AMD’s new K12 ARM project. No, its not Kindergarten!

 AMD announced last year that they were using ARM Cortex A57 as the basis for the Seattle SOC, some of us noted that AMD is the only vendor building a server SOC on the standard ARM Cores.  APMC, CAVM, BRCM, and others are heading down the custom core road, spending a lot more money to design their own chips with the hopes of out-perforing their competitors and building SOCs that have more differentiation than the  I/O and other “non-core” blocks.  A standard core is a great way to get into the market quickly, leveraging the common cores available from ARM Holdings.   But it is hard to differentiate on anything other than price and I/O if everyone can build a similar SOC for a few 10’s of millions of dollars.

As I consulted with industry players and investors,  I would often point out that AMD isn’t stupid;  they get it,  and are probably quietly working on their own ARM Cores,  leveraging the expertise and existing designs for GPUs, Memory, I/O, and Cores into a superior ARM server SOC part.  Well, today they stopped being quiet,  announcing their architectural license and intent to build a custom SOC 64-bit ARM SOC named “K12” in 2016.  Jim Keller, famous for his earlier chips at AMD and Apple,  is leading this effort.  Jim pointed out that they will be applying the DNA from their Bulldozer high-frequency part, so this is not another Wimpy ARM core, but is meant to compete with Brawny Xeon’s and will be positioned above their Cortex A57 parts.

AMD also announced Project Skybridge,  their plans to deliver a Seattle (A57-based) SOC that is pin-compatible with a future x86 processor on 20nm in 2015.  . The first generation of Skybridge products (ARM and x86) will focus on clients and embedded markets but AMD indicated that Skybridge is a framework for future designs.  Does this mean that AMD has NOT abandoned X86 servers?  While they did not announce a next generation X86 server part,  it is clearly a part of their ambidextrous  strategy. Pin-compatibility is cool because it allows AMD’s customers to reduce design costs to support their ambidextrous processor strategy.  They can amortize design costs across more volume and and lower costs and speed time to market.

So, AMD will do Standard ARM, Custom ARM, and x86 for servers, clients, and embedded.   Thats a lot of work, and a lot of risk.  But there may be enough synergies between these projects that they could pull it off and, once again, surprise the industry as they did in their 1st Opteron dual core chip, which quickly stole 25% of the market from Intel.  Jim Keller said that much of the new X86 and ARM SOCs are the same,   so they are already leveraging a lot of IP and engineering.

Who said processors were becoming boring!???


  1. We are looking to spin up android remote desktop as a service, which server hardware is available now that can perform this function?

    • Karl Freund says:

      At this time, the only thing you could have bought is a Calxeda box, and since Android is 32-bit that would have been a good fit. The 64-bit products are not generally available. Try contacting AMD or Applied Micro, and see if you can get a development system.

    • Would it not be simpler to implement this by running Android instances under qemu system mode on an x86 cluster, or is that still not completely functional?

      • Karl Freund says:

        Great question. I know that some very large companies (Telco and Cable) were looking at native Android server-based services, and I think it was primarily driven by security with Trustzones. But I may have the wrong.

  2. Sharad Mehrotra says:

    this is very exciting stuff; it is entirely possible to build very powerful SoCs with “standard ARM-developed cores”, particularly if one can leverage ARM’s big.Little technology creatively, and focus on the “uncore”. The uncore recipe is very straightforward – lots of PCI-e Gen3/Gen4 lanes and DDR3/4 channels, plus a good and deep cache hierarchy. Nothing that AMD doesn’t understand deeply.

    • Karl Freund says:

      Exactly, Sharad. Thats why I really like what they are doing: they bring a lot of non-core IP to the table. Execution will be key, but now they have Papermaster and Jim Keller, two guys with great track records.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: