WP_Term Object
(
    [term_id] => 13
    [name] => Arm
    [slug] => arm
    [term_group] => 0
    [term_taxonomy_id] => 13
    [taxonomy] => category
    [description] => 
    [parent] => 178
    [count] => 392
    [filter] => raw
    [cat_ID] => 13
    [category_count] => 392
    [category_description] => 
    [cat_name] => Arm
    [category_nicename] => arm
    [category_parent] => 178
)
            
Mobile Unleashed Banner SemiWiki
WP_Term Object
(
    [term_id] => 13
    [name] => Arm
    [slug] => arm
    [term_group] => 0
    [term_taxonomy_id] => 13
    [taxonomy] => category
    [description] => 
    [parent] => 178
    [count] => 392
    [filter] => raw
    [cat_ID] => 13
    [category_count] => 392
    [category_description] => 
    [cat_name] => Arm
    [category_nicename] => arm
    [category_parent] => 178
)

I’d give my right ARM to be ambidextrous

I’d give my right ARM to be ambidextrous
by Don Dingee on 05-09-2014 at 8:00 am

Baseball loves a good switch hitter – from Frisch to Mantle to Rose to Murray to Jones, they are a rare and valuable commodity. AMD is calling on ambidexterity for its processors in 2015 and beyond, this week tipping plans for 20nm “Project SkyBridge” parts in either ARM or X86 with a common footprint. What remains to be seen is where they bat in the lineup.

Upon first hearing of the announcement, I was afraid AMD might be revisiting Socket 7, where standard motherboards could accept either an Intel Pentium or an AMD K5. Cloning would be ill-advised, as AMD has demonstrated time and time again. Historically, they have pulled out to a 20% advantage at launch, then fallen behind the relentless tick-tock of Intel launches for two or three beats, then they would leapfrog to a marginal lead, only to fall behind again.

That vicious cycle kept repeating, with Intel mostly calling the shots, and a strategy shift was in order. It looks like leadership changes at AMD, including Rory Read as CEO and Mark Papermaster (of Apple iPhone fame) as CTO, may have finally broken out in a new direction. The first part of the shift is to capitalize on AMD strengths: using GPUs and lower power X86 cores, they are moving full force in the APU direction.

There is a lot of stuff on the fully unveiled AMD product roadmap (link downloads a PDF file), but my eye goes to one box on the mobility slide: “Mullins”, a 2.8W SDP part with two to four “Puma+” cores, a Radeon R Series graphics unit, and ARM Cortex-A5 core as the Platform Security Processor, all in a 28nm implementation.

SkyBridge is a potential category-buster, taking a swipe at the top end of Intel’s foundering mobile strategy – while leaving Intel to grapple with smartphones if they choose. The plan is to offer tablet and embedded customers a choice of either architecture, in what is promised to be a family of APUs and SoCs with pin compatibility between ARM and X86 parts.

On the X86 side of the plate, they didn’t announce much because they just announced “Mullins” the week before, but head of core development Jim Keller made an interesting comment indicating they are getting ideas from working with ARM cores that will appear in the next round. More on that shortly.

On the ARM side of the plate, the Opteron A-Series “Seattle” made its first appearance at this week’s announcement, sporting 64-bit ARM Cortex-A57 cores in a demonstration of Linux derived from Fedora. SkyBridge is not about servers or smartphones, and shrinking an A57 design to 20nm plus other power management steps has the potential to put a potent offering in the mix for 2015. For 2016, AMD is taking the ambitious step of an in-house designed ARMv8 compliant core called “K12”.

AMD may have the technology, but will anyone buy it? There are two fascinating details in all this. Perhaps most obvious is the addition of Android, for both ARM and X86, to the AMD mix for the first time. One of the major cycles of learning here is ambidextrous is good, but bi-endian isn’t all that great if you want to take Intel sockets away. The Linux and Android communities are finally converging on little-endian for ARM, cutting one of the major objections in moving to or from X86.

Ambidexterity is possible through a less obvious but very sensible approach: a complete, ground-up reengineering of the IP inside the SoC, presumably around an AMBA paradigm, that allows the processing and graphics cores to modularly plug-in. This is a plus not only for hardware design and the ability to spin and verify chips more quickly (did someone just yell “Remember Barcelona” from the balcony?), but creates a software framework that shortens the leap between ARM and X86 – much shorter than moving between other ARM providers and Intel.

This is definitely about on-base percentage for AMD. The ARM investment is obligatory given explosive growth and a power advantage, but the entire AMD machine and their current customer base is built around X86, and they cannot afford to suddenly stop moving forward with it. A shrinking PC base doesn’t necessarily mandate a shrinking X86 base, if more attention is paid to embedded and IoT platforms, plus playing out the game in 2-in-1 and tablet designs.

From a strategy perspective, this is one of the more sweeping makeovers I’ve seen, and a pretty brilliant set of moves. Of course, there is execution to worry about still – this is the bottom of the 1[SUP]st[/SUP] for a rebuilt AMD and their new idea, it’s a 9 inning game. I’m guessing the stakes have been raised internally, and AMD will no longer be content with being what baseball circles call a “AAAA player”, constantly on the shuttle between the majors and minors.

lang: en_US

Share this post via:

Comments

There are no comments yet.

You must register or log in to view/post comments.