I'm not sure I see the point of putting an ARM copro into an x86 system? The Medfield/Cloverfield devices have killed the "ARM is more power efficient" myths, and don't offer any performance advantages over just adding more Atom cores? Am I missing something?
I have spent quite a bit of time thinking on this one and IANACS, but hopefully my view is reasonably coherent.
The IA32 architecture has evolved in a very odd but powerful way over the last 27 years with the virtualization routine really being a big deal and economics trumps technical considerations time after time, after all RAID is premised on inexpensive devices and look at what SATA did to SAS (and more broadly how SCSI never quite got there). MIPS and SPARC and POWER and to many other architectures and processing systems to think of have all been around and apart from filling special niche markets have generally not managed to rival the great birth of 1985 (the 386) in what can be termed general purpose computing (and of course even there you have the fact that the 386 had a heads up because of the 8086 forming the base of the IBM PC).
Now on a basic level general purpose computing really is all about getting maximum pull of hardware, software compatibility and killer apps (not apps in the modern tablet computing sense but apps as an application that made somebody buy a computer like Excel). This means that when OS/2 was priced out of the market its death warrant was signed and when IBM put together the PS/2 and tried a MCA bus over ISA (which took a damn long time to die) it didn't do itself any favours (until ironically the PS/2 connectors were adopted later on as a standard by the then kings of general purpose computing Microsoft and Intel). In my view the big boys for general purpose computing remains today Intel and Microsoft (although not to the same extent as about 14 years ago) with the most pull and then a host of players - most of whom have been around for a long time - but that in reality the market does have bizarre checks preventing any one entity from going on its own without getting knocked into the cold: Microsoft can introduce a new harddrive system at its own peril for example. Apple is an interesting side player (and always have been) because they stay out of what is the general purpose PC but really are the most succesful general purpose computer manufacturer that isn't a PC but thats a bit of a distraction. The 386 wasn't exclusively used in general purpose computing though and there are embedded systems. And there was a progression from IBM PC to a fairly loose Wintel concept of the computer with PC99 as a good example of how much pull Microsoft and Intel have on the market but also of just how much of a market driven industry it is. And lets face it Intel failed to bring to the market a 64 bit extension that was accepted and the Pentium 4 was a disgrace so it had its ass handed to it and basically had to adapt with regard to the first issue by letting AMD have a headstart with AMD64 and introducing EM64T to follow the market and in the second instance to modify their development stream.
The only exception to a rule of path dependence leading down to the IBM PC has been on the front of mobile phones to smart phones joined up with tablets and phablets and all other vile beasts. In this instance a handful of processor manufacturers using an ARM base have managed to grow into general purpose computing devices with a vengence. Nothing demonstrates this more than Android and iOS except for Microsoft going in with guns with RT. So Windows 8 has dropped support for everything except x86 and ARM and if you look at major distros of *nix os'es the story repeats itself.
So if we look at it I would state in general terms that there are three quasi-arch (a somewhat nebulous term I am injecting) support routines running x86-32, x86-64 and ARMv7. Now I know that with ARM the architecture discrepency is probably larger than between 32 and 64 on x86 but in terms of software development - and the somewhat capricious criteria for a quasi-arch I am going with - there is a difference. Windows 8 quite specifically accomodates all three quasi-architechtures and there market position means that this reinforces path dependence.
Now this makes life a little more difficult for ARMv8 - which I am not sold on, but lets assume manages to get critical mass it still doesn't affect my basic premise because it would probably affect the market currently using HP-UX and Solaris and what not, so niche markets.
If Microsoft wasn't developing both RT and Phone with the correlated development spend I'd write ARM as dead in the water in the next generation but Microsoft is and Android and iOS are turning up new developers en masse.
Basically it boils down to path dependence acting in favour of ARM co-existing with x86-32 as quasi-archs for 32 bit software and x86-64 being the beast boy. Now your workstation computers, and servers, really need the tools which the x86-64 quasi-arch brings with it, especially in terms of virtualization and so on.
With Atom Intel is adopting the fact that more processing power does not mean better computing and lets face it the x84-32 quasi-arch is really matured and is beautifully virtualized. Big bad datacentres and monster computing will work like a bomb and for the most part it will make no sense for systems to virtualize other quasi-archs unless the market demands that they do - and here is where ARM in my view already has itself running. Already a PC can virtualize ARM and simply because of development this will continue to improve. And again the Microsoft factor kicks in and I anticipate that Server 2012 will get software to host RT apps.
The 86 quasi-archs don't appear to have gotten the power down to interaction time routine developed to quite the extent that ARM has and for user centric computing this is relly important. On the other hand I can't think of a production use of an ARM system which is running at high performance on a constant basis. Virtualization and "the cloud" is largely premised on the assumption of maximizing the utilization of processing power whereas ARMs strength has been in providing a quality user experience on demand - as in smart phones etc ...
Now when looking at this together I see ARM as the only quasi-arch which can be thought of for secondary software handling (with there being enough software written for the platform and OS support being a given). When you are running a desktop computer you don't adjust virtual machines etc ... you do however use different amounts of computing power and get highly annoyed when Adobe Photoshop is killing your web-browser, now one solution is to do your web browsing and must playing on another machine (a tablet no doubt) but if that could be handed over to a processor - which is cheaper than the tablet there is play.
My view was reinforced when I played with Windows 8 the Metro apps and the desktop stuff are being built to work well in parallel to each other and nothing will make Microsoft happier than having Zune or whatever music player runs in the backround being eligible to be handed over away from the Core i5 beast that is the desktop workstation. Now there are two things an ARM co-processor has going over an Atom co-processor - cost and time to live (and yes Intel might make improvements there but I am not holding my breath). The economics factor in my mind plays like this - if it is cheaper to have a Microsoft supported ARM coprocessor than the price difference between a Core i5 and Core i7 the market is going to have space for the coprocessor. Add to this the fact that you don't have heat being added and so on and life is good.
Add to this the fact that the Rasberry PI and other hobbiest systems are using ARM and the case grows.
