Can ARM hold the mobile advantage over Intel's x86 architecture?
- 25 June, 2013 13:17
When it comes to personal computing, most of the growth today is in the mobile market, not traditional desktop PCs and laptops. You could say that the PC concept is morphing to include next-generation smartphones, phablets, slate-centric tablets and convertible devices in many forms. To date, most such devices have been ARM-based. Is that about to change?
Intel has announced its next generation of x86-based Core chips (Haswell) and new Atom architecture (Silvermont). Many question whether Intel can truly compete with the myriad ARM-based designs now powering the vast majority of mobile devices from ARM licensees such as Qualcomm, NVidia, Samsung, etc.
I believe it can. But there is more to it than just producing low-power x86 chips.
Much of the competitive argument revolves around the decades-old disagreement over the overall benefits of RISC (reduced instruction set computing) vs. CISC (complex instruction set computing) architectures. While RISC has many advantages around simplified silicon and smaller chips (which translate to lower power requirements), it also has more difficulty processing complex instructions. CISC chips implement such processing in optimized hardware acceleration, while RISC chips use software algorithms.
When phones were "dumb" devices, this was a non-issue. They worked quite well within the more limited RISC-based architectures and processing constraints. But as smartphones and tablets have evolved to become more computer-like, the amount of complex processes has increased dramatically. This is why many cores are being designed into chips (especially the computationally complex GPU cores that handle graphics, multimedia, gaming, etc.). Indeed, along with the many 4- and 8-core CPU chip designs available, we are now seeing as many as 72 specialized cores (e.g., GPUs) implemented on the same silicon. This increases size and uses more power. Power management has improved, making each core and the full chip more efficient at power management. But complexity of functions continues to make this an issue as more hardware acceleration is added.
It is generally (although not always) true that CISC architectures handle complex calculations much more efficiently than RISC. That's why PCs on RISC chips never succeeded (even though several vendors -- Transmeta, MIPS, SPARC -- tried). It's also generally true that fewer cores are needed with CISC for equivalent processing efficiency as more accelerated special functions and instructions are implemented in silicon (i.e., doing more with less -- it's not about how many cores, but how they are used in complex computations). And if Intel can neutralize ARM's power and size advantage (as it is doing with advanced processing and architecture work in its latest generation of chips), it can enable more compelling designs, especially given Intel's substantial capacity to highly optimize software to run on its platforms (as it has done for Android and Windows).
Bottom line: It is not clear that ARM has a long-term competitive advantage because of its RISC-based approach against Intel's x86 architecture. Clearly ARM has momentum on its side based on design wins in current products. But as it must add increasing complexity to power more advanced functions, its architectural limits represent significant performance constraints. Increasing chip complexity to include hardware-accelerated functions will ultimately increase the power and size of ARM-based chips and negate the advantage it has over the more complex CISC implementations, while at the same time CISC implementations are making significant advances in power reduction and size. Longer term, the RISC advantage that ARM currently enjoys over Intel's CISC architecture will not be sustainable.
Jack Gold is the founder and principal analyst at J.Gold Associates, an information technology analyst firm based in Northborough, Mass.
Read more about processors in Computerworld's Processors Topic Center.