AMD to build ARM-compatible 64-bit processors

Well, this changes some things. AMD has just announced its intention to build 64-bit CPUs compatible with the ARM instruction set, starting with a multi-core SoC targeted at servers and eventually expanding into other markets. The first server-oriented chip will be branded as an Opteron and is expected to ship in 2014, complete with integrated support for AMD’s Freedom Fabric interconnect for dense servers.

I’m going to post the bulk of the press release below, since it offers some additional details worth knowing—and since this is pretty huge news in the CPU space.

SUNNYVALE, Calif. — Oct. 29, 2012 — In a bold strategic move, AMD (NYSE: AMD) announced that it will design 64-bit ARM technology-based processors in addition to its x86 processors for multiple markets, starting with cloud and data center servers. AMD’s first ARM technology-based processor will be a highly-integrated, 64-bit multicore System-on-a-Chip (SoC) optimized for the dense, energy-efficient servers that now dominate the largest data centers and power the modern computing experience. The first ARM technology-based AMD Opteron processor is targeted for production in 2014 and will integrate the AMD SeaMicro Freedom supercompute fabric, the industry’s premier high-performance fabric.

AMD’s new design initiative addresses the growing demand to deliver better performance-per-watt for dense cloud computing solutions. Just as AMD introduced the industry’s first mainstream 64-bit x86 server solution with the AMD Opteron processor in 2003, AMD will be the only processor provider bridging the x86 and 64-bit ARM ecosystems to enable new levels of flexibility and drive optimal performance and power-efficiency for a range of enterprise workloads.

"AMD led the data center transition to mainstream 64-bit computing with AMD64, and with our ambidextrous strategy we will again lead the next major industry inflection point by driving the widespread adoption of energy-efficient 64-bit server processors based on both the x86 and ARM architectures," said Rory Read, president and chief executive officer, AMD. "Through our collaboration with ARM, we are building on AMD’s rich IP portfolio, including our deep 64-bit processor knowledge and industry-leading AMD SeaMicro Freedom supercompute fabric, to offer the most flexible and complete processing solutions for the modern data center."

"The industry needs to continuously innovate across markets to meet customers’ ever-increasing demands, and ARM and our partners are enabling increasingly energy-efficient computing solutions to address these needs," said Warren East, chief executive officer, ARM. "By collaborating with ARM, AMD is able to leverage its extraordinary portfolio of IP, including its AMD Freedom supercompute fabric, with ARM 64-bit processor cores to build solutions that deliver on this demand and transform the industry."

The explosion of the data center has brought with it an opportunity to optimize compute with vastly different solutions. AMD is providing a compute ecosystem filled with choice, offering solutions based on AMD Opteron x86 CPUs, new server-class Accelerated Processing Units (APUs) that leverage Heterogeneous Systems Architecture (HSA), and new 64-bit ARM-based solutions.

This strategic partnership with ARM represents the next phase of AMD’s strategy to drive ambidextrous solutions in emerging mega data center solutions. In March, AMD announced the acquisition of SeaMicro, the leader in high-density, energy-efficient servers. With today’s announcement, AMD will integrate the AMD SeaMicro Freedom fabric across its leadership AMD Opteron-, ARM- and x86-based processors that will enable hundreds, or even thousands of processor clusters to be linked together to provide the most energy-efficient solutions.

We’re seeking clarifications on some points of interest and will post an update once we have them.

Update: Some folks have taken this announcement to mean that AMD will be giving up on x86-compatible CPUs going forward. That’s very much not the case. Even the press release says the ARM chips will come "in addition to its x86 processors for multiple markets."

Comments closed
    • sschaem
    • 7 years ago

    Why do I have a feeling that ARM will make more money out of that deal then AMD ?

    Seem like great news for ARM, it solidify their position. Any way you look at it,
    I dont see any negative for them.

    For AMD it could be viewed as “We give up, we cant make competitive CPU at the high end or low end” . Its also a risk , as it will suck R&D money for an undefined outcome and they are still going to be directly fighting Intel.. but also in direct competition with Samsung, nVidia, etc…

    This doesn’t feel like a well thought out direction, more like a knee jerk reaction.
    nVidia officially announced its ARM server effort in late 2010? Seem like nvidia had a better vision of the future
    Tegra, Testla… both segment AMD could have dominated but didn’t have a clue it even existed.

    So far we know AMD new management team can fire people well, but what else can they do 🙂

    • chuckula
    • 7 years ago

    I often disagree with the SemiAccurate spin, but sometimes Charlie is really on point:

    [quote<]Back to the present, in today’s press conference AMD CEO Rory Reed showed the depths of how poorly he understands what AMD is doing. A simple phrase that will probably be spun as marketing hype was that Reed called the SeaMicro Freedom Fabric (FF) a “supercompute fabric”. This is repeated in the press release, “The first ARM technology-based AMD Opteron™ processor is targeted for production in 2014 and will integrate the AMD SeaMicro Freedom™ supercompute fabric, the industry’s premier high-performance fabric.” The problem with this is that it is technically wrong. FF is many things, but supercomputing is not one of them. It is an amazing backbone tech for a big box of shared nothing (BBOSN) machines, but it is fundamentally unsuited for supercomputing tasks. Anyone who had the vaguest clue about the product would understand this and not make that basic mistake. It is like calling a bicycle a car because both have similar privileges on some roads, analysts might be mislead, but that does not make it true. It is extremely worrisome that Mr Reed does not understand the tech on this level, it isn’t that complex and it fundamentally affects his company.[/quote<] The issue is that the CXO types at AMD seem to have a big disconnect from the technology. It looks like they read shiny magazine covers with the word "ARM" all over them and feel the need to copy what they see. That level of misunderstanding seeps into calling freedom-fabric a supercomputer interlink when SeaMicro obviously never built it for that purpose and never billed its products as "supercomputers" either.

      • NeelyCam
      • 7 years ago

      What’s also worrisome is that Papermaster let this go through to the press.

      • samurai1999
      • 7 years ago

      Charlie is talking nonsense again
      – Seamicro call the Freedom fabric ‘the SeaMicro Freedom™ supercompute fabric’
      [url<]http://www.seamicro.com/node/251[/url<] and it's not just in marketing blurb - it's also in their whitepapers 'A supercomputer-style interconnected fabric was designed ...' [url<]http://www.seamicro.com/sites/default/files/SM_TO06_v1.1.pdf[/url<] So, as far as SeaMicro are concerned, the 'Freedom Fabric' is indeed a 'supercompute fabric'

        • chuckula
        • 7 years ago

        Hrmm.. that’s a good call. I never realized they abused the buzzword like that. I fully agree that FF is a nice interconnect fabric for a rack of low-end blade servers to talk to a SAN/NAS or a network, but it is a far far cry from a supercomputer interconnect. Looks like there’s buzzwords aplenty in these products.

      • bcronce
      • 7 years ago

      I could very easily be mistaken, but many of the top “super computers” are just a bunch of clustered nodes with faster interconnections. Isn’t this what SeaMicro offers? A bunch of CPU nodes all interconnected via some fast “fabric”.

    • dpaus
    • 7 years ago

    I strongly suspect there’s another shoe to be dropped in a few months: Seamicro with GCN as a back-end (not unlike the architecture of yesterday’s Titan announcement, but with a lot more flexibility in choosing between x86 and ARM as the front-end)

    • Silus
    • 7 years ago

    What a shocking surprise!!! AMD is following the trend that many other companies already did a few years ago, when the mobile market boomed! Always late to the party and this is what has brought AMD to its sorry state…extremely incompetent and horrible management. And they call it “bold”, when in fact it’s just AMD going “me too” 2-3 years after everyone already made the same decision…

      • Arag0n
      • 7 years ago

      No one is selling ARM chips for Servers in a compelling platform… AMD may have a niche market if they succeed and mainstream one if Intel is so stubborn to do so and ARM becomes popular on servers.

        • designerfx
        • 7 years ago

        It should also be noted that intel’s not even making ARM chips – they’re trying to compete with them.

      • David_Morgan
      • 7 years ago

      At CES 2011, when nVidia was announcing “Project Denver” and AMD was talking up their brand new Brazos chips, a little AMD birdie mentioned that Brazos was designed to be modular, and the ability to swap out the Bobcat cores on that chip with ARM based units was scribbled on the engineers whiteboards.

      Given that, I’d say that ARM plans have been in the works within AMD for quite a bit longer than yesterday’s announcement.

      Still, 2014 seems sooo far away.

    • Sam125
    • 7 years ago

    Interesting, AMD pursuing low-power was pretty obvious, even before the announcement of going “ambidextrous” but going straight for big iron; very interesting! I hope it works out for them as it’s a brilliant move -if- they can break into the market.

    Scott, please follow these new AMD developments closely!

    • ClickClick5
    • 7 years ago

    Scott, reading your twitter in remarks to Charlie, invite him to the next TR BBQ.
    I think the gerbils here would love to have him as a guest. The hours of conversation would be priceless with a Scott vs Charlie showdown.

    [sarcasam] Vrock would just LOVE to get his autograph! [/sarcasam]

    • ronch
    • 7 years ago

    Just a thought. I think the industry is seeing that Intel is getting more and more powerful and greedy, not wanting anyone else to do x86. They’ve also noticed that AMD is falling behind. At the same time, ARM seems to be gaining a lot of traction in the mobile space and ARM seems to be perfectly fine (for now, at least) with licensing their IP. This is in stark contrast with Intel.

    It was then up to Microsoft to try and shift the industry to a non-x86 architecture by producing an ARM version of Windows. Windows is a powerful brand, and most people couldn’t care less what an ISA is or what it means (heck, some computer ‘enthusiasts’ I’ve talked to aren’t even familiar with the concept of different ISAs). As long as they push their ARM version of Windows like they’re doing now, it will (hopefully) gain traction, and when that happens, they might slowly shift their entire ecosystem over to ARM. And, with the playing field being opened to more hardware vendors as we speak, it’s like the industry is resetting itself and getting out of an unhealthy duopoly where the success of a big company (Intel) can spell doom and gloom for the entire industry because prices will rise and innovation will slow down to a crawl. AMD seems to understand this because AMD also works closely with Microsoft and other industry players. Small conversations have a tendency to balloon into big things sometimes.

    Personally I couldn’t care less if the entire industry shifts over to ARM. With a more open playing field we can expect healthier competition and more innovation will happen. And as we’ve seen with Apple’s computers, shifting ISAs isn’t necessarily going to end the world. Windows will not work only with x86, and x86 will not work only with Windows. It’s the ECOSYSTEM (Windows). Most people don’t like Ubuntu because they can’t run their favorite apps and games anymore even if it’s running on x86, so switching the entire Windows ecosystem to ARM and providing some sort of emulation (the way Apple did it) for legacy x86 apps might very well work out for most people.

      • bjm
      • 7 years ago

      I think you’re making way too much of a story out of the industry somehow turning their backs against Intel. It’s hardly that. If anything, neither Intel, AMD, Microsoft, nor anyone else in the PC world saw the mobile device market exploding like it did. What you’re seeing has nothing to do with the industry being scared of Intel’s “power and greed” and everything to do with the race to place a foothold in the mobile device market.

      They couldn’t collectively come up with a competitive mobile device solution, so they each scrambled as fast as they could to find their own way. Microsoft jumped to ARM because Intel lacked a good SoC, Intel jumped to Android because MS lacked a good mobile OS, and AMD… well, their board fired their CEO, ran in a circle, and eventually decided to team up with ARM. The only PC player who jumped ship early was nVidia, and that’s why they’re sitting fairly pretty right now. Perhaps that x86 ISA debacle vs. Intel a few years back was a blessing in disguise.

      Anyway, I get the sense from your post that you’re portraying Intel as some sort of bad guy here. The PC industry as a whole took a smack to the head to adjust the mobile market explosion, but each big player is adjusting. Microsoft is making good moves, but so is Intel and nVidia. ValleyView/Haswell is sounding awesome right now, for both Windows 8 and Android/Linux.

      Let’s just hope AMD’s most recent move will be as promising for them.

        • ronch
        • 7 years ago

        It’s just a thought. First sentence right there.

          • bjm
          • 7 years ago

          …well, ain’t that the conversation killer.

            • ronch
            • 7 years ago

            Well, regardless, it’s obvious the industry is shifting. It’s not just AMD who is aware of this unhealthy duopoly but many industry players as well. I think the industry is moving in the right direction, deliberately or not.

    • ronch
    • 7 years ago

    Well, things seem to be looking better and better for AMD, having released their Trinity and Vishera chips and now making a bold move such as this. I have a few concerns, however, mainly in the area of funds as well as spreading their company and engineers a little too thin. Can they do it? Developing a 64-bit processor for the high end server market is no joke and a lot of validation will be required. And their continued battle with Intel will make sure those funds are stretched. I mean, they’ve had enough trouble handling Intel alone, what it will be like to compete with many (albeit smaller) ARM licensees? And lest we forget, Intel is also an ARM licensee and can choose to go this way any time they feel like it. With a bank-full of money they can do it on a whim.

    And as for competition in the ARM market, this article is worth a look:

    [url<]http://gigaom.com/2010/05/11/watch-out-intel-marvell-to-make-arm-based-server-chips/[/url<] "With 662 ARM licensees, the potential for competition in server chips extends far beyond the current two primary vendors making them today, although many ARM licensees are likely uninterested in the server space" I hope Rory read (pun intended) this article.

      • chuckula
      • 7 years ago

      [quote<] Can they do it? Developing a 64-bit processor for the high end server market is no joke and a lot of validation will be required. [/quote<] True, but the ARM servers will by no means be high-end. Instead, they will be competing with Atom micro-servers at the low-end of the market for things like cheap web servers. Don't let the 64 bit fool you into thinking high-performance, because Atom already is 64 bit and is obviously not high-performance. The 64 bit support is simply a minimum requirement needed for programs that access > 4GB of memory. That's not to say they don't need validation, but the good news is that these things will likely come in blade racks where you are expecting a few servers to fail here and there and you just swap out the blade cards to replace the faulty servers.

    • eitje
    • 7 years ago

    they should have leap-frogged the competition and done 128-bit CPUs.

      • TurtlePerson2
      • 7 years ago

      Not sure if you’re serious or not, but if so I’ll explain why that’s not such a great idea. A 64-bit computer has memory which stores data into 64-bit chunks. These chunks are capable of storing 2^64 possible numbers, which is about 18 quintillion different numbers.

      If we started using 128-bit computer we could store more than an undecillion different numbers. That’s such a large number I had to look it up. Obviously that sort of storage and accuracy isn’t even remotely necessary. The only thing that we use 128-bit numbers for is encryption and it turns out that we can handle 128-bit numbers with 64-bit machines, but it’s a little bit clunky.

      A 32-bit computer can store 2^32 different numbers in a memory address. This is about 4 billion or 4G in the computer world. Therefore a 32-bit system is inadequate if you want to use more than 4 GB of RAM. Going from 32-bit to 64-bit makes a lot of sense. Going from 64-bit to 128-bit makes a lot less sense.

        • just brew it!
        • 7 years ago

        For certain types of problems a CPU that can natively handle 128-bit (quad precision) floating point could be advantageous. But I agree, there’s no need for 128-bit memory addressing.

          • Prion
          • 7 years ago

          As long as we’re going there, don’t Sandy Bridge/Ivy Bridge already ship with 256bit SIMD registers for AVX?

          • Arag0n
          • 7 years ago

          It depends the point of view… a 512bit or 1024bit and even more SIMD engine can dramatically increase some operations, specially those related to video, image and audio processing. ARM developers use that a lot to speed up processes targeting the NEON engine and that’s just a 64bit wide SIMD. Usually Intel and AMD use a 256bit wide SIMD. The point is that with SIMD engines you can program using vectorial operations. A good usage for those SIMD engines is opacy layers. You can write an equation that represents the process as the addition of one image to another one with some levels of ponderation. You could do the addition as:

          for (int i=0; i<pixelCount; i++)
          {
          Pixel p1 = image1[i];
          Pixel p2 = image2[i];

          p1.red = p1.red*alpha + p2.red*(1-alpha);
          p1.green = p1.green*alpha + p2.green*(1-alpha);
          p1.blue = p1.blue*alpha + p2.blue*(1-alpha);
          p1.alpha = p1.alpha*alpha + p2.alpha*(1-alpha);
          }

          However, using vectorial SIMD of just 32bits instructions you could do as follows:

          for (int i=0; i<pixelCount/2; i+=2)
          {
          PixelArray p1 = image1[i];
          PixelArray p2 = image2[i];

          p1 = p1*alpha + p2*(1-alpha);
          }

          That’s because pixels are formed by 4 components of 8 bits, so you can simplify the operation as a single operation of 4 inputs and 4 outputs. Now think about a SIMD of 512bit… it can process 16 pixels at the same time. We can go up till 66355200bit… that’s the number of bits that you would require to operate with Full HD images as a single data input. Sure it has not too much sense since the operation can be partitioned in smaller parts and all… but you get the idea. Of course not every kind of application can benefit from that kind of big chunks of data processing, but still there is several that do. One good example is physics for game engines. You can program targeting every single component (X,Y,Z, Phi) at the same time. Usually games need 4x32bit’s = 128bit SIMD, but some other components of objects can increase the complexity and you may have some use for extra bits.

          The point is, the largest single data that is recognized by the CPU, does it have sense to be more than 64 bits? But CPU’s with inner data paths designed to process data in 512, 1024 or 2048bits wide buses have sense. It’s up to developers and compilers to make it worth. That’s why I laugh when I see compilers targeting new instructions and people expect dramatic gains… compilers can search for pasterns and vectorized operations, but developers need to program at least according to that expectation and usually they will be required to use intrinsics to get most of the performance.

            • willmore
            • 7 years ago

            Calling this 256-bit, etc. is a bit of a misnomer. At no point are these though of as 256 bit values, but rather they’re just bundles of smaller values. 4 64 bit values, 16 32 bit values, etc.

            The size of an architecture is determined by the largest single valued data size or pointer type that it can handle.

            • Arag0n
            • 7 years ago

            Didn’t I say that? Still, usually the memory bandwidth is related to the size of the architecture. A 64 bit processor will pack data every 64 bits in memory so memory controllers will be designed to be 64 to 128 bit wide, and memory bandwidth can starve a wide SIMD engine. SIMD programming is one of the most challenging ones besides OpenGL and DirectX3D… you are almost programming in assembler and sometimes you actually need to work on assembler. Once game engines start to use the new instruction sets given by AVX and FM3/4, the new cpu’s as bulldozer, ivy bridge, trinity and pile drive will shine. You can expect 20 to 40% performance gains.

            • just brew it!
            • 7 years ago

            [quote<]Once game engines start to use the new instruction sets given by AVX and FM3/4, the new cpu's as bulldozer, ivy bridge, trinity and pile drive will shine. You can expect 20 to 40% performance gains.[/quote<] Not sure I agree with this. A lot of the stuff that AVX and FM3/4 are good for in games is already being pushed down into the GPU. What's really needed for the newer multi-core architectures to shine is heavier use of multi-threading in the game engines. And this is a tougher nut to crack than just taking advantage of new SIMD instructions.

            • willmore
            • 7 years ago

            I think you’re both right. Arag0n, I was just trying to clarify your ‘depends on the point of view’ comment, not refute it. I +1’ed you!

            Memory bandwidth needs of a processor depend on the kind of workload it handles, so that’s a different kind of issue. Yes, higher performance processors are going to need more memory BW, but HPC workloads are going to need more memory BW than office apps and web browsing.

            I agree that games will benefit from better using the new instructions sets available. Heck, most games don’t even take advantage of the biggest ISA benefit that’s been available for years–x86-64! Games are still 32 bit. That means they have fewer registers available for integer code and it’s been shown time an time again that almost all code benefits greatly by the move to x86-64 even if they only use 32 bit integers, simply because they can use more of them before having to swap data to L1. just brew it! is also right in that games need to take better advantage of multiple cores. It’s pretty sad to see how few games take advantage of one or two cores. I wonder how many of the ones that can use two cores do so by letting the graphics driver use that core for its work. Other than SC2, what games use 4 or more cores well?

        • David_Morgan
        • 7 years ago

        Fun bit of trivia:

        While other internal portions of the chip were indeed 64-bit, the original K8 Athlon64 chips only used 40-bits for physical memory addressing, limiting users to a measly maximum of 1TB of addressable (physical) memory.

        Starting with K10 (Phenom) this address space was increased to 48-bits allowing for a scant 256TB of theoretical physical memory.

        *Perspective… the newly announced 18,000+ node Titan supercomputer has a combined total of around 710TB of physical memory.

        Sounds like it’s time to unlock all 64 bits of that glorious address space, AMD. Daddy needs 16 Exabytes of RAM in his new Excavator rig.

          • sjl
          • 7 years ago

          Further bit of funky trivia: AMD designed the CPU so that only the least significant 40 (or 48) bits were used to address memory. The remaining, more significant bits had to be set to match bit 39 (or 47). This effectively splits the memory into two halves: the top half, used by the OS, and the lower half, used by applications. It also means that any future CPUs will not be constrained by ad-hoc use of address bits to mark privilege levels or other such kludge, as happened in the days of the Motorola 68000.

          A nice piece of forethought there by AMD’s chip architects.

        • eitje
        • 7 years ago

        nah, I wasn’t serious. 🙂

    • spigzone
    • 7 years ago

    “Read argued that together, “AMD and ARM can change the industry landscape” because they “share a common vision about the industry” as well as a vision on “how to disrupt it together.”

    so … kick Intel in the balls and steal it’s candy?

    • spigzone
    • 7 years ago

    Dang, won’t be able to wipe the smile off JHH’s face with a brickbat.

    How much does he love this news?

      • Alexko
      • 7 years ago

      Why would he be happy to get more competition?

        • spigzone
        • 7 years ago

        Oh yeah, forgot the sarcasm thingie. Of course this is his worst nightmare, AMD and ARM teaming up to storm the markets he was counting on to save his a$$.

        All he needs now is is an announcement Amazon is buying TI’s Omap division to totally make his week.

          • Alexko
          • 7 years ago

          Ah, sorry, I didn’t pick up on that. Yeah, NVIDIA’s position in the ARM market is somewhat precarious. But at least I don’t think AMD will put up any serious competition for phones & tablets until 2015, they do seem to be primarily focused on servers.

    • Game_boy
    • 7 years ago

    We know from the A6 that ARM’s interconnect/uncore is the weakest part. HT and Seamicro’s thing will help a lot. If this was a 2012 product. Which is isn’t.

    • Tristan
    • 7 years ago

    AMD is unable to make any competing CPU, x86 or ARM
    They do not have money to upgrade current chips. They will design new ARM from the air ?

      • just brew it!
      • 7 years ago

      No, they will graft a coherent HyperTransport-style interconnect onto an ARM design that they license (and possibly tweak slightly). At least, that’s how I am reading this announcement.

    • Vrock
    • 7 years ago

    Why?

      • khands
      • 7 years ago

      Expanding the amd64 instruction set to include ARM so we can have x86-x64-ARM applications and rule the world.

        • just brew it!
        • 7 years ago

        Umm… no.

          • khands
          • 7 years ago

          Learn 2 sarcasm.

      • tbone8ty
      • 7 years ago

      to make some money which Amd needs to do lol

      they cant survive on us fanboys buying FX chips every yr

      • internetsandman
      • 7 years ago

      One word gives you two thumbs down?

      In any case, AMD needs to expand to different markets in order to pull in more revenue and hopefully more profits. What I’m hoping for is that AMD can make enough profit from their ARM business ventures that they can invest back in both ARM and x86, because if this is the start of a slow step out of the market for x86 (even though thats not what is said, but one cant help but be worried) then what little competition Intel has will be gone. On the other hand, should their ventures prove profitable enough, they could drive an increase in competitiveness for AMD in both x86 and ARM markets, not to mention that simply being a chip provider with that kind of flexibility is sure to make them a more attractive partner and company

        • willmore
        • 7 years ago

        Because there may not be such a thing as a stupid question, there certainly are stupid ways to ask the question.

      • flip-mode
      • 7 years ago

      To try to sell something.

    • Anarchist
    • 7 years ago

    hmm, … AMD arm cpu with radeon core? sounds like what nvidia is doing already with tegra. Once again day late and penny short. AMD desperately need a CEO with some sense of direction.

      • Alexko
      • 7 years ago

      They’re not really doing the same thing. NVIDIA is mostly relying on their mobile GPU IP to make consumer products, AMD is mostly relying on their interconnect IP to make server products.

      As far as I can tell, NVIDIA’s mobile GPU IP isn’t particularly better than ARM’s, Qualcomm’s or Imagination’s, and consumer SoCs tend to have low margins; however AMD’s interconnect is very good (even Intel was quite fond of it, back when it was SeaMicro’s) and server SoCs tend to have relatively high margins.

      It remains to be seen whether AMD can actually pull it off, but the strategy seems sound to me.

      I doubt AMD’s graphics IP will prove to be of much use (if any) in data centers.

        • OU812
        • 7 years ago

        How easily you forgot Project Denver.

          • jensend
          • 7 years ago

          Well, nV seems to have forgotten ‘Project Denver’ too. Absolutely nothing new on that front announced in the last 20 months.

            • tviceman
            • 7 years ago

            It’s called Tegra 5, and it was announced it would be manufactured at 20nm nodes. So why else would they have anything to say about it right now when they don’t even have Tegra 4 out yet?

          • Alexko
          • 7 years ago

          Kind of, yeah. But we know so little about it, I’m not sure what to make of it.

          We know it’s a custom core, but we don’t know in what kind of products it will end up, except for some heterogeneous designs with Maxwell GPUs, but those would not be targeted at data centers. That is unless I’ve missed something, of course.

          Perhaps NVIDIA will reveal more when they launch Tegra 4, or shortly after.

      • sschaem
      • 7 years ago

      I dont think the ARM transition was thought of today.

      And this might be a big advantage for Seamicro, and there is no 64bit ARM ‘SoC’ today.

      AMD will actually be in with the first wave if they get their product out in early 2014.

      • willmore
      • 7 years ago

      Aren’t the graphics cores in Tegra pre-OpenCL level designs? Like GF7xxx in nature fixed pipes? If they’ve gained the ability to do OpenCL like AMD can do with all of their GPU blocks, then that would be a big plus in the mobile space. You’d think we’d have heard them brag about it by now.

    • chuckula
    • 7 years ago

    Called it: [url=https://techreport.com/news/23786/amd-to-clarify-ambidextrous-strategy-on-monday?post=680414<]here a couple of days ago.[/url<] They want 64-bit to be "in production" in 2014... that basically means they are going with standard ARM 64-bit cores and doing most of the work on the uncore. I hope these things do well, because they will be going up against 14nm Atom microservers (good news is that AMD can still resell the Atoms too).

    • juampa_valve_rde
    • 7 years ago

    Doesn’t sound a bold move after buying Seamicro, they should give an edge over other platforms and ARM looks good to deliver high density/low power servers. Also it shouldn’t be that expensive to develop being ARM based with synthesized tools on a standard TSMC/GloFo process.

    • ludi
    • 7 years ago

    “Well, so far we’ve pink slipped the car, Craigslisted the bicycle, and eBay’d grandma, and yet none of these lotto tickets is paying off. What else have we got?”

    “The Farm.”

    “Great! Put the deed on the table.”

    • Bensam123
    • 7 years ago

    Wow, way to take things up a notch AMD… This most definitely could be extremely beneficial to both companies. ARM ends up gaining a bigger piece of the pie and AMD will inadvertently get help from ARM improving the efficiency of their chips, which is AMDs weak spot right now. None of that even takes into account what the future holds as far as what hybrids would turn into.

    I definitely could see this fusion almost being necessary to fight Intel now on multiple fronts. It could yield some very interesting results too.

      • just brew it!
      • 7 years ago

      Yes, the idea has potential. But that doesn’t automatically mean it will succeed. I suppose when you don’t have much left to lose you need to take risks.

        • Bensam123
        • 7 years ago

        I’m not really seeing the risks associated with this… They could make more BD flops with ARMs help, but I don’t think that’s all that different from what’s happening right now. ARM could take over the x86 market and push AMD under (extremely unlikely)… AMD may take over the ARM market and push ARM under (also quite unlikely)…

        Perhaps both companies will simply merge in the future (seems more likely then all of the above).

          • chuckula
          • 7 years ago

          [quote<]I'm not really seeing the risks associated with this... [/quote<] 1. AMD has limited resources and while this is not as intensive as designing Bulldozer (because ARM does the cores) you still have to have a bunch of people dedicated to building the system interconnects and doing a whole bunch of validation on these parts... people who are not, for example, doing validation on using Jaguar cores in these systems. 1.5 The negative flip side to the upside of ARM designing your cores in point 1: ARM is also designing everyone else's cores too. AMD has no differentiation at the core level, has no time to market advantage over the competition, and is going to have to contend with big players who have been developing ARM systems for years. 2. AMD might seem "small" in the x86 world compared to Intel, but they will be positively miniscule in the ARM world. The good news is that Seamicro gives them an outlet where they can push ARM chips without having to convince a big OEM to adopt their chips. That's why there is hope here, but by no means a guarantee of success.

            • willmore
            • 7 years ago

            [quote<].5 The negative flip side to the upside of ARM designing your cores in point 1: ARM is also designing everyone else's cores too. AMD has no differentiation at the core level, has no time to market advantage over the competition, and is going to have to contend with big players who have been developing ARM systems for years.[/quote<] This could be a positive becaue the core developed by ARM will be for one of (one presumes) GF's processes. So, GF would be able to offer an optimized core to more customers. The more profitable (and competetive in process) GF is, the more AMD can offer their customers. I hate to use the word 'synergy', but drink!

            • chuckula
            • 7 years ago

            Sorry, but GloFo != AMD. Their relationship has gone downhill rapidly in the last 2 years and there is very little loyalty left between the two of them. Will GF make a bunch of ARM chips? Sure, but AMD will be just another customer in line along with anyone else who wants to use GloFo as a foundry.

            Speaking of GloFo… where are all those 28nm chips they were promising? Say what you want about TSMC, but at least there are products out on the market that use their 28nm process.

            • willmore
            • 7 years ago

            So, you think AMD is going to move over to TSMC instead of using GF?

            • chuckula
            • 7 years ago

            Completely possible. Remember that every Radeon is fabbed by TSMC so there’s no reason TSMC couldn’t fab ARM server chips for AMD (the ARM chips will likely be much smaller, although they will have to wait for the 20 nm node).

            That’s not to say that AMD couldn’t get chips from GloFo, but instead that AMD won’t get special treatment from GloFo if there are other customers who want to pay for those chips.

            • NeelyCam
            • 7 years ago

            Also, Zacates have been fabbed by TSMC. AMD is familiar with the quirks of both fabs – there is no fundamental reason for them not to choose the fab with the best performance or best price/performance just because it’s the “wrong fab”

          • just brew it!
          • 7 years ago

          If this doesn’t pan out, they just wasted a third of a billion dollars acquiring Seamicro. Sounds pretty risky to me.

          I don’t think there would be much of a point to an AMD-ARM merger. ARM is essentially an IP licensing company; it doesn’t seem like a good fit.

            • Bensam123
            • 7 years ago

            They can use the Seamicro interconnect on anything, not just ARM chips… They also have the option of selling it off… They could sell it to Intel if they go under… There is really a lot of fluidity to that acquisition.

            ARM is as much a IP licensing company as AMD is.

            • just brew it!
            • 7 years ago

            AMD may have spun off their fabs, but they still have the overhead of contracting with GloFo (and others) to produce the actual chips. They are still in the business of selling physical silicon, and need to balance production costs against what they can charge for each chip.

            ARM Holdings doesn’t need to worry about the manufacturing side since they don’t sell any physical chips. Their revenue is based solely on licensing fees.

            It is a completely different business model.

    • MadManOriginal
    • 7 years ago

    I bet part of this is an attempt to do the same thing Nv wants to, that is have just sufficient CPUs alongside powerful GPGPUs. Could be a good combination when the workload is right.

      • willmore
      • 7 years ago

      Oh, hell yes! I’ll take a few 64 bit ARM cores with at least A15 core level performance and E-350 level graphics (but based on GCN) in any tablet/netbook/convertable over anything out there now. Of course, such a chip is a good year away–if it happens–and everyone else isn’t holding still.

    • just brew it!
    • 7 years ago

    Sadly, this feels like a bit of a desperation move to me. I’m feeling an almost uncontrollable urge to play [url=http://en.wikipedia.org/wiki/Buzzword_bingo<]Buzzword Bingo[/url<] with the press release. I hope it works out, but IMO it is risky.

      • A_Pickle
      • 7 years ago

      I don’t think they have many moves left, though. And I don’t think this is a stupid move — x86 has been selling less and less, and ARM has been selling more and more. x86 isn’t dead, by any means (nor will it ever be, unless ARM can actually compete with it in performance). But, if AMD can designed a good ARM CPU, to maybe capture some of that profitability, I think that’s probably a good move.

        • MadManOriginal
        • 7 years ago

        The ARM that has been selling more isn’t this kind of ARM though. Maybe in the future, but it’s still a risky move if they are betting on this heavily.

        • just brew it!
        • 7 years ago

        Per Damage (comment #2), I doubt they are designing their own ARM core. They’re taking an ARM core designed by someone else, maybe tweaking it a little, and grafting interconnect logic onto it to make it play nice with Seamicro’s interconnect. They’ve certainly got experience doing that sort of thing (HyperTransport and their past work with Cray), so they’ve got a fighting chance; but IMO it’s still a pretty big gamble.

          • designerfx
          • 7 years ago

          you know, they hinted that they were going to look into arm like 2 to 3 years ago. Them announcing it now seems to confirm they were pleased with the results, not “oh look! a new foray!”

          • dpaus
          • 7 years ago

          [quote<]maybe tweaking it a little[/quote<] Look what AMD has done with power management on their x86 CPUs and on their GPUs. I'm sure we'll see them bring a lot of that expertise and experience to ARM chips to create an extremely power-efficient server chip.

      • ludi
      • 7 years ago

      Take out the conjunctions and a few legitimate nouns, and it’s nothing but Bingo from start to finish.

    • destroy.all.monsters
    • 7 years ago

    The only question I have is whether the x86 Opteron is gone as a result. Having two processors, or in this case families, using the same name often hurts the overall brand.

    Just saw, in the Anandtech article, that they’re continuing x86 in the server space.

      • Airmantharp
      • 7 years ago

      Totally different markets… and they just went through the trouble of designing a server-oriented CPU!

    • jjj
    • 7 years ago

    There was a live stream of the event and the replay should be available in a few hours on AMD’s investor site.
    It also looks like ARM will detail tomorrow it’s first x64 core due in 2014 so more interesting news to come.

    • bjm
    • 7 years ago

    AMD will be as ambidextrous as Homer with a dumbbell.

      • dpaus
      • 7 years ago

      [quote<]Homer with a dumbbell[/quote<] I thought Homer [i<][b<]was[/i<][/b<] a dumbbell?

        • bjm
        • 7 years ago

        [url<]http://tinyurl.com/czl5apj[/url<] Ah, that episode was a classic...

    • Hattig
    • 7 years ago

    Well that’s interesting.

    I don’t know if AMD is really going “screw this x86 lark, Intel is whipping us and we can struggle to get by for the next ten years, so why not just try and push 64-bit ARM instead and maybe we will win big if the market does shift across to ARM in that timescale?” …

    AMD wouldn’t have made this decision without some knowledge of long-term plans of key partners such as Microsoft. Therefore we can assume that Microsoft’s Windows RT is just putting a toe into the water – the future will bring more.

    Linux is already quite ARM-ified, and each release of GCC seems to improve performance significantly to boot as the ARM compiler backend is optimised.

    But AMD is moving into a market that other players are already playing in. They might be targeting servers because of this, but there are other players doing ARM for servers. Maybe AMD’s Opteron brand will be enough to swing things.

      • MadManOriginal
      • 7 years ago

      It’s certainly a big gamble if they are truly going to give up on x86. They seem to have a hard time competing in performance/watt specifically, so maybe that’s what they intend for this, and will still have some absolute performance from x86.

        • Damage
        • 7 years ago

        Who said they’re giving up on x86. Everything in the press release talks about having both offerings going forward, and AMD keeps talking about the “ambidextrous” nature of its ISA strategy.

          • MadManOriginal
          • 7 years ago

          I was replying to this [quote<]I don't know if AMD is really going "screw this x86 lark, Intel is whipping us and we can struggle to get by for the next ten years, so why not just try and push 64-bit ARM instead and maybe we will win big if the market does shift across to ARM in that timescale?" [/quote<] That's why I used the reply feature of the comments :p

        • ronch
        • 7 years ago

        They may have considered giving up on x86 and letting Intel play alone in their sandbox, but it’s been their bread and butter for the longest time and giving it up is a little reckless even if they’ve almost always had trouble competing with Intel.

      • StashTheVampede
      • 7 years ago

      Microsoft, moving into the ARM world (with RT) is actually a pretty large step for not only Microsoft, but the industry as a whole.

      When Microsoft did Windows 95, they were one the leaders of the industry and got a whole lot of hardware changes made for many years to come. Microsoft was a large driver of Plug and Play (it was definitely Pray from everyone that was bleeding edge), PCI, AGP and getting a myriad of items onto the motherboard for standardized support (IDE controllers, USB, serial, parallel, etc). This was all a rough start, but it’s the base of how well much of our hardware interacts and plugs in, today.

      With RT, Microsoft is now delivering a spec of how RT will interact with all of the sub systems (boot, I/O, etc). This will not only bring more makers into the ARM market (with or without RT as the design goal), but it will also help linux-based makers because they can clone/copy a lot of what Microsoft does with the same low level functions.

      AMDs move into this space is basically a no-brainer: they are hedging their bets on being able to support multiple architectures on as much similar silicon as possible. There could be a point in time when you’re buying an “Operton” that has both x86-64 and ARM on the same physical socket.

        • chuckula
        • 7 years ago

        Microsoft has actually supported ARM since the 1990’s (Windows CE). The big change is that it is bringing back architectural flexibility to the NT-kernel. The big problem is that while the kernel can be flexible and some apps that are coded the right way can be recompiled, a whole bunch of stuff does not port to ARM and will not have acceptable performance if it gets run through a binary translator.

          • willmore
          • 7 years ago

          Yep, the original NT kernel was developed on MIPS machies to keep any x86 specific cruft from getting into it. At one time NT supported: MIPS, ARM, SPARC, PowerPC, Alpha, and x86.

          Unless they’ve changed their coding standards in recent years, the core of their OS should still be pertty architecture neutral. That’s not to say going back to ARM is just a recompile away, but it’s not a rewrite–they already did that with NT.

            • Scrotos
            • 7 years ago

            I think they really killed the HAL (hardware abstraction layer) with post-2K stuff to get better video performance. When they rewote(?) stuff for Vista/2008 I don’t know that it was written for portability in mind since all they were supporting was x86 and IA64 at that point. Then again they did come up with WinRT and Windows Mobile 8(?) is supposed to be based on the NT kernel so maybe they’ve kept it somewhat architecture neutral over the years.

            Would be interesting if that were the case.

            • willmore
            • 7 years ago

            They considered hardware independence very valuable back then. But, then again, they go through so many management changes, that’s no indication of what they’re like today–or were sometime between then and now. I guess we’ll just see.

      • Voldenuit
      • 7 years ago

      [quote<]But AMD is moving into a market that other players are already playing in. [/quote<] AMD has something the other players don't - a cross-licensing deal with intel. Now I'm not a lawyer, but what's to stop AMD from incorporating patented technology that it shares with intel into ARM CPUs? Is the language of the terms specific enough for intel to stop them? I'm just thinking that out of the thousands of computing patents intel and AMD have, there are some that ARM would just love to have access to.

        • just brew it!
        • 7 years ago

        Seems to me ARM + HyperTransport or an ARM-based Fusion chip (for GPU compute) would make more sense.

          • Airmantharp
          • 7 years ago

          Looking at the Cray big iron using Bulldozer CPUs, that makes more sense- AMD has the IP and experience to integrate an ARM CPU, especially now that they’re going OoO, into a GPU to create a true general-purpose computing unit. If they’re willing to throw their fab into it, they can build more efficient GPC modules for big iron than anyone else in the arena. And that’s actually pretty cool!

      • Flatland_Spider
      • 7 years ago

      [quote<]AMD is moving into a market that other players are already playing in[/quote<] Not really. ARM as a mobile chip is mature, but ARM as a server/desktop chip is just starting. Marvell and Samsung, I think the second player is Samsung, are the only ones that are really moving out of the mobile space, and they don't have the brand name, or server experience, AMD does.

        • TurtlePerson2
        • 7 years ago

        I was always under the impression that brand name wasn’t as important in the server space. I figured that people actually did research before dropping that much money on a system, but I could be wrong.

          • Airmantharp
          • 7 years ago

          Brand name is more important when you’re spending that much money…

          You need something that’s going to last, and if not something that will be supported. Persistence is the best forecast, after all.

          • willmore
          • 7 years ago

          It’s no necessarily brand name, but a proven track record of working on big problems. AMD does have chips in HPC and server applications. It’s not a stretch for a customer to assume that they’ll bring that knowledge to the problem of designing good ARM based system in the same space.

    • BobbinThreadbare
    • 7 years ago

    Well this is the end of competition in high end computing.

      • sjl
      • 7 years ago

      Yes and no.

      It does seem like AMD is giving up on the x86 space, or at least, acknowledging that they aren’t really competitive. But Intel does have competition in the high performance space: SPARC isn’t dead yet (although I think it probably is on life support; it remains to be seen what Oracle does with it over the next five or ten years.) Also POWER; IBM isn’t about to move away from that space in a hurry, and they do get good cross-pollination with their mainframe architectures as well. (to be fair, POWER is more of a competitor in terms of raw power than SPARC has been for some time; if you can scale outwards to multiple cores, though, there isn’t anything hugely wrong with SPARC. Except maybe Solaris.)

      If your focus is on Windows, then yes, there isn’t much out there beyond the Intel space; but if you broaden your gaze to less consumer-oriented systems, there is still some choice.

      Damn shame that Alpha (and Digital) kicked the bucket, though.

    • sweatshopking
    • 7 years ago

    yay? these sound like they’re custom chips. any confirmation?

      • Damage
      • 7 years ago

      Sounding like no, these will be ARM cores that AMD integrates, not an AMD-designed CPU core. Waiting for confirmation.

        • OneArmedScissor
        • 7 years ago

        That would make it a custom chip. :p

        The possibilities of what they may do with entire chips are much more interesting to consider than esoteric core tweaks.

          • willmore
          • 7 years ago

          Custom when dealing with processor IP has a meaning that seems to be a bit different than what you expect. There are three types of ARM cores:
          1) hard cores–which are designed and layed out by ARM and the silicon fab and the customer just pastes into their design
          2) soft cores–which are hardware design language version of the processor. These versions can be ‘compiled’ for any process the customer choses to use.
          3) custom cores–these are designed from the ground up by the customer using an architecture license from ARM. Basically, they can implement the ARM archetecture any way they want–even add bits to it. Apple, Qualcom, and Marvel do this.

          All ARM chips are custom chips. Ever since the ARM2 days, they’ve always been purpose built. So, the ‘custom’ in the ARM context is alwasy meant to refer to the cores on the chip, and not the SoC itself.

          What we have yet to see is if these cores will be hard or soft. I would sure hope they’re going to be hard or custom.

Pin It on Pinterest

Share This