necessarily make economic sense.. there’s a lot of initial investment and time that needs to go into those 32nm fabs before they can ‘break even’. 45nm is already paid for, and as long as competition is behind, you can maximize profits by offering “just good enough”

          • Fighterpilot
          • 9 years ago

          Neely Cam…why do you care?
          We get that you hate AMD….take a chill pill dude.
          Some of the trolling for Intel you do at places like Semi Accurate is almost disturbing.
          Someone from AMD steal your GF or something?

            • Buzzard44
            • 9 years ago

            “Someone from AMD steal your GF or something?”

            No, I believe AMD let GF go off on it’s own to help its debt problem.

            • MadManOriginal
            • 9 years ago

            Sounds about right: shed the GF, get rid of your debt and cashflow problems :p

            • NeelyCam
            • 9 years ago

            Huh. I thought my trolling here is much more disturbing than whatever I do at SemiAccurate.. At SemiAccurate I’m more factual and less extreme… The fanbois are more rabid here at TechReport, so trolling is more entertaining.

      • stmok
      • 9 years ago

      l[

    • KarateBob
    • 9 years ago

    Correct me if I’m wrong, but 8 Bulldozer Cores = AMD Marketing for 4 cores with 8 execution units. So is the Komodo with 8 next-gen Bulldozer “cores” a 4 core, 5 core, 8 core, or 10 core design?

      • OneArmedScissor
      • 9 years ago

      Ugh…not this again. They never should have mentioned “modules” or whatever they called them in the slides.

      It’s however many cores it says. The end.

        • NeelyCam
        • 9 years ago

        Yeah; marketing bs. No better than Intel marketing their quad cores as 8-thread CPUs.

        At least Intel’s are running faster.

          • mesyn191
          • 9 years ago

          Depends. Intel should be faster for single thread but AMD should be faster for multi thread. 1 module is supposed to have 80% the performance of 2 cores. HT gives you something like 20-60% the performance of a 2nd core IIRC. Each approach has real world advantages and disadvantages, its hardly marketing driven tech a la P4’s high clocks.

            • NeelyCam
            • 9 years ago

            It’s not a stretch to say that if AMD says each module is “2 cores”, Intel could say each of their core is in fact “2 cores (*”

            *) “Approximately equivalent to throughput of two standard, non-HT cores with appropriate loads”

            Either way, it’s somewhat dishonest in my opinion…

            • mesyn191
            • 9 years ago

            Actually I wouldn’t have too much of a problem if they (Intel) did that. So long as they’re honest about its performance it’d be fine with me.

            In the end 2 or more cores is what the software and OS will see and that is what really matters.

            Intel’s approach costs less die space but is supposedly harder to do and has less predictable and lower performance. AMD’s approach costs more die space than HT but is still much cheaper than a whole 2nd core while having higher and more predictable performance than HT.

          • shank15217
          • 9 years ago

          You discount a brand new processor architecture with significant redesigns in every part of the core logic and un-core components without a single benchmark, you’re the one that sounds bs to me.

            • NeelyCam
            • 9 years ago

            Same crap Intel proponents were saying about HT. Marketing BS 101

        • shank15217
        • 9 years ago

        Modules makes more sense than cores, read up on the architecture, there are several articles online. Bulldozer features a new form of shared resource threading unlike anything before it.

          • OneArmedScissor
          • 9 years ago

          No, it doesn’t, because modules don’t mean jack crap to how it’s going to work when people actually use them in a computer instead of “reading up on the architecture.”

          AMD specifically said they will not mention modules outside of those slides they had, and they would have done better to just leave it out of there and not create all this confusion and BSing.

            • shank15217
            • 9 years ago

            All ‘people’ see are # of execution threads the processors exposes to the OS, how they do it IS significant. You think these slides are for Joe ‘6’ pack. How dumbed down should they make their presentation? The module concept IS relevant as the FPU and the decode pipelines are shared between the two cores, I’m not even sure what you’re complaining about.

            • OneArmedScissor
            • 9 years ago

            A module is still two distinct integer cores – the only cores that will mean anything to anyone here.

            You can write a novel about how different shared FPUs are if you so please, but that still wouldn’t explain why there’s any need to stop referring to cores just because one particular aspect of them is in the process of being deprecated, while what’s left will function as always.

            They don’t need to dumb anything down. They need to not throw in pointless terms that don’t accomplish anything but confuse people. They even had to edit the Anandtech article that first talked about this because it was such a mess. That’s pretty bad.

            An eight core Bulldozer chip has eight cores. The end.

            • shank15217
            • 9 years ago

            right because enthusiasts don’t run programs that use the FPU… OK so simplify all you want but don’t assume by being simple you are being correct or less confusing. This is a tech site, not yahoo news, if you aren’t interested in the details why bother coming here?

            • Edgar_Wibeau
            • 9 years ago

            There are two FPUs (128-bit FMACs) per Module which can be fused to one 256-bit AVX unit. 2 Cores per module, 2 FPUs per module, AVX shared.

            See: §[<http://www.planet3dnow.de/photoplog/file.php?n=11784&w=l<]§ Shared are: - Fetch - Decode (4-issue which can all feed to one core if second is idle!) - L2 cache (double size of K10) - 256-bit AVX You have - 2 Int cores - 2 128-bit FPUs (FMACs) - 3 independant shedulers Make of it what you will, to me it's far closer to two than to one core(s).

            • poulpy
            • 9 years ago

            Sounds about right to me, nothing utterly shocking there.
            Although after all these years of massive marketing bullshit I’m not sure I’d be shocked by much nowadays.

            • [+Duracell-]
            • 9 years ago

            I do believe the 2 128-bit FMAC units are, in Bulldozer’s case, treated as one FPU unit since there’s only one scheduler, whereas each integer unit has its own scheduler.

      • stmok
      • 9 years ago

      AMD says “cores” => They’re referring to the integer cores. Each module has 2 integer cores.

      For example: a 4 module processor has 8 integer cores.
      => Which AMD will say it has “8 CPU cores” in their roadmaps.

        • KarateBob
        • 9 years ago

        So do we have 8 cores, 16 modules? 8 modules, 16 cores? 8 cores, 4 modules? 8 Modules, 4 cores? HOW MANY COMPLETE CORES ARE THERE? HOW MANY EXECUTION UNITS ARE THERE?

          • stmok
          • 9 years ago

          *[

            • KarateBob
            • 9 years ago

            So essentially, in 2011, with Bulldozer, AMD will finally have a desktop CPU to answer the Core i7 920/940/965, of 2008? In 2012, they’ll tweak it, and that’s their answer to the 4-core Sandy Bridges. But nothing to answer the 980X/970?

            • sschaem
            • 9 years ago

            It seem that AMD feel that if you need more then 4 core, you are an Opteron costumer.
            For consumer app that leverage 6 core, most likely would benefit from OpenCL. And if so, any GPU parall processing will be so much faster then 6 cores.

            AMD seem to have come to the conclusion that for consumer desktop 4 core is the sweet spot going forward.
            Workstation software the benefit from many cpu core & GPU would be designed on opteron flavor or bulldozer.(16 core per chip in 2011)
            Almost 3 times the execution units of a 980x … Who needs that at home?

            • Flatland_Spider
            • 9 years ago

            I think the sweet spot on the desktop will be 16-32 cores with 32GB of RAM. From there the OS could start partitioning resources creating different resource pools.

            The idea is to make the OS more like bare-metal VM hypervisior then the traditional monolithic OS we’re used to. (When I say monolithic, I mean one big program; I’m not talking about kernel architecture.) This pretty much going to be required since everyone wants backwards compatibility, but coders need to be able to break compatibility, when necessary, to achieve better security or performance.

            • smilingcrow
            • 9 years ago

            That all makes perfect sense.

Pin It on Pinterest

Share This