New AMD embedded roadmap shows 64-bit ARM Cortex-A57 chip

As Intel’s Developer Forum kicks into gear in San Francisco, AMD is making waves to ensure it doesn’t stay out of the headlines. The underdog has disclosed a new embedded roadmap that teases, among other things, an upcoming ARM Cortex-A57-based system-on-a-chip.

Code-named Hierofalcon, this SoC will apparently feature up to eight 64-bit ARM Cortex-A57 cores clocked as high as 2GHz. It will have dual 64-bit memory channels with DDR3 and DDR4 compatibility as well as ECC support. Also included will be 10-gigabit Ethernet, PCI Express 3, ARM’s TrustZone security technology, and a "dedicated cryptographic security co-processor."

AMD expects Hierofalcon to sample in the second quarter of next year and to hit production in the latter half of 2014. The company touts the chip as the "first 64-bit ARM-based platform from AMD targeting embedded data center applications, communications infrastructure and industrial solutions." You probably won’t see this puppy in smartphones or tablets, in other words, but it is nonetheless one of the first fruits of AMD’s partnership with ARM.

The new AMD roadmap also teases three other products—two x86 APUs, Bald Eagle and Steppe Eagle, and a discrete embedded GPU, Adelaar—which are all due in the first half of next year.

Bald Eagle is an embedded x86 processor for digital signage and "embedded digital gaming" applications. It will feature up to four cores based on AMD’s next-gen Steamroller architecture, and it will be available with and without integrated Radeon HD 9000-series graphics. AMD claims to have implemented "new power management features," including a configurable thermal envelope, in the chip. (The roadmap says Bald Eagle’s TDPs will range from 17W to 35W.)

Instead of Steamroller cores, Steppe Eagle will include up to four cores based on an "enhanced" version of the low-power Jaguar architecture. (Jaguar, as you might recall, premiered in Kabini a few months ago.) Steppe Eagle will also have "new features for increased CPU and GPU frequency." AMD promises performance-per-watt improvements at lower TDPs and a peak speed "above 2GHz." Steppe Eagle will be compatible with boards and software designed for the current, Jaguar-based G-Series embedded APUs.

Finally, Adelaar will deliver discrete graphics in a choice of MCM, MXM, and PCI Express form factors. It will feature a Graphics Core Next GPU accompanied by up to 2GB of "pre-qualified and integrated" GDDR5 graphics memory with peak bandwidth of 72GB/s. AMD plans to continue offering Adelaar for seven years.

Comments closed
    • dpaus
    • 6 years ago

    [quote<]Also included will be.... a "dedicated cryptographic security co-processor."[/quote<] Marketing name: "The LongARM of the NSA"

    • The Dark One
    • 6 years ago

    [quote<]Code-named Hierofalcon, this SoC will apparently feature up to eight 64-bit ARM Cortex-A57 cores clocked as high as 2GHz. It will have dual 64-bit memory channels with DDR3 and DDR4 compatibility as well as ECC support. Also included will be 10-gigabit Ethernet, PCI Express 3, ARM's TrustZone security technology,[b<] and a "dedicated cryptographic security co-processor[/b<]."[/quote<] Given that Intel's hardware-based random number generator is now suspected of being tampered with, will anyone with serious crypto needs want to trust a black box implementation like this?

    • Bensam123
    • 6 years ago

    So what is the GCN card supposed to be used for (why the long lifecycle too?)? Directcompute? OpenCL? Could something like this be used as a physics processor?

      • Bensam123
      • 6 years ago

      Anyone?

        • chuckula
        • 6 years ago

        Buehler?

        It’s used for things that require embedded graphics and, for whatever reason, are not amenable to using an IGP. The card looks like it will have somewhat greater performance than an IGP, at least for the next couple of years, and it probably is setup to drive a big number of displays if needed.

        Remember that companies like Matrox are still in business even though their products are not directed to gamers. These cards will likely go into similar niches.

          • theonespork
          • 6 years ago

          Something-doo economics.

    • OneArmedScissor
    • 6 years ago

    Wow. Bulldozer was a certified disaster, but in retrospect, now even Bobcat / Jaguar are a disappointment. There still aren’t any Jaguar tablets, and they’re already sidelining it for something that Windows 8 can’t run on.

    Their one advantage has been ceded directly to Intel, who will now walk all over everyone with $200-300 quad-core Atom tablets.

    All of that work AMD did developing “next generation” architectures was for nothing. They could have perpetually shrunk Phenoms and Athlons and had a more appropriate product all along.

      • theonespork
      • 6 years ago

      I wish I could argue this.

      Perhaps some chip designer or person of knowledge in chip design can explain why this would not have worked, because it feels right. They could simply have improved and added and embedded and shrunk and I feel like things might have worked out for the better? I am truly curious if anyone has the time and inclination?

        • maxxcool
        • 6 years ago

        All you need to do is look an Intel Banias… and you have the answer:

        At the end of the p4 prescott craze, They took a pentium-pro 200 from 1995 (what 8 years out of production?).. and made ‘Intel Core’ cpu’s out of it. Every single Intel cpu today, including atom. Were/are sourced from a pentium-pro that the Israeli team ‘tweaked’ and added modules to.. that was it.

        There is no way this could not have been done with the Deneb cores and 10.5 stars cores. but they gambled based on the idea they could cheat a little and get better thread execution. it did not happen now they are locked in. they devoted pretty much everything they had to do this so it either carries them to the next gen, or they get bought out by via or samsung…

        the absolute WORST part.. **is they did this before** (well ATI did) with a custom execution pipe on the old old ati 7000 (r100) series cpus. it had 3 texture units when everyone one had 2…. naturally nobody coded their games to look for this extra tmu that everyone else did NOT have, and the gpu was a utter failure, this module bulldozer crap.. is a repeat on a billion dollar scale.

        [url<]http://en.wikipedia.org/wiki/Radeon_R100[/url<] lesson unlearned..

          • theonespork
          • 6 years ago

          Thanks man. I knew about the CORE architecture, not the ATI story. I disregarded the CORE story b/c for some reason I guess I thought there was some industry reason it would not work again. The more things change…

            • maxxcool
            • 6 years ago

            You know right?? … circles and more circles 😉

      • tcubed
      • 6 years ago

      well the story goes like this:

      AMD R&D camp came up with a crazy idea (they were in the athlon era – 2002-2003) they said… look gpu’s have much better FP throughput then any x86 if we could leverage it we would stomp all over intel… then Hektor Ruintz goes out and buys ATI at twice the market value and indebts AMD to it’s throat…

      intel had no chance on beating athlon let alone athlon 64 or athlon 64 x2 so what they did was take the mobile chip they had and turn it into the core series.

      From a theoretical point of view AMDs R&D was right but merging the 2 chip techs cost AMD almost everything. The ATI buy ment R&D funds slashing so they now had all of ati’s pattents and man power but not the raw brains and money to munch it into something useful.

      So… they tried 1 time to merge x86 cores – phenom – a mild disaster … phenom 2 & 3 were a bit more successful but they fell short of any intel offerings…

      Then they went even more out into the unknown merging the FP unit of 2 cores under 1 decoder… stupid move, the decoder is slow the FP is not well managed – bulldozer – a huge disappointment – for amd almost the last nail in the coffin.

      AMD then realized the path it went on is going to the future but without better R&D and better brains they can’t pull it… Intel did what they know how best, they just packed more and more into the same package improved on process and ta-da better chips. Their chips got shared memory for the cores 2 generations after amd they don’t have true integration of the gpu – it’s just in the same package but different silicon… They realized they are getting their asses handed to them(on gpu side) so they put 128 dram on top of that…

      So from a theoretical point of view AMD is some 10 years ahead of intel – from a practical point of view intel chose the pragmatic approach amd chose a “creative” approach… intel is thriving while amd was until this quarter always in peril of being declared bankrupt…

      So what does the future hold?

      Well AMD brought a lot o brains back to it’s camp from nvidia, apple and others… They are now focusing more on process and operation then on innovating and pioneering thanks in large part to the very pragmatical CEO they now have…

      It will take a while for the new R&D and the software department to make a dent but I hope the steamroller & excavator cores will bring better IPC and I have a small hope they have learned the lessons of the phenom & the bulldozer and will finally start developing all these new and shiny ideas slowly and wait for them to mature. I put a lot of faith in the return of the K7 leaders…

      And, no, the development was not for nothing they achieved huge progress their chips are far ahead of their time but totally immature and don’t have any type of software support – what they did was reckless, they played the pioneer card with a less then stellar R&D, starved of money and with a merger on their hands (not to mention all the illegal actions intel did and the PC drought).

      Why I say that they are 10 years ahead… well for once the i series is still based on the pentium chip… that’s a very old architecture… they could keep up because they have a stellar execution on silicon but the chip is primitive at best, they just added new instructionsets like they always did…

      Now amds cores, bulldozer especially, is low in ipc because it shares a decoder & a FP pipe (I hear steamroller corrects the decoder part at least) they patched this horrible mistake with the resonant clock patents they bought that allowed them to increase the frequency staying in the same TDP… this is a last minute fix rather then an actual fix… It will however come in handy later on. They played the resonant clock card in the last richlands & FX too that’s why you can have a 5 GHz chip on air… The problem however is far from solved…

      You need to understand that in order for the bulldozer to work they needed 2 things:
      1 – double frequency decoders
      2 – good FP management

      bulldozer however failed at both… In order for the two integer pipes to be propperly fed @4 GHz you would need a 8Ghz decoder (yeah I know right… what were they smoking??)… Even if you create the meanest badest decoder in the world you wouldn’t be able to properly saturate those integer units… Then the FP … well that’s something they kind of solved lately with windows 8.1 and the new scheduler.

      Ok there was another problem with the bulldozer … the / operation… this operational block wasn’t working in the first revisions so they made a soft / operation that took 20 clocks instead of 1 or something of the sort… they sorted this mess I think in R3 of the bulldozer (or in trinity – can’t remember exactly) – and they covered it up like skillful doggies instead of manning up and delaying the launch…

      So… The idea behind the entire fusion project is rather simple… have a bunch of integer cores (that 2 by 2 can form a FP) and move the heavy FP lifting (parallel processing) to the GPU FP array… well easier said then done… First you need to bring the gpu on the same silicon… then you need to build a unified buss for the two types of cores and a unified l2/l3 … then you need to merge the memory allocation between the two so you can move processing from one core to the other seamlessly… After this the next logical step is to unify decoders to save on space… having slightly bigger decoders but less redundant – again get it wrong and you end up starving the cores… too many and you end up with loosing the size advantage…

      Now we are very close to the end of the road here – this road however should have taken 4-5 years not 10 and should have not been on production products – they should have just improved on what they had and play the same game as intel.. they should have… but they did not … they should have payed only 2 bn on ati not 4 – they should have but they did not…

      Alas they are now back in black and have upcoming ARM server chips and I suspect they will also do some smartphone chips for the next super-phones, a successful marriage between A57 & GCN would probably be a killer chip but we shall see… if they can prevail where nvidia failed with tegra…

      So yeah they are a few generations ahead of intel -theoretically- but they executed miserably on them. Being back in black means more flexibility & more R&D money which means we should see some kick-ass chips in the coming years I expect first real showdown with intel to happen no sooner then 2015 maybe Q4 of 2014 – untill then they will just drag along maybe claim a victory or two in the cloud server area because of seamicro & their A57 ARM chip but I don’t expect anything spectacular untill late 2014 or more certainly 2015. And all that IF and only IF Rory Read manages to improve on execution as he did until now!

      The consoles win is a big big thing for AMD it might not seem much but it really is… having their chips in all consoles means they will get money and orders on the long run some 5 maybe even 8 years. That means money without R&D costs just production – this is a very smart move… Even if it’s not that much money it’s a steady money flow that relies on Microsoft’s & Sony’s PR & marketing… so they can just enjoy the ride and make a money on the side…

      The verizon cloud is another important thing – amds first major win in server space since … forever… seamicro division might be another long term contract bringer…

      With 2 long term strategies (and long-term contracts) AMD can secure money for R&D and get fusion to fruition… but we shall see…

    • chuckula
    • 6 years ago

    On a note unrelated to AMD’s ARM daliances, the embedded APUs look very good. Not every server can take advantage of an IGP, but for workloads that can use the IGP either for graphics or compute, then AMD has a pretty good advantage.

    In a highly competitive server market, AMD should be able to carve out some territory in areas where it can offer alternatives to the competition instead of trying to do kamikaze runs directly into the competition’s strongest areas. That’s probably why we see the major shift away from multi-socket and high-core count parts into microservers & small APU servers.

    • sschaem
    • 6 years ago

    Nice slides! Lots of Attractive Marketing Details

      • internetsandman
      • 6 years ago

      I thought you were just overenthusiastic with your caps lock key. Well played good sir

    • tfp
    • 6 years ago

    AMD 2010 called, it wants its ARM back.

    • Unknown-Error
    • 6 years ago

    AMD must be renamed AMR = [b<]A[/b<]lways [b<]M[/b<]aking [b<]R[/b<]oad-maps In AMD and its fanboi world, road-maps, nice powerpoint slides, fancy code names alone can smash Intel to a pulp. GO AMD! AMD Rocks! What a pathetic excuse for a tech company.

      • Wildchild
      • 6 years ago

      So, who are you a fanboi of then?

        • chuckula
        • 6 years ago

        Pffftt… isn’t it obvious? (MIPS of course)

          • Helmore
          • 6 years ago

          No no no. He used to be an AMD fanboi, but he got severely disillusioned by AMD’s lack of performance and is now taking it out at AMD :D.

            • Unknown-Error
            • 6 years ago

            Bravo Helmore! I must admit I haven’t been an AMD fanboi since 2006. Now its off-course MIPS the new king in town which will smash it into pieces by 2113 :p

          • nanoflower
          • 6 years ago

          I’ve been a fan of MIPS since back in the SGI days but I certainly have no illusions about the place the company holds in the market today.

            • stupido
            • 6 years ago

            Wasn’t MIPS instruction set license bought by the Chinese? or something in that line?
            I mean they wanted to make their own processor based on MIPS architecture…

        • NeelyCam
        • 6 years ago

        Yeah!!

        You [i<][b<]HAVE-TO-PICK-A-SIDE!!!!![/b<][/i<]

      • Auril4
      • 6 years ago

      In the same pathetic bin with American automobile companies.

        • theonespork
        • 6 years ago

        OOOOOOOOOKAAAAAAAAAY, you want to explain this one, or are you just tossing out a useless opinion? Last I looked, things were looking up for the American car industry. But then, that would be based on facts and analysis.

        In the last 3 years GM has posted 18 billion in profit. Ford has posted 31 billion. Chrysler has had 41 straight quarters of sales growth. All three U.S. headquartered companies have reported lower overall incentive spending per vehicle and increased transaction prices 2 years in a row. All three companies have increased market share over the same period. All three companies have increased their perceived and real quality scores according to most publications.

        Pathetic? Sure, if you are living in the past and are unable to allow your opinion to change as facts change and the world evolves. In that case…pathetic indeed.

        Oh, feel free to post real, hard facts to counter the above. My information comes from the following sources: Wall Street Journal, Washington Post, FREEP, Automotive News, L.A. Times, and Bloomberg. All freely available on the internet through this useful tool I discovered: Google.

      • theonespork
      • 6 years ago

      What, exactly, is the point of your post?

      It is news. It was reported. You may not like the news delivered, but your post is dripping with hatred for the company and for people who might like what they read. Why?

      You attack AMD fanbois, but they are not even mentioned in the post. You call the company pathetic for offering up a roadmap that does not cater to you.You provide no analysis as to why you attacked AMD fanbois, nor as to why you so seethingly hate the company.

      If you hate AMD, great. Whatev. I would bet most computer enthusiasts are less than thrilled with where AMD has ended up. We needed and wanted a thriving competitor to keep performance high and costs low. We have lost that. Everyone lost there. I am not sure why they should make me, you, or anyone hate the company?

      Maybe instead of insults and attacks on low hanging fruit you could try something with a little more forethought and analysis. Unless of course you were just trolling. If you were trolling, well done. Sad you spend your days that way, but bravo. Your accomplishment is noted. Yay.

    • ronch
    • 6 years ago

    Of these, Hierofalcon (so awkward to type, so it’s “HF” henceforth) and Bald Eagle (“BE”) are probably of the most interest to me, not because I care one bit for embedded apps but rather because HF is presumably the first fruit of AMD’s ARM efforts, and BE is also presumably Kaveri sold off as a product for embedded apps (Steppe Eagle is just Beema disguised as a chip for casino slot machines). It would be interestly to see whether HF takes off and makes a dent in the server market or not, and of course, everyone’s wondering whether Steamroller will only be sold off in APUs such as Kaveri and BE. There are two possibilities here: (1) SR is shaping up really nicely (there were rumors not too long ago that AMD is ‘taking its time’ with SR to make sure Hasbeen doesn’t smack it silly out the gate) that AMD feels confident enough to ditch their 8-core plans with it, or (2) AMD is giving up not just single-threaded performance, but multi-threaded performance as well. It’s like they’re screwing us for booing the FX series too much.

    PS – For the record, I love my FX-8350 but if someone gave me an FX-9590 I’d probably clock it down to 4.0GHz and run it as an FX-8350.

      • chuckula
      • 6 years ago

      King to Kaveri Server Part: I dub thee, [s<]SteppenWolf[/s<] uh... Steppe Eagle. King to ARM Server Part: I dub thee, Hier-of-Pidgeon.

    • drfish
    • 6 years ago

    *sigh* How freaking awesome would all this be if it had managed to hit the market last year?

      • chuckula
      • 6 years ago

      You could say the same thing about Intel’s GPUs… if Haswell’s GPU had been out in 2009, where would AMD be? The word “if” is a lot of fun, but it doesn’t rewrite reality.

        • drfish
        • 6 years ago

        Sure, but the difference in this case is that the “if” means a lot more to AMD than it would have to Intel.

      • ronch
      • 6 years ago

      Why settle for ‘last year’? If this thing hit the market in 1990, AMD would’ve beaten the he** out of its competitors and boosted its brand value to Mars, assuming they keep it up to the present.

        • chuckula
        • 6 years ago

        Pfft… 1990? That’s practically last week. If these things + a large modern industrial infrastructure had been on sale in 476, then Rome wouldn’t have fallen! Seriously AMD, why are you such a bunch of Visigoth fanboys!

      • tipoo
      • 6 years ago

      Sarcasm? Internet and all…A57 was nowhere near on market last year.

        • chuckula
        • 6 years ago

        It was on-powerpoint, which is good enough for some people.

    • chuckula
    • 6 years ago

    Hrmm…. 4 – 8 ARM cores on a SoC.. interestly…

    Late 2014… well, that’s not spectacular but honestly AMD will likely be one of the first vendors with server parts out since AMD knows a whole lot more about servers than the other ARM SoC vendors…

    28 nm…. not overly encouraging here especially when late 2014 was supposed to be on schedule for 20 nm….

    and… the piece de resistance…

    15 – 30 Watt TDP…. OOPS. Remember what that “primitive” Avoton is getting for TDP using 8 cores *this year*? What was that, 20 watts at the *high end*??

    [Edit: For those of you who think this is an anti-AMD rant, read it more carefully. When it comes to ARM in the server world, AMD is easily going to have the best solutions since AMD has worked with servers for a long time and has its “freedom fabric” interconnect. I’m really pointing out that the supposedly magical ARM architecture is rapidly running out of miracles when it starts to scale bigger than your smartphone. It doesn’t matter if AMD/IBM/Qualcomm or even Intel made the chip.]

      • maxxcool
      • 6 years ago

      eh, it’s server hardware. I’m sure most wont care about the TDP *if* it has the chops to perform well at it’s tasks and is cheap enough.

        • chuckula
        • 6 years ago

        Ahh the use of the word “if” again…. Anyway, servers (and microservers in particular) are *very* sensitive to power consumption when deployed en-masse.

        This isn’t just the seemingly measly difference between 20 watts and 30 watts.. this is a *50% higher power budget* multiplied by 1,000, 2,000, 3,000 …. blades in a large server deployment. That *does* add up, and I’m sure not expecting any miracles out of the A-57 performance numbers where it will blow away Avoton in every dimension to justify a *50%* power premium.

          • Waco
          • 6 years ago

          This. Anyone who says “server people don’t care about TDP” knows very little about large-scale deployments. 50% is HUGE.

      • Wildchild
      • 6 years ago

      Intel, a company with vastly higher quantities of disposable income is somehow able to make a chip that uses less power than little AMD.

      Gee…. who would have guessed…?

        • chuckula
        • 6 years ago

        Sorry, this isn’t just AMD now, this is AMD + ARM, and I have it on good authority that anything from ARM is absolute magic. x86 is obsolete trash remember?

        Cortex A57 is the coming apocalypse for Intel because it will have at least the performance of Haswell and use so little power that it will run off of ambient air currents!

        Oh wait… you mean physics apply to ARM too? I think that’s just pro-Intel fud there!

          • cegras
          • 6 years ago

          Hey look it’s chuckula and his ARM ventriloquist act in an AMD/ARM thread! So novel!

            • chuckula
            • 6 years ago

            Hey look it’s cegras who:
            1. Never posts any technical content to a discussion.
            2. Pretends to be neutral while really being an ARM fanboy.
            3. Will likely say that nobody has ever posted pro-ARM predictions anywhere on the Internet.
            4. Doesn’t like it when predictions that I’ve made in a consistent manner going back a very long time start to come true*.

            * Actually I was wrong: I was betting that Intel would hit a power-performance lead at the 14nm process node. Now, it looks like the crossover is occurring while Intel is still at 22nm. I’m willing to admit when I’m wrong.

            • cegras
            • 6 years ago

            1. All you do is parrot the article in question, I’ve never once seen you post anything that I couldn’t have gotten from reading the actual techreport news article.

            2. LOL You’re really working that strawman!

            3. For the nth time, not on techreport. I.e. take your stupid crusade to other forums

            4. Who cares about your predictions? Do making predictions about a company good at execution and a company who consistently fails at execution give any of your statements weight, or you clairvoyance? You might as well be saying “I predict the sun will come up between 6 and 8 am tomorrow.”

            • chuckula
            • 6 years ago

            So you are saying that ARM consistently fails at execution? Because my post was an accurate insight into the struggles that ARM is having at scaling out of smartphones you know… the fact that ARM doesn’t make its own parts and relies on AMD + GloFo/TSMC is secondary to my argument.

            But then again, you never read what I actually wrote before you used your hackneyed “strawman” argument to deflect from the real issues.

            • cegras
            • 6 years ago

            I’m saying AMD has disappointed for a long time. I don’t know about ARM, it seems that they’re doing fine in mobile but have had problems breaking beyond that. I don’t really have any emotional investment in their success or failure.

            [quote<]But then again, you never read what I actually wrote before you used your hackneyed "strawman" argument to deflect from the real issues.[/quote<] The problem is that you keep posting, on techreport, posts that essentially make you out to be some sort of martyr fighting against the ARM freedom force, when in fact that is not the case. It doesn't even matter what you write anymore, your tone and style of writing is boring and pretty annoying. It's gotten to the point where as soon as I see any news report concerning intel or arm I *know* you'll be in there doing your usual schlong. Multiple people here have informed you of this but you don't seem to care. Then you go on to brand anyone who says so, like me, a supposed "ARM fanboy", when in reality all we're interested in is seeing you stop being so annoying, or at least go be annoying somewhere else where you can have a fair fight with the "ARMy" (get it? army?). You promised to stop after "hasbeen". Apparently you've moved on to "broadowell" (or whatever you call it). Oh, and I don't think I have ever downthumbed you, or anyone here.

      • gc9
      • 6 years ago

      AMD announced the server part, code-named Seattle, a few months ago. Hierofalcon seems like similar silicon binned and packaged for embedded applications.
      [url<]https://techreport.com/news/24972/amd-announces-arm-based-seattle-chip-for-servers[/url<]

Pin It on Pinterest

Share This