When Advanced Micro Devices (AMD +3.50%) launched its very first Opteron processor, it was an industry game changer. For the first time, not only did AMD -- Intel's (INTC +1.05%) much smaller competitor -- finally offer a meaningfully better product with respect to performance, power, and scalability, but it was rapidly gaining share against Intel. AMD caught Intel asleep at the wheel and took full advantage. However, Intel turned its server efforts around in a big way and took back the share that it lost -- and more. The question now is how well AMD's server efforts hold up today and whether things will get better or worse going forward.
Traditional server space going from bad to worse
In the traditional, higher-performance tiers of the server space, Intel's products are viewed as superior. The key metric used here is performance per watt per $0. Performance per watt is important because energy costs are a huge part of the cost of operating a data center, so a substantial lead in performance per watt allows a chip vendor to command a significant premium to competitors up front.
AMD's current Opteron line of products are competitive with the mid-range of Intel's older generation of Sandy Bridge-EP products on the basis of performance per watt per dollar, largely because AMD is willing to be very aggressive on price. But Intel's recently released Ivy Bridge EP processors -- which pack much more performance into the same or lower-power envelopes as the chips that they replace -- cause serious problems for AMD's offerings here. Indeed, AnandTech's recent review of Intel's next-generation processors had this to say with respect to the competitive situation:
At the end of last year, AMD was capable of mounting an attack on the midrange Xeons by introducing Opterons based on the "Piledriver" core. That core improved both performance and power consumption, and Opteron servers were tangibly cheaper. However, at the moment, AMD's Opteron is forced to leave the midrange market and is relegated to the budget market. Price cuts will once again be necessary.
The bigger problem here is that while AMD can cut prices in a bid to maintain its marginal market share, Intel's cost of production for its latest chips are probably lower by virtue of the following facts:
- Intel builds all of its chips in-house, while AMD has to pay GlobalFoundries' margins.
- Intel's 22nm process allows for die sizes that are significantly smaller at every price point. For example, AMD's Opteron 6378-a 16-core, dual-die processor sells for just $860 for what is about 630mm^2 worth of silicon. Intel's much higher-performance, top-of-the-line Xeon E5 2697 v2-a 12-core, 24-thread part sells for $2,750 and only sports a die size of 541mm^2. The smaller die parts that AMD's Opteron actually goes up against are even cheaper to build. In short, Intel packs much more performance per unit of area than AMD can.
How about micro-servers?
Much has been made about the rise of microservers -- servers focused on using as little power as possible for lightweight workloads. While the higher-end servers prioritize high performance with a drive to get those high performance levels in as small a power footprint as possible, microservers aim for a much lower power level. The goal is to achieve as much performance as possible within those much lower power constraints. This is an area that ARM and its licensees, such as AMD, Applied Micro Circuits, and Calxeda have been touting as their big opportunity to disrupt Intel's dominance.
AMD made an interesting move in 2012 by acquiring microserver vendor SeaMicro, which gave AMD the interconnect fabric required to connect all of the resources in a given system, usually either targeted at computing or storage. While AMD offers variants of its SeaMicro systems with its own home-grown Opteron processors, the ones that seem to be selling -- including the most recently publicized deal with Verizon -- seem to be the Intel-based ones, although AMD does claim that even the recent Verizon deal involved some systems with AMD processors.
The problem, unfortunately, is that while AMD's road map calls for its first competitive, ARM-based microserver part to sample during the first half of 2014 for production in the second half, Intel's recently released C2000 series of microserver chips based on the firm's latest Silvermont processor design are here today with 64-bit support, full compatibility with the wide gamut of X86 server software, and built on the very power-efficient 22-nanometer process. AMD's second half of 2014 parts, code-named Hierofalcon, will be built on Taiwan Semiconductor's 28nm process. These parts will go up against Intel's 14nm Denverton and Broadwell parts by launch, so the competitive situation from performance, power, and cost gets no better with these new chips -- in fact, being two generations behind in process technology is even worse.
AMD has a long way to go
Make no mistake, AMD has a very long way to go before it can become a real contender in the server space again. It's still fighting against a competitor hell-bent to preserve its market share, and many of the disadvantages that it had with its "big cores" don't go away in the "small core" space. Be very careful before buying the AMD server story.
