For $163 OEMs Can Have An 8-core ARMed CPU

According to Digitimes, that’s the going price of 8 cores of ARM.“The 8-core CPUs have been in volume production for over six months, with prices of 8-core smartphone now falling under CNY1,000 (US$163), and therefore the adoption of 8-core chips may not be able to effectively lift up smartphone prices in the end market” I remember running whole computer-labs on less CPU power (my first 30-seat LTSP lab had a single 32-bit core of Intel at 1.8 gHz). Folks are putting this in a smartphone.

Intel has responded to such pressure by supplying Atom processors at ever-lower power-consumption and prices. They are now competitive in tablets. The other half of Wintel is still floundering though trying to sell the OS for less on small devices and more for big devices. How many need big devices? Chuckle. Hint: It’s not the size of the device that matters for most of us. It’s the price. Not many care what OS runs on it as long as they can figure it out. Lock-in is almost gone for consumers and many businesses.

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology and tagged , , , . Bookmark the permalink.

34 Responses to For $163 OEMs Can Have An 8-core ARMed CPU

  1. oiaohm says:

    Robert Pogson you are correct its not a linear a relationship. The proto type produced arms at 20 and at 16nm compared to 28nm both allow the clock speed to double. 16nm more than double. When the design is the same performance is a linear relationship. Arm A57 reason at 28 mm for stopping at around 1.3-1.4 ghz is heat.

    The reason why A57 arm chips will have A53 sitting next to it is there is a problem with super-scalar designs. Light load a super-scalar has the habit of being more power hungry than a non super-scalar.

    So A53 arm is almost a Atom its missing hyperthreading. Yes a A53 arm does have performance matching Atom chips.

    This is the problem arm makers are able to put out something with super-scalar that will not destroy your battery life while idling

    Even with 4-6xsuperscaler A57 as long as load is above what requires just over 2 A53 cores to process is lower power to use the A57.

    Basically you can optimise a cpu design for performance and power usage at high load or performance and power usage at low load. Not both why not both out of order processing is costly when you only have a single item to process. Arm solution to the issue 2 core designs both with the same instruction set in the same bit of silicon.

    Yes you may see arm chips that are pure A53 but it will be really rare to see anything pure A57 due to the issue.

    Basically intel cannot just release a Xeon chip for mobile phones because if they did it battery life would be poor. Very poor. Intel would have to make a working hybrid. Problem here atom and Xeon are not like A53 and A57 that have identical instruction set functionality.

    There are still scheduler issues with A57 and A53 in the same chip but this is OS issues.

    Arm has done something very special.

  2. dugman wrote, “you have to purchase additional software to make it work correctly”.

    That’s a bizarre concept in a free market but just another form of lock-in in a monopoly. Fortunately many consumers now get it. They can buy a smartphone and forget about all the hoops M$ made its slaves pass through.

  3. dougman says:

    Windows, so good, you have to purchase additional software to make it work correctly. Now that’s optimization!!

  4. oiaohm wrote, “Then you have to remember as Arm reduces its nm to 20 and under it clock speed will double.”

    It’s not that linear a relationship. Also, CPUs are mostly idling. Most of the cores are shut down except for brief bursts. The extra cores and higher clockspeeds are mostly selling points. “Ours is faster than theirs” works in the market but the current products are insanely fast and definitely good enough for consumers. My smartphone is ancient and I have no problems using it as a general-purpose PC. FTP is just as fast as the network allows. GPS is something I don’t have on a legacy PC. Browsing is crisp too and my smartphone is single core 32bits, 1 gHz, 512MB… I am the bottleneck with my smartphone. It’s about the same power that was amazing when P3s stomped across the planet. No one complained they were slow. M$ complained that they were good enough to prevent consumers replacing them so they had to manipulate the market into replacing them by making software incompatible with old hardware.

  5. Deaf Spy wrote, ” Please enlighten us, Mr. Pogson. What sensible human being would spend the same amount of money for an inferior product?”

    The folks who found only that other OS on retail shelves, mostly. They don’t even know they have a choice on legacy PCs still. Fortunately the small cheap computers are now selling good software at great prices.

  6. Deaf Spy wrote, ” Companies which make their money from selling software, try to make sure their product behaves properly, and efficiently. These companies have their developers optimize.”

    Ah! That’s why Lose ’95 had a mean lifetime of a few hours and why I moved myself and my students to GNU/Linux which ran like the Energizer Bunny. The truth is Lose ‘9x shipped with ~50K bugs because no matter what the developers told the salesmen, the salesmen were in charge and they had promised certain delivery dates to OEMs.

  7. oiaohm says:

    Deaf Spy if you think developers make sure product behaves properly, and efficiently you are kidding me.

    Yes kurkosdr yelling over here is fully caused by software developers not following API instructions. There was a optimisation the developers took it. Is the optimisation safe or sane does not cross developers minds at times.

    These companies have their developers optimize.
    This is the correct statement. Problem is developers optimising are not always sane.

    Also optimise for the target hardware is also correct. With x86 this has changed over time. Older programs not allocating memory before using can in fact be a performance optimisation on x86. Some older x86 cpus if the memory was allocated would copy the allocated memory into the cpu before performing operations on it even if the result would be a straight over write. So yes geekbench 2 is faster on old x86 processes due to the so called bug. In fact benchmarking for cpu handling of unallocated and allocated operations are interesting to know if it causes the cpu to jack knife. Just to be fun a Xeon Phi likes unallocated yet entry level atoms don’t. X86 cpus in the middle are mixed between generations what way is the fastest. Hello stuffed developers trying to optimise generically. Of course you miss anandtech uses geekbench and other benches. Also has shown the same horible trend for x86. In int and floating point x86 is not keeping up with armv8 chips. Entry level chips are not competition.

    Why are armv8 chips clocked so low. They have to run fan and heatsink less. So as the nm goes down and the heat production reduces the arm chip clock speeds will just keep on going up at this stage. There is a practical limit around 4.5Ghz. x86 chips got to the practical clock limit first.

    Super-scalar limit is about 8 instructions/theads per clock cycle core this is found in IBM POWER8. So arm still can expand a little. This is the problem Xeon Phi is not the best Super-scalar processor in existence. Just Power 8 chips are not cheap due to lack of production.

    Armv8 is brining a lot of high end techs to low end hardware. AMD armv8 chips are going to be interesting. At long last arm chips with heatsink and fan and able to have clock speed pushed. Remember 2.6-2.8Ghz will be able to be done without a fan or heatsink. We are going to see armv8 chips pushing the 4Ghz+ range no problems. Next 12 intel and arm will become more level. Items that have restricted arm will not be in place.

  8. Deaf Spy says:

    Mr. Pogson speaks of software… Companies which make their money from selling software, try to make sure their product behaves properly, and efficiently. These companies have their developers optimize. Yes, optimize. Make sure you squeeze all the performance you can get from the hardware you target.

    GeekBenchmark is obviously sloppy development. I don’t blame the developers, mind you. When you go multi-platform, you always target the lowest common denominator. You get all the drawbacks, and lose all the benefits. But such a benchmark is meaningless. That is why serious analysts like Anandtech (who are very pro-android) never compare apples with oranges.

  9. Deaf Spy says:

    Mr. Pogson clutching at straws… Please enlighten us, Mr. Pogson. What sensible human being would spend the same amount of money for an inferior product?

  10. oiaohm says:

    The scary part is an Apple A7 is 28nm. In scale of production its huge and should be insanely power hungry. A8 for apple devices at end of year double the clock speed as they reduce nano meter.

    This is the problem. Armv8 chips are not slow that slow now. 1.3-4 Ghz is a A7 and a A8 is 2.6-2.8Ghz. Armv8 first over doubled performance due to design changes. Now will double again due to nm changes. Yes x86 was already losing in some benchmarking.

    Those apple A7 benchmarks have a 28 nm chips.

    This is the problem armv8 is fairly competitive with a x86 at 28 nm or a very long way behind in nm compared to intels 14nm. Then you have to remember as Arm reduces its nm to 20 and under it clock speed will double.

  11. oiaohm wrote, “The reality intel does not have the raw power of the new arm chips in the low end x86.”

    Intel is barely competitive but only because it does not charge its usual monopoly prices down there. I have two Atoms. One is OK but the older one is pathetic. The newer ones should be pretty good now that Intel uses 14nm but they aren’t power-houses. It’s informative that ARM at 22nm is competitive with Intel at 14nm. They are both approaching the point where power-consumption is OK but Intel still costs more: silicon, fiddling fabrication, and marginal volume. ARM is shipping ten times as many CPUs as Intel, which is good. That prevents Intel from monopolizing the market.

  12. oiaohm says:

    This is armv8 vs the existing arm 32bit chips. Duel core armv8 A57 with lower clock-speed beats the existing quad core 32 bit A17. Something nasty is that apple production A57 are 6 way super-scalar. So 12 threads per cycle in a dual cpu get processed.

    Remember Cortex A57 is smaller on silicon than a Cortex A17.

    So in one development cycle arm chips have in fact more than doubled their performance.

    Reality is you are looking quad core something x86 to stand a chance against these new arm chips in quad core. Mostly likely Xeon.

    Intel is not talking raw power in low end chips any more but how power effective. The reality intel does not have the raw power of the new arm chips in the low end x86.

  13. oiaohm says:

    Deaf Spy using unallocated memory is a defect a lot of Windows applications suffer from. Sorry real world application performance testing includes testing with common coder errors that the cpu does not reject.

    Mind you its not only geeksbench where intel has problems. Quadrant and Linpack also show a major advantage to the Arm. Those don’t have the unallocated memory issue. Geeksbench is not a large out abnormality even including the bug The thread stall that accessing the unallocated in Geeksbench 2 make the work load more real world ish.

    Deaf Spy really sometimes it pays to sit back and wait. Sorry googing your way out of this I will kick you teeth in. The reality is you have been wrong.

    The more threading and cpu focused the test is the better the powerful arm chips do compared to anything bar a Xeon Phi in the x86 line. Yes and the cpu number difference relates directly to the fact the arm cores are handling more threads than the non Xeon x86 chips. Intel does not bench atoms against arm chips in server workloads that are heavily cpu bound. Instead they put arm head to head with the Xeon.

    Super-scalar method of instruction handling is just many many times better than hyper-threading.

    Yes the older 32 bit Cortex A17 beats your general x86 chips when it comes to data processing. Cortex A17 are locked to one endlan mode only same way as x86 most common.

    The duel endlian is a new feature of the armv8 chips. Equal super-scalar to Xeon chips is also on armv8. Higher clockspeeds also in armv8 chips.

    Remember the armv8 chips are a lot cheaper than Xeon chips.

  14. dougman says:

    LOL, for $250 I have a safe and secure device free of malware that is always up to date.

    Did I mention that it boots fasts and will last me 10 years?

    The Windows device is good for maybe years at best and regarding the malware thing, well, we all know that Windows = Malware.

  15. Deaf Spy raved about a Wintel device…

    Yep reduced $100 from $350 because it wasn’t selling. Belongs in the “clearance” bin. I quoted similar products from the same supplier. BestBuy sells Exynos Chromebooks for $200-$247 so my observation applies to BestBuy too.

  16. Deaf Spy pointed me to stuff I already knew about…

    “Two of the Geekbench 2 floating point workloads (Sharpen Image and Blur Image) have a fencepost error.”

    That’s just like real software then, eh? That’s rich, M$’s “friends” complaining about unfairness. Where were they when M$ made certain applications run poorly on other DOS? Provided faster APIs to its own software so that competitors would seem slow? Told the world that FLOSS cost more? Told the world that Linux violated unspecified patents? etc. Why didn’t they complain about unfairness then?

  17. dougman says:

    Re: “The Office license is life-time.”

    As MicroSh1t, does not make software that lasts a lifetime, you must be either lying or using pirated software, perhaps even both.

    M$ software has a lifecycle, then it is killed off. DO you seriously think that they would let something work forever?? M$ wants you to upgrade software every 3-years.

  18. Deaf Spy says:

    Thanks for the numbers, Mr. Pogson, but I expected a better research from a scientist like the one you claim to be:

  19. Deaf Spy says:

    Mr. Pogson, your searching skills are failing. Take a look:$pcmcat247400050000&cp=1&lp=3

    Touch screen, 4GB RAM, 500GB HDD – $250.

    For the same money, you can get a Chromebook, which has 2 GB RAM and only 16GB local storage.


  20. oiaohm says:

    Deaf Spy geeksbench includes some network traffic handling testing. Differences have shown up.

    Atoms rarely beat any powerful ARM chips in any bench. i3 and i5 min to out bench large percentage of the older 32 bit arm chips.

    Usage of arm by HP was delayed due to the major differences in performance being 64 bit did compared to 32 bit. Even so arm 32 bit chips put up a hell of a fight against systems running 64 bit x86.

    The new armv8 chips the A57 type not the A53 type have a few new tricks and a few old ones. superscalar in a armv8 does like hyperthreading. Armv8 is 4 threads per core compare to hyperthreading 2 threads. The other thing superscalar starts becoming interesting. The fact that the 4 threads per core can in fact be processing instructions per clock cycle.

    “quad-issue superscalar pipelines” Yes this is arm solution instead of hyper-threading. The thing here is “superscalar pipelines” are not new to armv8. Every arm chip to give intel chips a benchmark beating in different areas has it. Cortex-A15 has it but its limited to only 3 way.

    Now you see arm 32 bit chips fall behind in workloads that give advantage to 64 bit as well.

    This is the problem armv8 is truly able to give x86 chips a true run for money.

    Exynos 5 Octa for example is in fact able to handle at peek instructions from 16 threads at the same time. 4×3 giving 12 for the big processors and 4×1 in the little processors.

    The newer big processors 4×4 so 16 just in the big. Exactly what x86 are you looking at for this Xeon Phi. Intel does make a “quad-issue superscalar pipelines” processor. Its not the atoms or the i3 or i5 chips. Xeon Phi also don’t have hyper-threading.

    Yes the chips that have been giving arm chips the biggest benchmark trouble have been Xeon Phi yes the x86 chips with superscalar pipelines. Problem is compare to 32 bit arm that only have x3 Xeon Phi had x4.

    Armv8 sees arm chip draw level in lot of base level tech. This is the problem arm is putting Xeon Phi level tech in phones and tablets.

  21. Deaf Spy wrote, “Let it be Atom, not i5.”

    I have two systems with Atoms. The older/slower one definitely is slower than some smartphones struggling barely to render video with certain codecs. It’s sleeping at the moment or I would give you some specs. One can just line up the MIPS: Exynos 5 at 1.6gHz clearly beats Atom with dual cores at 1.6gHz.

    Here are some benchmarks:
    Intel Atom Z2760 at 1.8gHz = 1355 Geek Bench 2
    Exynos 5 Octa 5410 at 1.6gHz = 3596 Geek Bench 2
    Intel Core i3-3217U at 1.8 gHz = 3590 / 4617 Geek Bench 2
    NVIDIA Tegra 4 at 1.9gHz = 4493 Geek Bench 2 competes well with
    Intel Core i5-3317U at 1.7gHz = 4693 / 6497 Geek Bench 2

    So, there you have it. Too bad there’s not price/performance numbers…

    “The i5-3317U is used in the Microsoft Surface Pro.” $349 here.

    Meanwhile, “The Exynos 5250 is used in the Samsung Chromebook and the Nexus 10 tablet.” which sell here for $270 and $347. Of course, M$ doesn’t charge itself for use of its own software but its device is ~$70 more than the ARMed thingy.

  22. Deaf Spy wrote, of my assertion that x86 costs less, “Judging from the prices of laptops and hybrids, you cannot say so.”
    Low-end laptop:
    Wintel = $470
    HP Chromebook with Exynos 5 = $330.

    Note that is the only ~12 inch Wintel notebook HP sells in Canada. They can’t even try to compete against */Linux on ARM otherwise.

    The lowest price of a dual-core Intel processor at my favourite supplier is $65. The cheapest Core i3 is twice that. Most others are $300+ Meanwhile folks can buy good ARMed processors for peanuts and they include almost everything on the chip whereas Intel requires additional components.

  23. Deaf Spy says:

    Mr. Posgon wrote: “x86 setups cost more…”

    Do they, Mr. Pogson? Judging from the prices of laptops and hybrids, you cannot say so. See Ivan’s example.

  24. Deaf Spy says:

    Doug, I own such a device. The Office license is life-time. Not trial. Nothing new, as usual, you don’t have a clue what you are talking about. But you please Mr. Pogson.

  25. Deaf Spy says:

    Ohio, would you care to show me a single test in which ARM blows i5 out of the water? Let it be Atom, not i5. You will get bonus points if the test involves these Big-endian-based protocols.
    Because, you know, endianness if often handled by the hardware servicing that protocol. But don’t let this stop you, just show me a test.

  26. dougman wrote, “M$ will not survive in the long-term, unless they adopt Linux in some manner”.

    They are much more likely to adopt a BSDish OS as did Apple. The licence suits their style and they can hire programmers to polish the GUI rather than worry about security and performance of the underlying software. They might use GNU/Linux on internal servers for commodity hardware. It will be interesting to see how the world of peripherals responds. Will they produce BSDish drivers and make BSD a “new Linux” or will M$’s universe of “partners” contract? It is even possible they could merge the OS-part of their operations with Apple. They can only do that if GNU/Linux is more widely accepted on the desktop because of anti-trust law… It is possible for them to ship a GNU/Linux OS with all the applications going through some non-Free API, Software as a Service, so to speak. That’s more or less what Android/Linux does with their BSDish code layers.

  27. dougman says:

    ON a tangent, but related nonetheless.

    Read Moore’s law: computer chip density doubles every 18 months. That was in 1965, when Gordon Moore of Intel made this observation. Today, it’s close to every 12 months.

    I take seriously Ray Kurzweil’s estimates on information costs. His article on the law of accelerating returns (2001) is a classic. It has influenced my thinking.

    Moore’s law is accelerating. Kurzweil wrote this in 2001: “In line with my earlier predictions, supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing. By 2050, $1000 of computing will equal the processing power of all human brains on Earth.”

    M$ will not survive in the long-term, unless they adopt Linux in some manner, that is what I expect them to do sometime down the road.

  28. oiaohm says:

    Ivan you need to read the price again.

    The 8-core CPUs have been in volume production for over six months, with prices of 8-core smartphone now falling under CNY1,000 (US$163)

    That is $163 for a full phone including at least 16GB of solid state storage and screen. Most are only 32 bit.

    Mind you the does have issues. Its only A53 processors and does not support USB 3.0 and a video card is not 4K. So you would say that is a phone.

    Really you would use the 810 snapdragon in tablets.

    Allwinner has Oct 32 bit arm out and will have Oct 64bit arm in 2015. Same with a lot of other soc makers. So prices are going to stay low.

  29. dougman says:

    Ivan meant, “M$ Windows, and it’s retarded bullshit”

    Office preinstalled, that is the ONLY way that M$ gets people to use it’s 60-day trialware. Been using LibreOffice, Scrivener and Writebox for quite a while now and have not looked back.

  30. Deaf Spy wrote, ” For the same money OEMs can buy a 4-core i5 will blow ARM out of the water in performance anytime”.

    As you know, the first release of cutting edge CPUs is accompanied by high prices. Remember the $1000 AMD64’s? You can now buy them for less than $100. The 8-core ARMed CPUs will come down quite a bit in a year making Deaf Spy’s observation irrelevant. The fact is that Intel CPUs cost more to make and Intel charges huge margins compared to ARMed run-of-the-mill CPUs. Then there’s the cost of PSU and the motherboard. x86 setups cost more than double the cost of ARMed setups.

  31. Ivan wrote, “I can get a brand new two-in-one with Office preinstalled that comes with the benefit of not having to deal with linux and it’s retarded bullshit.”

    Will it fit in a smartphone or a pocket? What BS? Why won’t M$’s software run on that CPU? It was just a few years ago, folks were telling us that GNU/Linux had a hardware problem. Was that just BS? Run LibreOffice and leave M$’s BS behind, where it belongs.

  32. Ivan says:

    For double the cost of just that CPU I can get a brand new two-in-one with Office preinstalled that comes with the benefit of not having to deal with linux and it’s retarded bullshit.

  33. oiaohm says:

    Deaf Spy it depends on what you are benching. The Internet by all stupidity has a lot of protocols that are big endian and x86 is little endian. Guess what arm64 is. The answer is in fact both. So x86 blow arm64 out the water it depends on the workload. Network traffic processing give to a arm chip over a x86 chip any day.

    Arm is one of the odd designs that you can have a 2 int both in different sections of ram both different endian and do a single instruction compare between the two giving the correct result even addition or subtraction. Due to duel nature being in the core of arm there is almost no overhead in case of changing between endian.

    Arm has some unique tricks. Its always interesting that Intel avoids all the benchmarks that play into Arm strengthens.

    Some ways in a server the best solution will be a mix of x86 and arm as both chips do have particular strengths in their instruction sets and design.

  34. Deaf Spy says:

    So what. For the same money OEMs can buy a 4-core i5 will blow ARM out of the water in performance anytime. And have x86 compatibility.

    Unless you need a low-power server, but then for 50USD you can have a 4-core Atom that will still blow ARM out of the water, and still have x86 compatibility.

    Intel is all good, and shows no signs of decline.

Leave a Reply