Along with an update on their x86 plans, AMD has also presented an update on their ARM plans at financial analyst day today. The news there is a mixed blessing, depending on one’s point of view.

AMD’s “Seattle” SoC – the ARM Cortex-A57 powered Opteron A1100 – will finally be shipping in H2 of this year, after first sampling towards the end of last year. This unfortunately is almost a year behind AMD’s original schedule, though AMD seems moderately optimistic about catching up once they have their first ARM silicon out the door.

But the real focus of AMD’s comments on ARM for FAD involve K12, the AMD-developed ARM CPU core being designed alongside Zen. With AMD having opted to prioritize Zen development, K12 has been pushed back from 2016 to 2017, essentially taking the near-simultaneous launch of the two parts off the table. AMD for their part is attributing this change in schedule only to the decision to focus on Zen, however with Opteron A1100 delayed, it’s entirely possible this is also a knock-on effect that has pushed back the entire AMD ARM roadmap.

In any case, even with the delay AMD is still eagerly moving ahead with their ARM plans. A1100, even though it’s late, will be the pathfinder for AMD’s ARM efforts, serving as a platform to further develop the AMD ARM ecosystem, getting developers acquainted with the technology and getting software ready for it. K12 in turn will come in after that ecosystem has already seen some development, allowing AMD to get their new hardware out to the market and already have software support for it. I hesitate to say that this makes A1100 entirely a pathfinder product – until it’s available, it’s not clear how many customers might purchase it for production work – but clearly AMD’s big play is K12, not A1100.

And though K12 is delayed, AMD tells us that this hasn’t changed how it’s being developed, which is to say that it’s being done in concert with Zen. So virtually everything AMD gets right with Zen will be integrated into K12 as well, just on the basis of the ARM ISA instead of x86. AMD still believes that the time is right for ARM in the server space – along with a very obvious place in the semi-custom SoC space – and that K12 will be an interesting alternative to Zen in that regard.

POST A COMMENT

15 Comments

View All Comments

  • Samus - Wednesday, May 06, 2015 - link

    2017? Damn. Reply
  • MartinT - Wednesday, May 06, 2015 - link

    Maybe AMD should consider dropping a few projects, this one included, until they get their core businesses, x86 CPUs and GPUs, back in shape? Reply
  • eanazag - Wednesday, May 06, 2015 - link

    They can't afford to drop this project in investors eyes. AMD brings legitimacy to ARM on desktop and server. That being said, AMD's plain ARM chip being behind is sad and obviously ties up development efforts. They need to get the first one out the door and poach some Apple employees too.

    It makes more sense for AMD to push for x86. It is a tried and true market. People would buy AMD x86 CPUs for consumer and enterprise if the stuff wasn't 4 years old and too slow to justify even with pricing. I want to buy AMD, but I can't justify their stuff with any reasoning.
    Reply
  • spikebike - Wednesday, May 06, 2015 - link

    @MartinT AMD has a suite of problems. Seems mostly related to their board following the hype and not making clear engineering and business decisions. They ditched the previous CEO who was doing quite well and had turned AMD around and made them profitable. The board did this because they "must be in smartphones/tablets", even though they couldn't answer the CEOs question about how to make such an attempt profitable.

    So they ditched their FAB, and now much use FABs that follow the most profitable chips. Sadly that's not ideal for competing with Intel. So now AMDs in a position where they don't match performance or performance per watt. Intel's stopped innovating on performance of their CPUs and GPUs, all have only seen very small improvements. A 3+ GHz quad core sandybridge isn't that different than a 3+ GHz quad core broadwell. Nor has GPU performance increased much from the HD4600. For normal use cases the sandybridge, ivy bridgbe, haswell, and broadwell have only made small incremental improvements in performance (clock rate, ipc, or ipc per clock). Intel has been pushing hard on performance per watt though.

    So without being able to beat Intel AMD can't charge much for a CPU. FABs are optimized for things like smartphone CPUs which are not designed to compete with Intel's laptop/desktop chips.

    Sadly the APUs and GPUs don't seem to be doing much better, and AMD drivers don't seem nearly as good as Nviida's. The more heavily you use your GPU (professionally or for gaming) the more likely you are to end up with a Nvidia GPU. This is particulary sad because AMD has access to the same fabs that nvidia does, so they don't have a handicap of a few generations of process improvements.
    Reply
  • testbug00 - Thursday, May 07, 2015 - link

    Intel's performance increases slowing down are due to them only having a really advanced micro-architecture where all the "low and medium hanging fruit" has always been picked. Anything to increase performance from Intel is very expensive and challenging. Lowering power at the same performance now, that has more low/medium hanging fruit (at least, it did, I think Intel has gotten most of it by now) allowing for large improvements.

    As for drivers and Nvidia/AMD... That's FUD from 4 years ago on the consumer side. The drivers are more or less the same, either slightly favoring Nvidia or even.
    On the professional side Nvidia has better drivers, but, more importantly, they have a lot better industry support. Although, AMD is making progress.
    Reply
  • przemo_li - Monday, May 11, 2015 - link

    Actually, AMD APUs have better perf (gpu/games wise) then competing Intels.

    And Intels recent iGPUs are weak cause of too low heat room. Meaning they would do better (in notebooks where it matter) if only had better cooling. So OEMs choose baseline models.

    Fails happen.

    Though if one took Your words as "Intel stagnated a bit in APU space recently, AMD can outdo them in single gen still", then its correct. AMD can still put good chips that would fight with Intel offerings on performance terms.

    Now lets see if:
    a) they can put out better perf APUs
    OR
    b) DX12/Vulkan change PERCEIVED performance (meaning - cheaper AMD chips suddenly provide same FPS as expensive Intels, so why bother with blue?)
    OR
    c) Software momentum for HSA start to provide clear and tangible results in software people use everyday

    So AMD do have some attack angles. Lets see if something pan out this year!
    (I bet Vulkan, followed by HSA)
    Reply
  • TheinsanegamerN - Tuesday, May 17, 2016 - link

    Iris pro in broadwell beat out every AMD APU according to this very site. Reply
  • Dmcq - Tuesday, May 12, 2015 - link

    I think what AMD is doing is right. The ARM server market won't be worth much in 2016 but they need something there to develop the market and form a presence, it doesn't have to be anything too wonderful though. As to the future depending on x86 would be to continue a losing strategy. I believe the ARM architecture will deliver a worthwhile improvement in performance over x86 with the same development effort and so can give AMD an edge rather than always pricing down and trying to fill gaps. They already have experience in enterprise computers and customers who know them so have that advantage over other ARM chips. Also it provides them with IP for developing chips like for the games machines they've done. Reply
  • MrSpadge - Wednesday, May 06, 2015 - link

    Seeing the extremely quick and impressive progress ARM made moving from A57 to A72 I think AMD should simply put A72 cores into a A1100 successor in 2016. Afterwards they can still evaluate whether K12 still makes sense.. or whether ARM already passed them with whatever they have by then. Reply
  • beginner99 - Thursday, May 07, 2015 - link

    Exactly my thought. This Seattle thing is already late. By then the first A72 will probably already be available. And K12 is 2 years out. Must be a monster or else there is little hope. Reply

Log in

Don't have an account? Sign up now