Mobile vs. Server = just another variant of client-server debates
Someone suggested ARM/Qualcomm should just cede the server market to Intel and that Intel should just cede the mobile market to ARM/Qualcomm. I think this is not going to happen for a bunch of reasons. (A "bunch" meaning I don't have the time/incentive to write some long-winded analysis).
1. Almost all mobile users (Android, iPhone, Windows phone, etc.) and even laptop users (Apple, Dell, Lenovo, H-P, etc.) are contacting some kind of "server" (in the broad sense, not necessarily in the old "client-server architecture" terminology which was once so in vogue).
2. That is, while mobile devices are indeed of what some jokingly call "a supercomputer in your pocket," I doubt anyone is using their mobile device to serve other requests. (There's the idea of mesh networks, but nobody is currently having their iPhone 6 "wake up" as incoming requests are arriving....I doubt this will be a big thing anytime soon.)
3. Yet there remains a vast amount of "server" infrastructure. All of the stuff the mobiles are connecting to, including this board, other such sites, Google, and on and on.
4. The fundamental idea is of distributing computation very widely, as economic incentives take things in various directions.
5. Currently, Intel is not making much headway in mobile phones, a little more headway in tablets, and of course it dominates laptops (Apple, Dell, Lenovo, H-P, etc.). Likewise, Intel dominates servers and supercomputers.
6. But there is no fundamental, "basic physics" distinction between processes/architectures for mobile devices and for servers. Sure, there are different "dialings" for power vs. computational horsepower situations. (There is no longer a "high power process" as there was in the mainframe days when ECL and GaAs filled supercomputer niches, and a "low power process" for NMOS and CMOS processors intended for desktops and traffic controllers (joke). Today, the processes in use for SOCs are basically all minor variants of the 14-16 nm FinFET/3D, or 22 and larger earlier versions, whether from Samsung, TSMC, or Intel. Or the handful of much smaller makers.
7. And this is architecturally the same sort of thing we saw when desktop PCs were connecting to servers. (A pretty good thing for Intel that it didn't just conclude in 1985-90 that they should cede the "server" opportunities to DEC, IBM, Fujitsu, H-P and other mainframe and supermini makers.) That is, one class of software was running on the client side (terminals, PCs) and various server-side classes of software were running on the other side (Oracle, other DBMSes, RESTful servers, etc.)
(A side note: the compelling advantages of CMOS (and HCMOS variants) for all aspects of computation were summarized by a guy at Livermore who predicted that the Intel sort of approach would eventually sweep away the GaAs, Cray-type supercomputers and even the superminis. "Attack of the Killer Micros" was the name of the paper, as I recall. Around 1990. And the trend was already happening by then.)
8. So, obviously, neither the ARM/Qualcomm faction nor the Intel faction has any reason to just say "Let's just divide up the market and be happy!" For simple game theory reasons, as well as the technology issue (no natural division in the processes), this won't happen. Each "side" will move as it can. (There are antitrust issues as well, but I won't touch that one.)
9. I've seen this situation several times before. Intel faced scarier competition back during the periods when other architectures were mounting pretty serious efforts to dethrone the x86. No need to repeat the examples here.
10. I'm not even mentioning the "Internet of Everything" stuff. But computation is spreading widely, and the distinction between "mobile and server" will just be lost in the process. As it has always been.
--Tim
