Slashdot Mirror
Intel Challenges ARM On Power Consumption... And Ties
GhostX9 writes "Tom's Hardware just published a detailed look at the Intel Atom Z2760 in the Acer Iconia W510 and compared it to the NVIDIA Tegra 3 in the Microsoft Surface. They break it down and demonstrate how the full Windows 8 tablet outperforms the Windows RT machine in power consumption. They break down power consumption to include the role of the CPU, GPU, memory controller and display. Anandtech is also reporting similar findings, but only reports CPU and GPU utilization."
Despite repeated claims that x86 is beating ARM here, they look neck and neck. Assuming you can make a meaningful comparison.
Despite repeated claims that x86 is beating ARM here, they look neck in neck.
It's neck and neck.
It's "neck and neck" as in a pair of horses very close together at the finish line.
Sigh
Oh for crying out loud: Neck and Neck.
Often used when describing two racers that are nearly even in position.
No brain, no pain.
If two processors are Neck and Neck in power consumption and one of them is x86. It means x86 is ahead. It's got better clock speeds and it's got more software going for it than arm. Yes we have a lot of android apps, but I would rather have my windows applications to those "apps" and their private internet. Unless Neck and Neck is for a processor intel does not produce any more, it's clearly advantage intel.
Even if true (watch out for cognitive dissonsoance with respect to Intel power efficiency claims) it does not mean a thing if Intel cannot match the price. Currently something like $1 goes to ARM holdings per chip. Lets see a bloated old monopolist get by on that.
When all you have is a hammer, every problem starts to look like a thumb.
http://www.tomshardware.com/news/intel-arm-processor-soc-atom,17476.html When that story was posted i said that all ARM was doing was poking the bear. Didn't take long for Intel to get there either. Just shows you don't piss off a company with a lot of $ for R&D
First, those articles are very interesting, thanks to Intel for making them happen.
Second, it's a good thing that Intel is catching up. I'm not a great Intel fan (rooting for the underdogs and all that), but still, I'm impressed.
Third, isn't the OS choice biasing the results a bit ? Would ARM fare better under a more ARM-oriented OS such as Android ? Or is power consumption profile, in the end, fully OS-independent ?
The Cloud - because you don't care if your apps and data are up in the air.
The only issue here is that this is not an apples for apples comparison. 40nm vs. 32nm should give a huge benefit to the 32nm Atom. We need to compare the same technology node for this to make any sense. Also, looking at the idle cpu power consumption from the anandtech article, the Atom SOC used 10x more power. /half speed or with the SOC sitting idle?
So the real question is what do most tablets spend the majority of their time doing? Running a benchmark at full
Example numbers: ARM CPU 0.0038 W vs.. Atom 0.02.
NVidia GPU 0.21 W vs. Imagination 0.11 W
The part that wins isn't from Intel, and it is available for ARM and it probably is the part that would lose badly in any benchmark.
Yay for biased benchmarking.
So far Intel wins by undersizing the GPU.
I have said it before: with ARM, you can choose from multiple, competing chip vendors, or you can license the ARM technology yourself and make your own chips if you are big enough; with x86, you would be chaining yourself to Intel and hoping they treat you well. So, if low-power x86 is neck and neck with ARM, that's not good enough.
Intel is used to high margins on CPUs, much higher than ARM chip makers collect. Intel won't want to give up on collecting those high margins. If Intel can get the market hooked on their chips, they will then ratchet up the margins just as high as they think they can.
The companies making mobile products know this, and will not lightly tie themselves to Intel. So long as ARM is viable, Intel is fighting an uphill battle.
lf(1): it's like ls(1) but sorts filenames by extension, tersely
Interesting that they are not comparing to a *modern* ARM chip (Cortex-A15), like the Exynos 5 (5250) or even a Qualcom Krait S4 (perhaps MSM8960).
So the news is that Intel has mostly caught up to an old ARM based chip based on designs/specs years older still and only running under MS-Windows. Yawn....
http://www.tomshardware.com/reviews/snapdragon-s4-pro-apq8064-msm8960t,3291-4.html
Atom isn't here, but perhaps because it is too new, but it's clear from this graph that at least Tom's Hardware seems to agree that the Snapdragon eats Tegra's lunch.
I have a Nexus 4 (Snapdragon S4) and a Nexus 7 (Tegra 3), and the 4 is WAY, WAY faster than the 7 in almost every experience.
On the Nexus 4 I can leave a movie playing in the background and keep listening to it while I check an important email that just came in or make a move in a game of Words with my wife. Attempting the exact same thing on the Nexus 7 results in the movie skipping and the user experience slowing to a crawl.
Perhaps there are some significant architecture differences between the two, but at least from a real-world user experience standpoint, I would not characterize the OP's assertion as "random conjecture" at all.
Arm draws 10% of the power of Atom at idle, and Android runs mostly at idle even when you're using it to do stuff because its designed from day one that way. Windows uses a lot more processing power, and 'idle' on those Windows, literally means not using it at all, and even when you're not using it, the Atom is still drawing > 1W.
My God, I'm an insensitive clod... sorry mate.
Why are they a node ahead all the time? Because they spend billions in R&D. When the downturn hit everyone in the fab business cut R&D, except Intel. So now they have a 22nm fab that has been running for awhile, another that just came fully online, and two 14nm fabs that'll be done soon (one on 450mm wafers).
They do precisely what geeks harp on companies to do: Invest money in R&D, invest in tech. They also don't outsource production, they own their own fabs and make their own chips. Most of them are even in the United States (8 of the 11).
The payoff is that they are ahead of people in terms of node size, and that their yields tend to be good (because the designers and fab people can work closely).
If other companies don't like it, well the only option is to throw in heavy on the R&D front. In ARM's case being not only fabless but actually chipless, just licensing cores to other companies, they can't do that. They are at the mercy of Samsung, TSMC, Global Foundries, and so on.
One area in which Intel is significantly more open than any manufacturer in the ARM ecosystem is in graphics hardware. Although Intel hasn't opened all their GPUs fully yet (from what I've read), this seems to be mostly because providing all the documentation takes time, not because they are against making everything open.
This contrasts dramatically with every single ARM license in existence. ARM's own MALI GPU is tightly closed (probably because MALI was a licensed technology) so the Lima team is having to reverse engineer a Linux driver. All the ARM licensees who provide GPUs seem to be either unable to open their GPU information because their GPU core has been licensed from a 3rd party, or else are simply disinterested in doing so, or else vehemently opposed to it for alleged commercial reasons in at least a couple of cases. So, the prospect of open documentation on SoC GPUs appearing from ARM manufacturers is vanishingly small.
This gives Intel at least one possible opening through which they can be fairly certain that the competition will not follow. Although that may be worth a lot to us in this community, the commercial payback from community support tends to be very slow in coming. Still, it's something that Intel might consider an advantage worth seizing in the mobile race where they're a rank outsider.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
That's probably a combination of the piss-poor GPU on Tegra 3 (barely good enough to render one thing at-a-time, and you expect stutter-free multitasking?) Along with the pathetic memory bandwidth (DDR3, but only a 32-bit bus).
Snapdragon S4 has nether of these issues!
Man is the animal that laughs.
And occasionally whores for Karma.
One thing to keep in mind is that the ARM is much more general purpose while the Intel chips tend to have a more complex assembly instruction set. So for adding one number to another (x=y+z) I suspect the simpler ARM architecture is going to win on power consumption. But many Intel chips have assembly instructions specifically for crazy things like AES encryption. This is used as the basis of many encryption protocols, hashing, and random number generation. So if a machine is basically serving up all encrypted data then it is possible that an Intel chip will be much faster and consume much less power while performing these operations. Depending on whether he software will take advantage
So I thing this is a case where you really have to look at the significantly broken down performance results to see if your use case fits one chip better than the other. A normal consumer example would be if your OS is encrypting your file system and using these cool Intel instructions. I suspect that it would then be a night and day difference in battery drain. But the drag is that you probably have to pretty well buy a device with both chips, set up your standard configuration, and then test it out. This is generally only something an IT person about to provision a department might be expected to do.
I guess that the overall benchmark is all we really have to go by which really doesn't tell the whole story.
I dont give a flying crap how much juice it sucks just give me 75 gigahertz CPU and a damn drive that can keep up.
Oh and make it AMD prices not intel.
I suspect you're in the minority here (as in wanting power regardless of power consumption). For me, desktop (and laptop) processors became fast enough about 5 years ago. Probably more. The laptop i'm using now is about 5 years old and any performance problems it has aren't CPU related. A hard drive that can keep up with my 1.8Ghz CPU would be nice - something that could keep your proposed 75GHz running without waiting would be just a little awesome :)
The Tegra GPU eats the S4 GPU. Don't make uninformed claims. The difference is that the Tegra, like it's predecessor, needs GPU code to turn it on because it's such a massive generator of heat that all cores except one are disabled during normal use (it has 16 Unified Shaded cores iirc). No benchmark has currently paid Nvidia the fees to get the Nvidia SDK to make a Tegra benchmark so far. I doubt it's cheap.
Now go find a Tegra HD game and gawp.