Can be used and is used is not the same thing though. From an end user perspective all that matters is the actual measurable performance from actual use.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple's Next-Generation A8 Chip Said to Top 2 GHz, Remain Dual-Core
- Thread starter MacRumors
- Start date
- Sort by reaction score
Can be used and is used is not the same thing though. From an end user perspective all that matters is the actual measurable performance from actual use.
You said that the first batch only could use 4, but the latest version can use 8. That's theoretical. Software performance doesn't automatically scale with more cores, and not all software can be written to get performance enhancements from parallelization.
The first batch had a known bug that was ironed.samsung just yesterday added a 5th revision to the exynos and added there first ever built in lte modem to there exynos 5 soc.
Every octacore sold for the last year supports full 8 cores full blast when the demand is there.they only ones that had the problem where the first chrome book that used the first Gen ones.
Androids task schedule can take full advantage of 8 cores and will spread the load to all the cores.it's not like windows that needs software to be threaded for 8 cores
Last edited:
I'm not disagreeing with you. I'm just pointing out that more cores isn't automatically "better", or leads to more performance in actual use.
Apple makes their own cores, Samsung doesn't, which means that many cores is their more crude answer to get more performance. In reality, it may be close to the (mostly) meaningless megapixel race in cameras. It may look good in marketing material but the actual benefit isn't necessarily there.
I was editing mid post.On android it will get a huge performance boost.
I just scored 400 sun spider test on my octacore tab s and that is using a15 instruction set from arm and not the new 64 bit a57 set that has huge instruction per clock increase over the a15.
Once samsung goes a57 64 bit and 8 cores you will see record breaking performance boost over any other soc.there is already a leaked bench of the alleged 64 bit octacore and it's stupid fast on how hi it benched. It beat the k1 Gpu benching in 2k while the k1 was benched at 1080p
Perhaps on synthetic benchmarks, but Android isn't exempt from general principles in computer science or computation theory, I hope your realize this.
Edit: I do get that there may be benefits, just that it may not be the most efficient way to get more performance. That is, I'm not saying that there are zero benefits with more cores.
No samsung works with arm (they helped making big little) and uses reference arm designs that by nature since arm made them and there instruction sets will out do any miffed custom core like what apple has and what Qualcomm uses with krate.
Apple is using a custom core but is using arms latest a57 64 bit instruction set.when samsung decides to move to there next instruction set (arms 64 bit) and stays reference it will absolutely destroy any custom chip based off it.
Apple is just a Gen ahead with 64 bit but when everyone else goes to arm a57 you will see how far behind apples soc will be.
Exynos is always pure arm cores and has always beaten custom cores ever Gen cycle starting from the first snapdragon
A reference a57 core from arm clock for clock will destroy krate or the a7 soc from apple.I mean they are the makers lol so I think they know a few things lol
The instruction set are the same, that's why they are both ARM chips.
Not at all. There's no reason at all for this to be true my friend.
I don't have time to explain in detail but you just don't know.yes the new 64 bit is just an instruction set but the cores using that set are arm a57 cores.they are there newest Gen cores.I'm not going to get into it but there is a lot going on in the new cores.
Arm went a9 to a15 and now on a57 with major core improvements along the way.
Qualcomm in its krate core took a9 and a15 and made a custom core that was a hybrid of both those cores.apple used a custom a57 core in its a7 soc.
Samsung uses pure bread arm cores.the trade off is efficiency as the reference arm cores are made for raw performance and will kill efficiency for it and is the reason people make custom cores based of the arm reference design to fit there performance and power profile.
The a7 soc is based off the arm a57 core and nobody yet is using that core as samsung spent a lot of time working on getting a15 big little working right.
Some ARM A57 performance estimates for 20nm are already out. http://www.anandtech.com/show/7995/arm-shares-updated-cortex-a53a57-performance-expectations
Apple's almost year old custom A7 already benchmarks similar or better numbers, and you still can't buy a consumer mobile with an A57. The A8 may yet beat it to market. Most pundits suspect it has even better performance than the A7.
It's you that don't know. The remark about the instruction set referred to this:
"there instruction sets will out do any miffed custom core like what apple has"
ARMs instruction set is the instruction set Apple use, it has nothing to do with micro architecture, which is the implementation of said instruction set.
Ok, so what? I'm not disagreeing with this, but it also adds nothing to your argument.
Reference ARM cores are more likely made for general purpose use, that is they must work in many different applications, that comes with some trade offs naturally. A custom core can be customized to meet demands in a specific application to closer degree, and that may be more performance or energy efficiency or what ever requirement that application has.
Not true.
There are no general-purpose mechanisms to allow a program written for a single thread of execution to take significant advantage of multiple cores. You could spread one out to execute about 1/8th of the time on 8 different cores... but, depending on your hardware architecture, you may not want to do that. In some cases there's a significant cost to switching processing core (or CPU).
You're obviously just pulling arguments out of your... hat... hoping something will stick.
The advantages of many cores are sometimes exaggerated. It was a natural way for CPU makers to go as they had to struggle more and more to get performance increases out of a single core. But the advantage drops off pretty quickly after the second core. You just don't have more than two CPU intensive tasks to do at one time that often (especially on a mobile device). There are some notable exceptions, so if your use cases are right it may make sense to get something with four or more cores. But the reality is most people are not using most of those 8 cores virtually all the time.
Those first two cores are where almost all of your real-world computing happens (and most of that on the first core alone).
There so many errors that I don't even know where to begin. Is this a flame bait? For instance there's no such thing as "a57 64 bit instruction set" nor is the A7 or Krait's core based on the A57.
To think that you began the first post with "I don't have time to explain in detail but you just don't know." Wow.
"miffed custom core like what apple" LOL! Oh, you meant muffed. As in "muffed custom core like what apple", again LOL! Very evident that you are attempting to sound important or well versed on this matter. Please stop because you are showing quite the opposite.
Apple's Next-Generation A8 Chip Said to Top 2 GHz, Remain Dual-Core
Miffed custom core? Lol. Apple designing their own cores is a testament to their SoC design prowess. Samsung can't compete on that level so they glue together off the shelf parts.
Apple is a year ahead and it shows how advance the A7 really is. It's not Apple that needs to catch up.
Miffed custom core? Lol. Apple designing their own cores is a testament to their SoC design prowess. Samsung can't compete on that level so they glue together off the shelf parts.
Apple is a year ahead and it shows how advance the A7 really is. It's not Apple that needs to catch up.
Apple must really be magical if they can be a generation ahead AND "far behind". >_>
I hate when hardware is compared to Apple and Android. Its a bad argument. Apple designs its hardware and its operating system. "Android"(Google) does not. Samsung, HTC, Asus... They design hardware and they run Android. I think a lot of people forget that.
So yes "Android" is a great os that is later skinned and bloated by the manufacturers making it very inefficient. We have also seen a lot of benchmark cheating on the android side so I don't trust any benchmark test, show me real world results.
Apple has the advantage here because it designs the hardware, OS, and core services to run together. This is why for most of my devices I use Apple, that doesn't mean I don't love my nexus 7.
Final note: Is apple using IBMs roadmap of faster, smaller, faster, smaller now?
So yes "Android" is a great os that is later skinned and bloated by the manufacturers making it very inefficient. We have also seen a lot of benchmark cheating on the android side so I don't trust any benchmark test, show me real world results.
Apple has the advantage here because it designs the hardware, OS, and core services to run together. This is why for most of my devices I use Apple, that doesn't mean I don't love my nexus 7.
Final note: Is apple using IBMs roadmap of faster, smaller, faster, smaller now?
Back to the main article, TSMC has estimated that their 20 nm tech should allow a 30% increase in clock rate over 28 nm, while also reducing power. If one were to "turbo" clock a processor to use some of that power savings, an increase of 42% might be possible, which would provide in an increase of 1.4 GHz to 2.0 GHz... with no additional tricks. This would "just" be a "tick", using Intel's tick-tock nomenclature, for moving to the next newer fab process.
So, the open question is whether Apple is adding any more innovative tricks other than a basic "tick" update. (Perhaps more GPU/GPGPU power to support Metal performance?)
Also, the A7 has the same instruction dispatch width as Haswell. So the main differentiator seems to be that Haswell is designed for a bigger heat sink (and/or fan), thus allowing a higher power, higher bandwidth memory subsystem and semiconductor process tuning. This would imply that the A7 arm64 core is absolutely "desktop class", but that the arm64 core has been stuck inside an ASIC that has been detuned for less heat dissipation (thus allow using a much lighter and thinner battery as well). If Apple has in mind a system with a bigger heat sink, much bigger battery, and room for more memory chips, then they absolutely should be able to tape out some sort of A7/A8Turbo ASIC that would roughly match Haswell/Broadwell in performance (potentially even exceed for Apple's tuned mix of applications).
As for dual core: Given that most apps (almost all for except photo and movie editors...) use only 1 or 2 threads, using 2 faster cores will provide a speed boost to far more iOS apps than adding any additional cores. Any speed boost also allows apps to "race to sleep", which improves battery life.
Those companies selling you mobile octo-cores are mostly "innovating" by selling wasted silicon to people who are easily fooled by bigger specification numbers.
So, the open question is whether Apple is adding any more innovative tricks other than a basic "tick" update. (Perhaps more GPU/GPGPU power to support Metal performance?)
Also, the A7 has the same instruction dispatch width as Haswell. So the main differentiator seems to be that Haswell is designed for a bigger heat sink (and/or fan), thus allowing a higher power, higher bandwidth memory subsystem and semiconductor process tuning. This would imply that the A7 arm64 core is absolutely "desktop class", but that the arm64 core has been stuck inside an ASIC that has been detuned for less heat dissipation (thus allow using a much lighter and thinner battery as well). If Apple has in mind a system with a bigger heat sink, much bigger battery, and room for more memory chips, then they absolutely should be able to tape out some sort of A7/A8Turbo ASIC that would roughly match Haswell/Broadwell in performance (potentially even exceed for Apple's tuned mix of applications).
As for dual core: Given that most apps (almost all for except photo and movie editors...) use only 1 or 2 threads, using 2 faster cores will provide a speed boost to far more iOS apps than adding any additional cores. Any speed boost also allows apps to "race to sleep", which improves battery life.
Those companies selling you mobile octo-cores are mostly "innovating" by selling wasted silicon to people who are easily fooled by bigger specification numbers.
Last edited:
I know you think this is some sort of wicked burn, but Intel's single-threaded improvement at a given frequency is 25% in the six years since Nehalem shipped, about 3% improvement every year. The reason your MBA today feels faster than your MBA from 2009 is
- the 2009 model was pre-Nehalem, when Nehalem still burned too much power
- faster memory
- faster SSD
- much better turboing.
Forgetting the snark, an A7 is, at the same frequency, about the same single-threaded performance (ie IPC) as an i3/i5/i7. An MBA today sports a 1.4GHz i5, so is (single threaded IPC) the same performance as an iPad Air.
Why doesn't it feel the same speed? Primarily because the i5 is capable of turboing up to 2.4GHz, and does such a good job that anything that requires snappiness runs at that higher speed. If the A8 maintains IPC (which will surely be the case, especially since I predict the core will be largely unchanged) but runs at 2GHz much of the time (again a reasonable extrapolation from the process leap from current 28nm to the A8's 20nm) then some, but not all, of that gap is covered.
- i5 still has 2.4GHz on A8's 2GHz --- Apple needs to fix this with a PMC (power management controller) and aggressive turbo'ing circuitry, but this sort of addition to the SoC is a lower payoff than various other things they can do, so I expect it will not be part of the A8; maybe with the A9.
- i5 has hyperthreading compared to the A8. Intel's hyperthreading is worth about a quarter of a core, so the i5 is about 2.5 cores. A8 could counteract this if and when they want by adding a third core. I'm more on the fence than many people about this. For example I could see them shipping A8 and A8X cores, where the iPad A8X core comes triple core, the iPhone A8 is dual core. But, like so many theoretically reasonable ideas, you have to look at the big picture. They're moving to a new process, and they're doing everything very fast compared to the speed at which Intel moves. Two layouts and two different chips is twice as many things that can go wrong compared to one layout, and if you have to optimize, you optimize for the largest selling product.
On the third hand, with 20nm you get twice as many transistors as you had before. If you expect the same yields, and are willing to accept the same chip size, that means you can double your GPU size, improve your L3/memory controller/uncore and still have plenty of space left over to just add a third CPU, even if for thermal or power reasons you disable it on phones.
- i5 has AVX2 and so longer vectors. This is a theoretical win; it's not at all clear that it's a win that matters or that Apple cares about. My belief is that Apple sees the future in aggressively HSA terms, and so vectors should be handled on the GPU not the CPU. Apple COULD double the length of neon vectors easily enough if they cared, but they don't think it's a sensible use of resources. Personally I agree with them --- I think Intel started down a stupid and crazy path with Larrabee, and now they're trying to convince themselves there's a sensible plan here as they keep going with AVX2 and Phi, even though the direction they're going just doesn't make sense in the broad scheme of things. IMHO they'd do far better to just go with HSA, rather than being forced to copy AMD five years after the fact AGAIN.
I hate how people think the iPhone is underpowered because it's dual core...
Those two cores take up a similar amount of die space that a quad core would take up that's one of the reason why the 5S' dual core A7 smashed the quad core phones at the time.
They're essentially making two big cores rather than four small cores. Going quad core would reduce single threaded application performance.
Although the time to go quad is approaching as apps become multithreaded and the ones that are only single threaded don't need the extra performance.
Those two cores take up a similar amount of die space that a quad core would take up that's one of the reason why the 5S' dual core A7 smashed the quad core phones at the time.
They're essentially making two big cores rather than four small cores. Going quad core would reduce single threaded application performance.
Although the time to go quad is approaching as apps become multithreaded and the ones that are only single threaded don't need the extra performance.