I all-too-often get accused of writing positively about positions that I happen to hold (which is funny since, well, if I'm positive, I should own it, right?). But today I saw a particularly face-palming announcement from Advanced Micro Devices (AMD) that it won the "FireFall" Mobile Gaming Center design win with its SeaMicro dense serving products.
That should mean good things for AMD, right? Well...no. And here's why.
SeaMicro: Powered By Intel
What do you do if you're bleeding server market share and now have less than 4% of the X86 server CPU market? Well, the obvious answer is to expand into the new, untapped high growth areas of servers. Dense servers (microservers) are the new, untapped, greenfield opportunity that every chip designer with an ARM (ARMH) license seems to be getting giddy about. AMD, in what seemed like a good move at the time, purchased one of the leading microserver startups known as "SeaMicro." They have a neat interconnect fabric called "Freedom Fabric," and the idea is that this fabric combined with a bunch of low-power CPUs could lead to substantial TCO reduction for certain types of workloads that are not particularly single thread intensive.
The important part of all of this was that AMD could not only sell the box and the fabric, but it could sell a bunch of CPUs into these things to help market share. I mean, owning a "leader" in a growth area and selling CPUs into it should help reverse the fortunes of a struggling company, right?
Yeah. That's where this gets ugly. From the press release:
AMD's SeaMicro SM10000-XE server contains 256 CPU cores and the 1.28 terabits per-second Freedom™ fabric, which removes the constraints of traditional servers while delivering up to 50 percent reduction in power consumption. It is the industry's highest density, most energy-efficient x86 server.
256 CPU cores, eh? So, selling those Opterons, AMD? Nope. The SeaMicro SM10000-XE is powered by Intel (INTC) Xeon E3 processors:
Now, I can understand why SeaMicro would use Intel's chips: they're much more power efficient. Intel's CPU core is inherently a better performance/watt design for most workloads, and Intel's 22nm process affords it additional power consumption savings over AMD's 32nm "Piledriver" chips. In short, without Intel's chips, SeaMicro wouldn't sell anything.
It Only Gets Worse
Intel unleashes both "Haswell" and "Avoton" microserver parts onto the world in 2H 2013, built on the 22nm process. The former uses the higher performance core, and the latter uses the redesigned, next generation Atom. Intel will beat all of the ARM folks to market with this product, including AMD's own ARM efforts, and quite frankly, Intel's advantages over AMD still hold true for the ARM vendors (but the ARM ones don't even have X86 compatibility, so it's even worse).
So what's AMD going to do? It will probably offer both Intel and AMD processor based solutions, but it is more than likely that the Intel ones dramatically outsell the AMD ones. AMD's own solutions will compete on cost, but the TCO proposition for the Intel stuff (dramatically higher performance/watt) will likely negate any advantage that AMD's own designs will provide.
What was AMD thinking?
AMD seems to be fighting for its life by throwing things at the wall and seeing what sticks. Maybe there's decent margin in selling the full microserver solution, but Intel still wins because it will get to sell the chips into AMD's - among other vendors' - microservers.
What really bothers me is that AMD's upcoming microserver part won't use its own apparently very good "Jaguar" core but will instead use ARM's A57. What? Does AMD think that throwing away compatibility with X86 and the ability to be a second source to market leader Intel is a smart move? Would it rather compete against the hordes of other unproven ARM server folks with no competitive advantage other than a fabric that they paid $300M for?
Additional disclosure: Short ARMH.