.. army of ARM...
Did you really HAVE to do that
The ARM RISC processor is getting true 64-bit processing and memory addressing - removing the last practical barrier to seeing an army of ARM chips take a run at the desktops and servers that give Intel and AMD their moolah. At the ARM TechCon conference in Santa Clara on Thursday, the top brass at ARM Holdings, the company …
"Intel has a lot of time to make Atom a better chip, and AMD still has time to do whatever it is going to do. ARM needs to move faster – or AMD will."
Intel haven't moved very fast as yet. If this doesn't chivvy them up a bit I don't know what will. It's almost as if Intel's strategy has been to wait for ARM to pose a serious threat and then pull out all the stops on a proper low power development.
That seems awfully risky. Intel have to make huge improvements all round because ARM's performance / Watt is beginning to look seriously good. AMCC are talking about 3GHz quad core ARM64 in the middle of next year! There's no word as yet on power consumption, but if it conforms to the norms of what ARMs seem to achieve then it's going to be a very cool running chip indeed. Whereas Intel have to pull off a large step change in their chips' power figures and get it right first time.
Look at it another way. ARM64 fails, ARM and Intel survive as they currently are. ARM64 succeeds, Intel lose out, ARM win big time. Intel won't be threatening ARM in the mobile market anytime soon. They way I see it Intel are probably in it for survival at the moment.
And look at it another way again. Itanium must surely be a dead end now. Intel can't possible afford to waste engineers on Itanium when the ARM threat has to be addressed. Very soon no one is going to be at all interested in Itanium servers unless Intel can somehow make them low power too, and I don't think we've even begun to hear the merest peep about that on any roadmaps.
Operating systems still use the whole address range (or can), just as in the 32 bits days the OS would take the top one or 2 Gig's for itself and the low memory for user space, 64 OS's do the same so any address in system space or on the stack will use those high bits.
Anyway with many GBs of ram why would anyone be assing around encoding data in to the pointers! That is just stupid.
Which is why Intel and AMD both stopped that pissing match.
I'm surprised ARM is chasing speed when they can produce much smaller(cheaper) cores than x86 or pack a lot more into the same space. Virtualisation is easier with more cores, not with more speed.
The reality for raw grunt is that the target hasn't been x86 for a long time, it's GPUs. If you can pack 900+ cores in the same space as a Cuda, run almost as fast and use a lot less power then there will be a lot of paths beaten to ARM's door.
(FWIW: x86 and Sparc were always among the least power efficient CPUs - that didn't matter as much as being cheap - until people started hitting data centre infrastructure limits.
(Try buying ~1MW in central london - you'll be told it can't be delivered until 2020 or so. The existing power distribution network is all tapped out. That's starting to impact a bunch of businesses and the transport system too(*). My suggested solution to $orkplace was to stop being so stupid and build the $enormous data/compute centre on cheap land near a power station.)
(*) Check out the power draw of the average commuter train.
Not with Bulldozer but with Brazos and the Fusion APUs. By giving you "good enough" X86 while giving fast and lower powered GPU cores you can speed up a heck of a lot of math thanks to openCL. And just to remind everyone that is while they are still using VLIW, they are in the process to switching to vector based GPU cores which should be even faster and allow the GPUs to be like super FP cores. Currently doing double precision FP takes as much of a hit as slowing the GPU to 1/4th, whereas the first gen vector units will be less than half and the goal is to get full DP FP by the third generation.
while I can see a niche for the ARM in the server room I think there is still too much code based on X86 for it to take a big bite, oh and I'm bettin g win 8 (MSFT's bid for ARM) will make Vista look light as a feather and fast as a rocket!
Alan wrote: "I'm surprised ARM is chasing speed when they can produce much smaller(cheaper) cores than x86 or pack a lot more into the same space."
I'm sure ARM is doing just that. They are just betting on multiple horses at the same time. As well they should, for different people have different needs: Some need Ferraris and others need commuter buses.
You mean just klike Windows NT ran on: MIPS, ALPHA, Power and x86.
Alpha lasted longer that the rest as Microsoft uses Alpha for all its initial 64 bit porting, but they never realy supported them to any credible level (eg: no native Office products).
Until they prove otherwise there is no reason to not beleive it will last any longer this time, or be anything other than a niche OS for Phones and FondleSlabs, just like Windows CE has been for many years.
There is a notable difference between now and ACE/ARC (NT on Alpha, MIPS, PPC, etc).
The market wants ARM and MS know that. The reverse was true with multiplatform NT - the chip+system vendors (thought they) needed NT, and wanted MS to help.
Now MS need ARM. The boot is on the other foot.
If Windows on some flavour of ARM does catch on, one of the things it will magically do is introduce some compatibility into those parts of the ARM market. Which is likely to benefit the Linux people just as much as it benefits Bill. Which is nice.
The low power Xeon E3 chips that Intel announced at the beginning of the year (and that's positively ancient in CPU land) had quite reasonable power consumption figures, given that they were quite capable x64 chips. More expensive than Atom, of course, but to my mind a low end Xeon is a rather more useful bit of gear than a top end Atom...
...why haven't MIPS got their act together.
They have had 64bit cores since the 90's, it was not that long ago that SGI were still making workstations with them.
Though they may just be waiting for the Chinese to do all the work for them, they have new MIPS derived coresboth 32bit & 64bit. They have versions for embedded to servers and they are already quite low wattage (4cores @ 1GHz using about ~15W using 65nm chips).
When the Chinese are manufacturing these by the billion Intel, AMD, and ARM could be in trouble.
For those unaware of any of this try looking up "Loongson" and "Godson".