Intel, king of the desktop- and server-processor markets, sees opportunity in the burgeoning mobile market; ARM, king of the mobile market, believes it can take a sizable bite out of the data center. But will the upcoming battle leave a single dominant architecture (Intel’s x86 or the ARM architecture) across domains, or will the two chip giants reach a stalemate, possibly with some minor incursions into enemy territory? Here’s a look at some of the market dynamics that could affect the outcome of the struggle.
One major factor fueling ARM’s strategy is the growth of the mobile-device market. Almost daily, some pundit is claiming the PC is dead—being replaced by the duo of tablets and smartphones. Clearly, the market for tablets and other mobile devices is expanding, likely in part at the expense of PCs and notebooks. Many users are simply looking for entertainment rather than serious content-creation capability, so a tablet or smartphone (or both) may offer a better option than a PC. Call me a skeptic, but I just don’t see tablets replacing desktops and notebooks as real productivity devices. Eventually, the market will balance as users decide which products they need and largely cease buying those they don’t.
The result of this dynamic, whatever the case, is that ARM is riding a big wave that is highlighting the ARM architecture’s low power consumption. But if the company can bring low power to mobile markets, why not bring it to “fixed” markets like the data center? With computing and storage demand exploding, particularly as users move more services to the cloud, data centers are bearing more of the burden. But not everyone is pleased, particularly because of the data center industry’s growing portion of world energy production. As calls for greater efficiency crescendo, ARM seems well positioned to strike into the heart of Intel’s realm. The question, however, is whether a lean, fit ARM processor can handle the heavy lifting of the beefier Xeon processor. The answer may be what many already suspect.
Intel and ARM ISAs
In the 1980s and 1990s, chip designers struggled over which architecture—RISC or CISC (we’ll avoid the details)—was superior. According to a recent paper (“Power Struggles: Revisiting the RISC vs. CISC Debate on Contemporary ARM and x86 Architectures”) presented at the 2013 IEEE Symposium on High Performance Computer Architecture, the debate previously focused on chip area and processor design complexity, but “today, energy and power are the primary design constraints.” This territory is familiar to those following the ARM-versus-Intel struggle: ARM claims it can bring the power efficiency of its architecture to bear in the data center server market, stealing share from Intel’s x86. But one of the paper’s key findings is that neither the ARM nor x86 instruction-set architecture (ISA) is fundamentally more efficient. How these ISAs are implemented (the microarchitecture) affects efficiency, but this implementation is a design choice that involves a tradeoff: “ARM and x86 implementations are simply design points optimized for different performance levels.”
In other words, as one might expect, ARM’s task (i.e., the task of designers using the ARM ISA) is more complicated than beefing up its mobile processors for use in servers. Not all the efficiency features of the mobile devices will simply carry over to server processors while delivering high performance. Similarly, Intel has historically focused its processor designs on performance rather than efficiency; as such, it has seen little uptake in mobile, but it has dominated the data center. Moving to more-efficient designs, however, requires sacrificing performance.
The above-mentioned paper, which is based on a study by researchers at the University of Wisconsin-Madison, confirms that neither ARM nor Intel (x86) has a fundamental advantage that could give one or the other dominance across all markets. Instead, each company has historically focused largely on one design consideration: efficiency or performance—typically at the expense of the other. As Intel tries to increase the efficiency of its offerings and ARM tries to add more performance, the two companies are likely to meet somewhere in the middle. Design skill may determine who gains a slight edge, but neither will offer a single chip that does all things well.
Momentum May Decide Winners
Although ARM has been on a marketing offensive to promote its foray into servers, it smacks of at least moderate hype. At a minimum, design considerations probably won’t be the sole deciding factor: Intel has the benefit of momentum in the data center. A move to ARM requires a move to new software as well, so switching architectures requires a greater investment than simply dropping in a new chip in place of the old. Other factors, such as prices and business models, may play a greater role in determining the market distribution once ARM-based servers reach significant production levels.
Intel could make inroads to the mobile market with Windows and associated software, but it must deliver both the efficiency to provide competitive battery life and sufficient performance to keep Windows-based software running at a user-acceptable speed. But ARM’s incumbency in mobile, by itself, still creates a significant barrier for Intel.
End of Moore’s Law
Yes, every year someone predicts Moore’s Law will end by the following year (or within a few years), but the real end—at least for traditional silicon technology—may indeed be in site. One of Intel’s major advantages has been its lead in semiconductor process technology; if Moore’s Law finally fails, x86 and other chips could soon end up on equal footing with regard to transistor size, speed and so on. At that point, any edge granted by process node (which includes power efficiency) would dissipate, leaving design as the main factor determining chip characteristics. (At this point, a much-needed focus on software efficiency would probably arrive as well.)
ARM chips generally lag Intel’s offerings in process technology, so the end of Moore’s Law—assuming it were to arrive within a couple years—could benefit ARM. Again, however, this situation doesn’t address the matter of microarchitecture rather than ISA as the determining factor in performance versus power, so the battle between ARM and x86 would likely continue unabated.
Neither Intel nor ARM will likely emerge as the sole dominating force in both mobile and servers (the data center). ARM brings efficiency capabilities that serve mobile well, but those capabilities come at the cost of performance, limiting its use in the data center. (Nevertheless, an ARM-based server could be well suited to certain lighter-weight tasks.) Intel’s performance lead, although superior for heavy workloads, sacrifices efficiency in lighter tasks, such as mobile computing. Efforts to improve efficiency will come at the expense of performance, bringing the two companies to a meeting in the middle.
An end to Moore’s Law, should it finally arrive soon, could aid ARM slightly, although other research efforts in semiconductor manufacturing may allow Intel to maintain an edge. Interestingly, however, the work done by researchers at the University of Wisconsin-Madison indicates that neither ARM nor x86 has a fundamental advantage. The RISC-versus-CISC debate may still be at a stalemate, leaving implementation effort as the deciding factor among specific products.
Thus, although ARM has momentum owing to growth in mobile markets and a certain level of marketing hype surrounding its server efforts, it has no fundamental edge over Intel in energy efficiency. Even though we are unlikely to see a world dominated by ARM processors, the battle between the two architectures should yield some interesting business and design adjustments that could greatly benefit processor consumers.
Photo courtesy of bmjames