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 A14 Chip Rumored to Become First Arm-Based Mobile Processor to Exceed 3GHz
- Thread starter MacRumors
- Start date
- Sort by reaction score
Imagine a possible future when Final Cut Express for iPadOS allows you to add your iPhone 12 to a list of network renderers for your A14X iPad Pro and with both devices using 802.11AX, making it feasible to actually dole out the work to another device wirelessly. Science fiction? Possibly.
No, it’s a ridiculous flaw of physics. Keeping things in memory means that memory must be refreshed, because the memory is dynamic RAM. Refreshing memory requires power, and only horribly inefficient designs, like what Samsung does, would include more memory than is necessary. Thankfully, unlike the terrible engineers at Samsung, Apple’s engineers chose to make a much smarter design decision.
[automerge]1584369207[/automerge]
Not really science fiction, but not a good idea, either. If the phone is on battery, that would obviously kill it pretty quick. And if not, it would still be pretty inefficient because the phone thermal solution will always require extensive throttling. So the iphone wouldn‘t contribute all that much vs. what the iPad Pro is already doing.
Fair enough...I’m projecting my own desires into a scenario that looks good on the outside, but may not be on the inside.
Still valid, that is why I said L3 and L4 caches. L3 is SRAM, L4 is DRAM. I was hesitant to get into too many technical details.
Note that I also mentioned increasing the number of cores, either CPU or GPU. The L4 Cache can lay outskirt from the cores, which should be alined like a loop. Inside the loop, we put the L3 cache and other small supporting blocks such as hardware decoder, motion co-processor, etc. Between the cores, we put the L2 cache. Within the core, we augment the L1-Instruction cache and L1-Data cache, which are private to the individual cores, and only addressable by the corresponding cores.
The compute, decode, and graphics cores should share the L4 Cache. via the loop-shaped bus. The L3 cache should only be addressable by cores within the same efficiency zone, which shares the same variable clock speed. those L3 Cache should be grouped into blocks, which can then be assigned to efficiency zones based on computational needs.
Therefore, L1 and L2 caches are core addressable. L3 cache is zone addressable. L4 cache is globally addressable.
The die should be significantly bigger than the current CPU architecture equivalent, but still smaller than the combined size of the typical CPU and RAM silicon modules. It should be significantly more performant than the equivalent, especially at sustained load.
Around the die, 4 soldered-on PCI-e NVME silicons should be directly connected to the bus as L5 Express storage. On the two diagonal corners, there should be two encryption Co-processors, each controlling two NVME silicons, the L6 storage and below, and the motherboard firmware. The other two edges should have, two dedicated neuro-network processors.
There has to be a balance between what is a reasonable amount of DRAM in a device, the user’s needs (NOT their wants and desires, which engineering can never satisfy), power draw, et al. Exactly what @cmaier is describing. Samsung putting 16GB in a phone is meant to look good on a marketing tear sheet, woo the spec hounds and compensate for any deficiencies in Android, Samsung’s own shortfalls, etc.
Makes me think of the 3GHz G5 promise back in the day. (Not that Apple promised a 3GHx A14 but it brings that to mind.)
Ill bet nobody can prove their asertions, although the OP used to design cpus so he might know a thing or two about this. Edit: correct autocorrect
Last edited:
False statistical assumption. The huge apps that most people don't restart most of the time can use more energy from near constant dynamic memory refresh, than the energy occasionally used to read them back in from static storage.
[automerge]1584380339[/automerge]
Or you can rent a 64-vcore ARM processor from Amazon. By the hour. Already.
[automerge]1584380677[/automerge]
Yup. It's bored gen-X-ers who give up on any web site or app that takes longer than a few hundred milliseconds to load, run, or render. Apple needs to give them multicores with 3GHz+ to keep them from swiping to the next thing, or driving their car into a light pole.
Last edited:
https://gall.dcinside.com/board/view/?id=smartphone&no=5859372
https://gall.dcinside.com/board/view/?id=smartphone&no=5859992
The geekbench score for A14 is FAKE! Someone from South Korea confessed that he made a fake score.
https://gall.dcinside.com/board/view/?id=smartphone&no=5859992
The geekbench score for A14 is FAKE! Someone from South Korea confessed that he made a fake score.
They’re not necessary, but as can be seen with async/await, making concurrency and parallelism easier to write can go a long way towards making it more common.
You said 16 cores — that’s useless with the status quo of “concurrency works OK”. Ten of those will be bored much of the time.
(Maybe Apple should look into something akin Turbo Boost?)
Sure. My main point is that concurrency is as ”easy” (or hard) on Arm multi-core as it is on anything else.
Android phones have more memory largely because most apps are Java, ergo JIT, and ergo require more RAM.
Bad decisions beget bad decisions. JIT begets need for faster processors (without ability to deliver them) and more RAM, which begets need for more battery, etc. etc.
If A14 Geekbench scores have started to leak, it won't be long before we get some leaks of the future Laptop ARM CPUs too. I am really curious to see what that's going to look like from a performance viewpoint.
Apple's unreleased A14 chip is rumored to be the first Arm-based mobile processor to officially exceed 3GHz, according to a new report by Research Snipers.
Apple's A14 processor, the successor to the A13 chip in both the iPhone 11 and iPhone 11 Pro, is expected to debut this fall in Apple's "iPhone 12" models. The report highlights the suspected Geekbench 4 score of the A14 chip, with a frequency reaching 3.1GHz. This would be 400MHz higher than Apple's current A13 Bionic chip with a frequency of 2.7GHz.
At such a frequency, the chip's Geekbench 5 running points have surged. The report mentions that the A14's single-core performance shows a score of 1658 (up 25% from the A13), and a multi-core score of 4612 points (up 33% from the A13). The extra processing power will be helpful in running simultaneous workflows, navigating through apps, and more.
Apple chipmaker TSMC is expected to ramp up production of Apple's 5nm-based A14 chipsets in as early as April of this year.
In addition to the A14 chip, rumors recently have mentioned both Arm processors with Mac Pro level performance and a Mac with an Apple-designed Arm processor are in the works.
Article Link: Apple's A14 Chip Rumored to Become First Arm-Based Mobile Processor to Exceed 3GHz
I am planning on replacing my MBP13 very soon so I probably won't wait until then but it's interesting nonetheless.
I'd rather have more DRAM than more MHz. My Apple mobile devices that have become prematurely obsolete were due to DRAM starvation rather MHz starvation. Apple knows this that's why they market MHz while keeping DRAM minimal so people will upgrade more often.
Classic example is while the iPhone 4S had 512MB DRAM that reloaded apps like crazy the competion, Samsung, had 2GB DRAM (4X more) which allowed more advance OS features like split screen multitasking, PiP, etc. and is still usable to this day. iPhones now have 4GB while competition has 16GB (4X more).
I'd take 4X more DRAM future proofing over a minor MHz bump. Arguing against more DRAM is like arguing against more pay. Who does that?
Classic example is while the iPhone 4S had 512MB DRAM that reloaded apps like crazy the competion, Samsung, had 2GB DRAM (4X more) which allowed more advance OS features like split screen multitasking, PiP, etc. and is still usable to this day. iPhones now have 4GB while competition has 16GB (4X more).
I'd take 4X more DRAM future proofing over a minor MHz bump. Arguing against more DRAM is like arguing against more pay. Who does that?
Last edited:
Not science fiction at all insofar as that sort of thing has been going on with 'regular' computers for years, although there's a slight matter of network bandwidth if you want to do it wirelessly.
The question is, why you'd want to use an iPhone of all things (expensive, redundant display, limited battery power, extremely thermally limited) as a 'remote' render engine when (say) a Mac Mini, a cheap-as-chips generic PC or even a rack of 20 Raspberry Pis would run rings around it for less money.
It was hypothetical as much as anything...imagine being out with just your iPad Pro, iPhone and, possibly, your video camera/DSLR. You’re done with your shoot and you’re already editing the video on your iPad Pro in Final Cut Express (or whatever Apple ends up calling it) or in LumaFusion (or Premiere Rush, blech!) and you need a little extra oomph for the render portion. Turn on Wi-Fi, add the iPhone as a network render node and continue working.
Your examples assume that I am going back to an office where I have access to that sort of kit when in reality I may be cutting in the field that is remote enough that I have cellular coverage, but not at all convenient to a desktop. I have my iPhone with me at all times, so I can put it to work.
We’re past the point where journalists, videographers, YouTubers, et al. should be concerned with getting back to an “office” to cut video, title it, narrate it, upload it and move on to the next project.
I think it’s a bit more of cause and effect.
Apple can optimise their OS better to take advantage of less ram, and I suspect they do this to justify including less ram in their devices, which in turn lets them cut down on costs.
I have seen the speedtests and acknowledge that the iphone pro max could probably do with a little more ram, but the reality is that most people are not constantly opening and closing huge apps.
Conversely, android OEMs have no such control over their software, and thus rely on increasing ram to get around any inefficiencies in the OS.
Just different ways of skinning the same cat.
https://gall.dcinside.com/board/view/?id=smartphone&no=5859372
https://gall.dcinside.com/board/view/?id=smartphone&no=5859992
I already attached it before. A guy from Korea is quite nervous about this cause he modified the setting with the developer tool to change the result and he already confessed the score is fake.