Go ARM!
Getting ready to kick some chipzilla's arse.
ARM Holdings' high-performance, low-power Cortex-A15 processor design will appear in products in late 2012 or early 2013, when it will begin to muscle in on territory long dominated by Intel's x86 architecture. "With our upcoming Cortex-A15 processor, we are definitely moving closer to the day when your smartphone or tablet …
Once software is written with 8+ cores in mind (note that many are simply dual-core or quad-core capable and gain no benefit from, say, an Phenom II X6), then we can have 16 cores. Currently, the better poke would be from a more capable pipeline and faster GHz. Hence why the Sandy Bridge parts are rather tasty....you can pump a 3.4GHz part to 4.6GHz reliably. Gives substantial performance gains across the board.
Hopefully ARM's entry into the low end (which will likely hurt AMD more than Intel btw), and AMD's potential (please say it's so!) counter-punch with Bulldozer, will cause Intel to innovate and stop holding it's punches. It likely has the ability to pull a Pentium-to-Core2 jump again, but is trickling out the performance gains due to lack of competition in the high-end. No sense playing all your cards at once, right?
I have written on several occasions that I am not interested in an iThingy or equivalent as any kind of "supplement" (ie waste of time and money), I am only interested in a tablet when it is an all singing all dancing pc and can *replace* kit here at Arctic Fox Towers. When the tablet concerned can replace my lappie, the front room pc and my Kindle (which will be getting a little aged by then) I will be willing to part with some serious hard earned for the kit concerned. A little docking station in our TV bench connected to our home theatre system and we would be all set. Seven inch form factor could live in my coat pocket between home and work very nicely thank you.
Servers are all fine, There are already some ARM based server products.
But imagine no need to sync
addresses
bookmarks
documents
The "phone" docks onto a keyboard screen (wirelessly from your pocket using BT for keyboard and Wireless HDMI, or into a charging dock).
It then can do everything a real laptop can do. The ultimate portable workstation. Without Cloud Computing (which is mainframe/big corporation control instead of your own computer).
Needs 30Gbyte to 300Gbyte storage though. Oddly a Toshiba 1.8" HDD uses less power than an SSD. Samsung make a 120Gbyte 1.8" HDD at least. Toshiba have a 250Gbyte model. These are not much bigger than a battery. Would fit a 4" WS (maybe wider than 16:9 to make it slimmer for pocket and more page width?)
What they want is access to their "stuff".
As wireless mobile networking gets faster and cheaper, cloding makes more and more sense.
The tin-foil hat wearers will be worried about third party service provides but worried folk and organisations can manage their own data and can make their own clouds.
ARM & their licensees have a very large hurdle to overcome to beat Intel/AMD/x86 out of server markets, there is compatibility, integration with legacy kit, staff knowledge and plain old fashioned luddite-ism.
I'll believe it when I see it basically... Intel chips are fast, regularly improving and power usage on some models has dropped a lot. Besides, an ARM chip can't help with the rest of the server's power draw, it's just one component.
Virtualisation is already consolidating light workloads onto a single server, and there is a big attraction in running/managing a few servers rather than many with light workloads on each. Not many workloads benefit from extremely threaded solutions either, so spreading jobs over loads of ARM cores won't necessarily be a winner...
An low power ARM server sounds cool, whether it presents a valid/better solution to any actual needs remains to be seen...
Indeed it is, but I can point to one of my racks, which pulls 2kW. There are 4 Xeon 5450 dual-socket Proliants in there, and each of those chips has a TDP of 80W. Well, that's 640W right there, and since they're VM hosts I do try to run them hard.
So if 25% of my power budget is CPUs, and I can maybe save 50% of *that* by going ARM, that's a significant saving. Then there are X5400s that pull *twice* the power of mine, but the rest of the server is the same. Sure, there'll have to be a significant speed boost for an ARM to keep up with that, but the potential for energy saving is very real.
(Oh, that cabinet is also full of drive enclosures, tape library, switches, blah blah - it's not just those 4 servers.)
If you're going for dense servers in a rack, the benefits of saving a little power on each server can be significant. I don't know if you've dealt with colocation, but for that you're restricted on the power consumption per rack. If you can save some power you can fill the rack some more. If it's your *own* rack, you save once on the electric bill, and once again on the cooling bill.
Legacy stuff needn't be a big issue. Plenty of Linux distros run on ARM, so you're covered for those applications. Windows is coming for ARM (apparently, though I've heard that in the past), and if that means Windows Server, that means SQL and Exchange etc etc. Who cares what's happening inside the box so long as what goes in and what comes out is the same.
Migration could be a pig - I don't know if there would be endian issues with the data, and I'm too lazy to look.
It won't be nearly such a big hurdle if energy costs keep rising, and the amount of processing you can do on an ARM system per watt-hour remains many times the same for an i86_64 system.
Of course, adoption of ARM has to be made reasonably practical. ARM boards have to become available as commodity products at a comparable price to an Intel one (say £60 for ARM CPU + motherboard competitive with Atom CPU + board of similar performance). They have to have standard interfaces (SATA, USB, ATX PSU connector, etc). And they have to run mainstream O/Ses. Windows 8 is going there. Linux already is there, but we'll need RHEL for ARM and a more standardised ARM board ecosystem before it takes off.
I would argue that as competition from AMD forced Intel to design better CPUs, a little competition from ARM will help both Intel and AMD to design more power efficient chips.
There is too much code already written for x86 architecture to port to ARM, so whenever possible, people will buy the hw to run their existing sw, not the other way around. ARM only got in because Intel and AMD had nothing to offer for that power efficient segment - now that both Intel and AMD have waken up, I expect them to give ARM a hard time expanding into their territory (servers, real notebooks and desktops).
As ARM already has a lot of code written for it, replacing it in existing apps will be a very tough sell fro Intel, but it will not replace x86 anytime soon.
"we'll need RHEL for ARM and a more standardised ARM board ecosystem before it takes off."
MontaVista and other non-IT non-consumer (but well known in their sector) Linuxes are there already with the software. And if the Window box folk are serious about Windows 8 on ARM and aren't using it just as PR talk/idle threats, that'll get you some kind of more standardised board ecosystem. It would be a laugh if Windows 8's dependence on standardised ARM hardware was what finally allowed a real ARM/Linux ecosystem for desktops and servers as well as the embedded stuff which is everywhere already, if you look for it.
"I don't know if there would be endian issues with the data, and I'm too lazy to look."
Doesn't ARM swing both ways, I forget? I'm pretty sure gcc for ARM does. OK it's a bit trickier once it's in a system, but... Either way, folks have been surviving this problem since the days of M68K and SPARC vs DEC and x86.
"I expect them (x86 vendors) to give ARM a hard time expanding into their territory"
Maybe so if ARM want to attack legacy x86 territory head on, but why should they? Equally, an Intel/AMD powerdown move can't be done in the foreseeable future. There are architectural features in ARM which enable OK performance in a low power system. You can't put these features in an x86 system without losing x86 compatibility. And if you lose x86 compaitibility you might as well recompile anyway, just like Microsoft are planning to do and just like the open source world already did.
And then besides that, there's the years of SOC design and build experience that the ARM licencees have. Intel used to be one of them too. Even Intel can't grow experience like that in six months - but Intel do have other ways of encouraging their more valuable customers not to stray from the One True Way (tm). Just ask Dell.
x86 isn't going to disappear overnight, but there are interesting times ahead for anyone or any organisation that is totally x86-dependent.