other chip aficionado's), to change the subject a little. I read an article on Engadget claiming Intel was planing on taking the Atom processor into new low voltage designs and having them in phones. Since what I think I have read in this thread is that x86 is dying and Arm is taking over. What is your opinion of moving x86 into the lower power user space and trying to push to equalize it with Arm chips in terms of battery life. Is it even possible for them to do this without seriously lagging behind in performance? Are they just trying to keep their chip's alive longer by fooling people? Or could they really eventually make them powerful enough and battery friendly enough to make it viable?
You have a _lot_ of factors in play here.
Earlier in the discussion I made the mistake of thinking Intel had shifted to SOI instead they've been trying to compensate with other technologies like stressed silicon and were first with high-k metal gates which doesn't change the connotation of what I went with as a result. Silicon-on-Insulator is more expensive then bulk silicon, but it has significant benefits. ARM did a test in 2009 with the 45nm process to test this and found a 40% reduction in power consumption and the ability to make a 7% circuit area reduction. IBM, AMD, Toshiba and otherwise for the most part have stuck with Soitek's Silicon on Insulator technology while Intel has snubbed it. These 28nm ARM parts should also benefit from high-k metal gate technology if fabbed at Global Foundries, as they've put in the effort specifically for that.
High-k metal gate technology has helped give Intel an edge in recent years, but by the end of they year everyone will have access to it. The Global Foundries project roadmap from here on, if followed, will not give Intel as much room as they would prefer and with the TSMC will be providing the majority of production for ARM processors.
As far as rating performance between ARM and x86 processors with numbers instead of appeals, nVidia's
new Kal-el with four A9 processors tested out on their synthetic benchmark of choice slightly edges out a Core 2 Duo. I'm was unable to find if Coremark takes full advantage of 64-bit verse 32-bit, but if that is the case this reflects well on the ARM Cortex A9 as it is ARM's first, still 32-bit, attempt at making a dual-issue out of order processor cores like those currently offered by Intel and AMD in a 64-bit flavor. Coremark is however noted as a small benchmark that doesn't reveal issues with L2 and greater cache or memory access.
Intel's bet basically boils down to they can get power usage down enough to be acceptable for the device/application, and that then the momentum of x86 software will take over winning them the market. As usual for Intel they love to overconsolidate things. This is why their approach to SMT is half assed, from my perspective. For their application SMT involves making their consolidated scheduler even more complicated and harder to debug, so they can fix the problem of the scheduler missing because it's not adequately debugged. It has nothing to do with handling and retiring more threads then they would if the scheduler was perfect. I personally really don't like the fact it reports to the OS as if the "extra threads" are cores, and hence just asking to cause problems.
AMD's bet is based around something that's been progressively developing in various realms that drive computing technology development, known as General-Purpose GPU technology. This is in addition to the long standing practice of AMD involving having independent FP and integer schedulers thus allowing greater flexibility and easier debugging/refinement/modification. This in turn will be augmented with the new dynamic Turbo Bulldozer will use that, within stability limits, allows all power to be used for a task regardless of how many cores it prefers and take advantage of a previously unexploited 30-40% overhead.
They've also continued with both ATI and AMD products their emphasis on max throughput. GPGPU technology handles FP and integer operations across what by current x86 or ARM concepts is an insane number of processors via Stream Computing, with flexibility increasing as the hardware and software developers have continued to go back and forth with refining the overall technology. Prior to this maturing Sun Microsystems RISC design effectively championed this approach while being a little too proprietary, unlike GPGPU will be. In effect their intent is for everyone to have a personal supercomputer, and to drive supercomputing to the exoFLOP range. Despite what Exophase may think about AMD's server share waning into oblivion while waving the Intel flag, the reality is a Dell product with Opterons was awarded the Server of the Year and Opterons were rewarded as the best for max throughput with highly parallel loads in the server sphere.
Both will benefit dramatically in FP performance from the AVX extension to be introduced in Sandy Bridge and Bulldozer respectively. The AMD implementation with Bulldozer currently appears to be the most refined.
nVidia is going hardcore on ARM now, see the linked Kal-el artile, but their current offering do not appear to have any emphasis on GPGPU technology yet, like those of AMD. Their GPGPU CUDA cores are four way hyperthreaded like Intel's to correct misses, so if you think that is a wonderful technology they could easily do it to ARM processors. Texas Instruments and Qualcomm will undoubtedly at the very least do their best to keep up.
Technologically ARMH (ARM Holdings) have the advantage of being able to start out with a clean sheet of paper and thus avoiding the need all the logic circuits translating old instructions from a time when just how one makes a effect CPU was being figured out, and newer extensions, to the instructions of a fundamentally different internal processing core and then retranslated on the other side. These translators have traditionally had to run constantly and use quite a bit of power as a result. There are currently some tricks to try to make things more power efficient by stuttering this ala the current Race-to-Execute strategy favored for throttling processor cores. ARMH also has the benefit of using a reduced instruction set verse the x86 processors and being subject to a hardware version of Linus' Law given how their licensing setup works.
The problem for ARMH is that the processor penalty for translating instruction sets in software ala qemu works both ways, and mobile devices are intrinsically different from more traditional client or server computer applications. While in a purely FOSS universe that would not be an issue, as things stand inertia is against ARMH products in these spheres. Yet it must also be noted that world governments have increasingly been breaking away from the Computer giants because they don't care for that degree of foreign control over their computer systems. Hence Linux is displacing Windows strongly in use by world governments not out of any performance issue on way or the other, but one of local verse foreign control. ARMH designs undoubtedly will appeal to the same interests.
Windows 8 based on patent filings will be based on a Hypervisor and there is talk of some interesting tricks around that which would improve backward compatibility dramatically and let it run Linux apps natively. So overall the best conclusion seems to be to quote a recent game trailer "It is a time of rapid technological advancement" and to leave it at that as far as anything definitive.
The next years will involve a lot of competition between the various approaches. Personally my money is on AMD for client and server space as their approach has the highest flexibility offering all the advantages of consolidated and Stream processing approaches with what has become the industry standard ISA. That Anandtech's die size estimate for nVidia's Kal-el is 80mm^2 @40nm versus AMD's Zacate at 75mm^2@40nm, both from TSMC, and that fab price is associated with die area tends to point towards things also not being as good for ARMH as those hoping to invade Client and Server space would prefer. Versus Intel the ability to play games well trumps Super Pi performance any day, and Fusion dominates in that easily particularly in terms of the -book market. Other then a 5-watt C-50 setup running a tablet for demonstration purposes and their manufacturing arm, Global Foundries, is set to turn out the latest and greatest ARM processors it's unclear what the actual plan is for tablets and smart phones beyond they want back in after selling the division that had been doing that to Qualcomm.
Investmentwise I wouldn't buy ARMH stock at this point as their stock price is probably best described as hyperinflated. Their [stock] price/earnings is around 100 and the company isn't really staged to explode in the short term, which to me indicates an excess of speculators and hangers on versus likely real growth.
Yeah, whatever. If you think I'm not worth arguing with then there's no need to continuing to argue with me, only while talking to someone else right in front of me. Every post you've made in this forum contains a flame to me. I've tried to play nice and reasonable but some people are just determined to be jerks no matter what.
Wow, people sure are anti-MS and anti-x86 for mysterious reasons.
If I really wanted to I could keep going with the abject lack of civility or reasonableness, or just point out your habit of warping and railroading what people say was enough to get Linux-SWAT up in arms over it, but as you say in your needlessly vulgar way "some people are determined to be jerks no matter what."
I'm not sure how pointing out the failures of your argumentation is flaming, which seems rather non-conducive to any form of debate if it was the case. I was under the impression that involved things like calling people "asshole" and "jerks" in a vicious manner. Much like you were doing while being "nice and reasonable" particularly towards Linux-SWAT, who by that spin on the metric one assumes you have a crush on.
And then there's FUD (Fear, Uncertainty, and Doubt) apparently now means pointing out real acknowledged by manufacturer defects that were severe enough to cause them to pull the product from the market and remain unresolved. They also apparently now don't mean trying to create fear towards the viability of purchasing that which is not Intel. Or trying to allude to uncertainty regarding performance of that which is not Intel. Or to strongly express doubt regarding that which is not Intel. Or culminating all of that Fear, Uncertainty, and Doubt in campaign that's pure pro-Intel.
