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 Says A13 Bionic Chip Was Designed With Performance-Per-Watt Focus
- Thread starter MacRumors
- Start date
- Sort by reaction score
The chip guys need their priorities set from management. As a journalist and now in the chip biz, he sets the priorities because he presumably understands the marketplace.
For example, Apple moved to on-chip ML years ago because at some level raw compute performance isn't really all that necessary anymore. That's something that engineering probably wouldn't think about. That decision is paying off now, and will for years.
Incrementally, the next step is to have flags in Xcode so that developers can tie into the hardware better. Why not have the developer tell swift that they need the high-performance cores for this thread instead of the low-performance cores? I suspect that background apps are already foisted onto the low-power cores automagically, but why not let devs give hints directly? That's something that Apple can do because it owns the whole stack. The A series is already swift-optimized in some ways, so this is the next incremental step.
For example, Apple moved to on-chip ML years ago because at some level raw compute performance isn't really all that necessary anymore. That's something that engineering probably wouldn't think about. That decision is paying off now, and will for years.
Incrementally, the next step is to have flags in Xcode so that developers can tie into the hardware better. Why not have the developer tell swift that they need the high-performance cores for this thread instead of the low-performance cores? I suspect that background apps are already foisted onto the low-power cores automagically, but why not let devs give hints directly? That's something that Apple can do because it owns the whole stack. The A series is already swift-optimized in some ways, so this is the next incremental step.
Presumably it’s referring to the SE device. It wouldn’t makes any sense to cut down any feature on iPhone Mini maybe except the not really Pro lens.
That would definitely be fun... I think the biggest hurdle right now, once macOS supports mobile processors, would be SSD prices coming down.
You almost certainly want a 1tb drive for a desktop... assuming you are ok with 128gb for your mobile, that’s still a large leap in total capacity- and obviously, you don’t want the extra storage in a “convertible” device to raise the price to the same as buying both separately.
Way to miss the point! We are getting an important hint here.
The previous run of chips (A7..A12) had the CPU code named after winds; with A13 that changes. The point (IMHO) is that we have reset the underlying design framework, changing from the (extremely successful!) width-oriented framework that began with the A7 to a framework that is built on, and prioritizes, hundreds of power domains and hundreds of thousands of power-gating transistors. This gives Apple something that can grow even further than the previous series of cores, both downward to ever lower power, and upward to desktop and server.
Apple rarely give their plans at this sort of detail. But they do try to signal their priorities...
[automerge]1568911380[/automerge]
That’s 1TOPs LEARNING!
The A12 NPU is 5TOPs inference, bumped up to 6TOPs for A13.
I have to say - an SE/Mini with no home button but same form factor would be very compelling!
I’ve got it now...you’re the Spock from the Perpetual Piss Pot Mirror Universe and you’ve been stranded here forever due to missing the window of time the dimensional phases were aligned. Amiright?
I am certain you are right, Apple would prefer us to buy two devices... and for a while they developed the iPod along with the iPhone knowing the iPhone was and would cannibalize iPod sales. Two devices. Now, mostly the one (iPhone). The same will happen with Apple's mobile devices, they will continue to get smaller and more powerful to the point for the most part they ARE our computers and we will connect when required to different input and output (keyboard and monitor) devices.
Yes. But I don’t check them out on an apple forum.
[automerge]1568913638[/automerge]
I think it’s you that have missed the point. There are LOTS of ways you can sell your product without resorting to silly names and superlatives like all the time.
Apple’s X variant CPUs are reserved for the iPad Pro at this point and are not designed for the same efficiencies that the iPhone is with regard to battery life, meaning performance is more of the focus.
Apple designs their own CPUs and has for quite some time. They also have plenty of their own chip designers and are not “farming” it out.
If you’re referring to the 10.2” iPad that was introduced at the event, it uses the A10 Fusion instead of Apple’s brand new CPU. No company is going to use their flagship CPU in their lowest cost tablet. It goes in their flagship device and the previous flagship CPUs start migrating their way down the chain.
My point was you can't complain about Apple using two codenames in one interview when Intel uses theirs absolutely everywhere, including marketing materials and technical/variable names, names every single little piece of the system (Intel Springville, Oak Peak, Islay Canyon, Harris Harbor) and are an order of magnitude more confusing.
Last edited:
You have dramatically overestimated Apple.
It's not two years behind, and it really depends on what part of the SoC you are talking about.
The SD855 should be compared to the A12
The SD865 should be compared to the A13
Qualcomm is ahead of Apple in GPU design. The SD855+ is behind the A13 but is an older design.
Comparing the the Spectra 380 IPS to the ISP in the A12, the SD855 is able to do depth calculations and 60fps video along with there features.
But the fact remains that all chips look at PPA -> Power, Performance, Area tradeoffs and ALL mobile processors look a performance per watt. Apple isn't doing anything the test of the industry isn't also doing.
Agreed. I’m also hoping that a future Apple-designed modem chip might be as impressive as the Apple-designed A-series and show some significant gains in power efficiency. Screens are getting more efficient, A13 has got more efficient; I’m not sure how far it’s possible to go but efficiency gains from the radios and antenna would be really helpful for battery life because that stuff is in use constantly to keep the connection with the mobile masts.
You point to ARM's design - where does it say that Apple is basing their chip on it? AFAIK, Apple has been going its own way for a long time.
[automerge]1568917547[/automerge]
Both the OP and you seem to know something I don't - can you point to a link that corroborates that Apple's Bionic A13 is based on ARM's A77+A55? I was under the impression Apple had gone its own way a long time ago.
You two are just embarrassing yourself for spewing nonsense. Not only it is wildly known from die shot of every previous chips that Apple’s custom cores have always been vastly different in size and spec compare to Cortex cores. (Edit: Fundamental aspects such as decoder width and pipeline depth are completely different) It is also simply against ARM’s ToS to use or base anything off Cortex design without declaring Cortex® Branding on product website and technical documents.
ARM has two types of licenses. Cortex license and Architecture license. Architecture license allow custom design base on ARM ISA, but it must be completely design from scratch requiring no reference to Cortex design. Apple is using this type of license exclusively for many years. Other companies such as Qualcomm and Samsung use a mix of both from time to time.
Cortex® is a specific trademark for a few lines of designs sold by ARM. Cortex designs are distributed by previous mentioned Cortex license. According to Arm Branding Guidelines, Any use of such technologies require declarations of Cortex® Branding on product website and technical documents.
Sources:
ARM licenses types: https://www.anandtech.com/show/10366/arm-built-on-cortex-license
ARM Branding Guideline: https://www.arm.com/company/policies/trademarks/guidelines-brand
Cortex® Trademark: https://www.arm.com/company/policies/trademarks/arm-trademark-list/cortex-trademark
Last edited:
The new forums look nice, but as long as people like that guy are still here, it's just lipstick on a pig.
That means nothing. The layout of the exact same chip design can vary between different nodes, fabs, etc. Ask anyone in the chip design industry.
It's based largely on ARM reference design with minor customizations to things like core configuration, cache size, image processor, etc.
No it’s not and you’re still spouting this lie.
For the thousandth time:
Apples cores are 100% custom and have NOTHING in common with ARM cores other than they run the same ISA. We actually have a chip designer here on MR (@cmaier) who also agrees with this. Yet you’re going to go against him and countless others (including a chip designer who works for Anandtech who has actually written specific code to probe the inner workings of Apple processors) who also agree with this?
Your claims are beyond ridiculous.
Which is very convenient since almost everyone beside Samsung use TSMC‘s latest node every year for the past few years. It mean there are many Cortex cores build on identical nodes and fabs for comparison. Also I mentioned spec which are vastly different. Stuff like different core config and flexible caches size are part of the Cortex design and are allow to be tweaked, but key fundamental aspects such as decoder width and pipeline depth are basically fixed or very narrowly defined. Even Apple’s older cores from A10 SoC has decoder almost twice as wide as ARM’s latest A77 design (7-width vs 4-width), and deeper pipeline. They are simply not the same design. Claiming A13 is based on A77 is as laughable as trying to claim AMD’s Ryzen is based on Intel’s design. Even though they use the same ISA (whether ARMv8 or X86) they are nothing alike.
Last edited:
I heard my name used in vain.
As a chip designer, we usually divide the design process up something like this:
- instruction set architecture (i.e. opposes, operands etc)
- architecture (how many cores, what high-level blocks perform what functions, etc.)
- micro architecture (how big are the caches, how wide are the on-chip buses, how many execution units per core, is there multithreading, how big is the TLB, how big is the physical register file, how many pipe stages per instruction, etc.
- logic design (what logic gates are used to implement a particular structure)
- physical design (how are the transistors in the gates sized, where are they located, what does the metal that connects them look like, etc.)
There is some fuzziness about what goes in which tier.
But Apple’s A-series processors are similar to the “reference designs” only at the instruction set architecture level. Arguably some similarities at the architecture level. Most everything is completely custom below that level.