Become a fan of Slashdot on Facebook

 


Forgot your password?
Close
typodupeerror
×
Desktops (Apple) Apple Technology

Future Apple Silicon Macs Will Use 3nm Chips With Up To 40 Cores, Report Says (theinformation.com) 97

Posted by msmash from the closer-look dept.
The Information today shared alleged details about future Apple silicon chips that will succeed the first-generation M1, M1 Pro, and M1 Max chips, which are manufactured based on Apple chipmaking partner TSMC's 5nm process. MacRumors adds: The report claims that Apple and TSMC plan to manufacture second-generation Apple silicon chips using an enhanced version of TSMC's 5nm process, and the chips will apparently contain two dies, which can allow for more cores. These chips will likely be used in the next MacBook Pro models and other Mac desktops, the report says. Apple is planning a "much bigger leap" with its third-generation chips, some of which will be manufactured with TSMC's 3nm process and have up to four dies, which the report says could translate into the chips having up to 40 compute cores. For comparison, the M1 chip has an 8-core CPU and the M1 Pro and M1 Max chips have 10-core CPUs, while Apple's high-end Mac Pro tower can be configured with up to a 28-core Intel Xeon W processor.
This discussion has been archived. No new comments can be posted.

Future Apple Silicon Macs Will Use 3nm Chips With Up To 40 Cores, Report Says More

Future Apple Silicon Macs Will Use 3nm Chips With Up To 40 Cores, Report Says

Comments Filter:

  • I'd rather have 4 cores that are 10 times faster. Heck I'd accept only 5 times faster.

    • I'd rather have 4 cores that are 10 times faster. Heck I'd accept only 5 times faster.

      And we would all like Faster Than Light space travel and over-unity energy production.

      Unfortunately, they are against the Laws.

    • Same. Other than a few specialized applications, the ability of software to use all these cores hasn't kept up with the silicon. Looking at my system usage, it appears one highest-performance core, and few efficiency cores would offer the best resource utilization.

      Perhaps it's easier to make 40 cores, than a single core twice as fast, so that's what they do?

      • Perhaps it's easier to make 40 cores, than a single core twice as fast, so that's what they do?

        I guess "easier" is one way to put it. GP is requesting a 10x faster core (but will settle for 5). Why get a suburban to carry 8 people at 70 mph instead of 1 car that goes 560 mph, since that way you could also carry just yourself faster when that's what you want to do? Why not, right?

        Damn I love car analogies!!! The best ./ tradition.

    • I'd settle for RAM that had faster cycle times than CPU clock speed. You know, like in the olden days when 6502s operated at 1MHz and the RAM chip access times were 250ns?

    • They should offer a version suitable for immersing in a room filled with flourene. As well as higher core speeds they'd make a fortune on accessories - scuba gear etc.

    • Well good news! Appleâ(TM)s cores are already the fastest single threaded cores in the business!

  • Not particularly deep/insightful (Score:5, Interesting)

    by Dixie_Flatline ( 5077 ) <vincent@jan@goh.gmail@com> on Friday November 05, 2021 @11:01AM (#61960193) Homepage

    40 cores is 32 performance cores and 8 efficiency cores, which is four times what the M1 Pro/Max chips have. (This is the so-called Jade 4C-Die configuration already speculated/leaked previously by Mark Gurman.)

    So basically this story is saying that manufacturing will get better, Apple will make bigger chips with more cores, and everything will get more efficient. A pretty safe bet.

    • Now just solve "supply chain problems" and we're in business.

    • 40 cores is 32 performance cores and 8 efficiency cores, which is four times what the M1 Pro/Max chips have...So basically this story is saying that manufacturing will get better,

      Holy cow, are we really so blase about moving to 3nm and going to four times the number of cores in just a matter of years?

      Yes to some extent we expect manufacturing to get better, but come on that is a pretty impressive jump!! We should celebrate it a little bit instead of just writing off soemthing that took a lot of work by a l

      • Maybe wait for it to actually happen before you start the celebration? Speculation and future plans often fall apart in the real world. Apple isn't infallible.

        • Speculation and future plans often fall apart in the real world. Apple isn't infallible.

          Maybe Apple isn't but it sure seems like TSMC is close.

          It seems way more likely any hiccups would come from the fab process than the chip design.

          • Re: (Score:2)

            by dfghjk ( 711126 )

            It "sure seems"? How so? What "sure seems"? That TSMC is infallible?

            "It seems way more likely any hiccups would come from the fab process than the chip design."

            It seems way more likely TO YOU, but that's because of who you are. This isn't even about "fab process" vs. "chip design", this is about people trying to make money off of pure speculation on what companies might do in the far future. It works, here you are being your pure tribalist self promoting idle gossip because you think it benefits your s

      • 40 cores is 32 performance cores and 8 efficiency cores, which is four times what the M1 Pro/Max chips have...So basically this story is saying that manufacturing will get better,

        Holy cow, are we really so blase about moving to 3nm and going to four times the number of cores in just a matter of years?

        Yes to some extent we expect manufacturing to get better, but come on that is a pretty impressive jump!! We should celebrate it a little bit instead of just writing off soemthing that took a lot of work by a lot of people (not all or even most of them Apple) to make happen!

        Now we know what Apple has been up to at its new Silicon Design Facility we heard about a couple of years ago. . .

        • Re: (Score:2)

          by dfghjk ( 711126 )

          "Now we know..."

          Do we? Because someone made some claim about the future that appeals to you?

          • "Now we know..."

            Do we? Because someone made some claim about the future that appeals to you?

            Just a random thought.

            Lighten up.

      • Re: (Score:2)

        by dfghjk ( 711126 )

        I don't recall you having this reaction when Intel published its latest performance claims. Why didn't you, SuperKendall, "celebrate it a little bit instead of just [it] writing off"? We know the reason, of course.

        "...took a lot of work by a lot of people (not all or even most of them Apple) to make happen!"

        Hasn't happened yet, SuperKendall, it is mere speculation of what may happen two generations from now.

        Once again, delusional bullshit from a tribalist. Just come out and say what you really mean, Appl

      • I mean...yes?

        At this point, it's news when fabs can't shrink their processes any further and need to move to a different technology. Right now, we're just looking at the incremental improvement that we expect.

        Don't take this to mean that I'm not wholly impressed with the M1 series—I'm typing this on an M1 Mac mini, and I plan on buying the 27" iMac (or whatever) when it's finally released. But yeah, the real magic of the M1 series of chips is already in the design; the part where they make it faster b

      • Holy cow, are we really so blase about moving to 3nm and going to four times the number of cores in just a matter of years?

        It's well-known that TSMC plans to go to 3nm. It also seems that Apple has already booked and pre-purchased their complete 3nm output for a very long time :-)

        And 4 times the cores... Is that unexpected? It's like they have done right now, they duplicated what they have. All it does is create a really big chip. There's the question whether that is optimal, or whether 2 or 4 of the existing chips wouldn't be better for the money, but then what we have here is a rumour. Not even a rumour, but just a very pr

      • This is Apple and Slashdot. Half the knobs here were impressed with Intel’s new desktop chips barely getting past AMD and the M1 Max while needing the Colorado River to cool the fucking things.

      • Re: (Score:2)

        by tlhIngan ( 30335 )

        Holy cow, are we really so blase about moving to 3nm and going to four times the number of cores in just a matter of years?

        Yes to some extent we expect manufacturing to get better, but come on that is a pretty impressive jump!! We should celebrate it a little bit instead of just writing off soemthing that took a lot of work by a lot of people (not all or even most of them Apple) to make happen!

        What's so surprising about that? 64 core processors are common these days in very high end machines. "In a few year

        • What's so surprising about that? 64 core processors are common these days in very high end machines

          Yes but in laptops on battery? Or an iPad/iPhone?

          I just feel like a lot of nerds here have lost sense of scale, and really are trivializing something that is actually quite impressive (and again, because people seem to be confused on this point, it's not even an Apple achievement as much as it is TSMC).

          I guess I'll just continue to keep my sense of wonder at the continued rapid pace of processor technology.

      • Do 8 efficiency cores make sense? If you need that much compute why not use a performance core? What I am saying is there may be an upper bound limit to efficiency cores, maybe you donâ(TM)t need more than 2 or 4. In fact, that may be the reason the M1 Pro and M1 Max have only two efficiency cores in comparison to the M1 which has 4.

        • It may get to the point when you have mid-efficiency cores and then weâ(TM)ll get towards each core being slightly different. Like, some cores for AI, some for GPU, some hybrids, at varying levels of efficiency, performance, and capability. All top chip design companies use one of the three top semiconductor EDA software (cadence, synopsys, mentor) and they make it easy to do stuff like that.

    • I will start saving my allowance money. Definitely want one.

  • You can fit about 20 carbon atoms within a line of 3nm. So how many silicon atoms go into a transistor in this architecture? 2,000 maybe?

    It is hard to believe we are building structures at this scale.

    • Re: (Score:2)

      by crow ( 16139 )

      Considering that they all use different metrics for reporting the feature size, perhaps we need a new metric: atomic weight per transistor. This would probably be reported in 'kaw' for kilo atomic weight or something like that.

    • The problem with chip making is not just the ability to mass manufacture smaller structures which poses challenges. The problem is also that at smaller and smaller sizes, unique issues start occurring. For example as the transistor gate got smaller, engineers has to use structures like FinFET [wikipedia.org] to counter problems like leakage. My understanding is that more and more power would be required to operate a single gate otherwise as the electrons would simply jump the gates if structures like this were not built.

  • pci-e lanes? max ram? storage not on die?

    the Mac pro will need stuff like that and maybe even video power to drive 3-4 8K displays.
    And an pro system will need pci-e slots / sata storage ports Not locked to the MB raid 0 setups. As well as ram slots.

    • pci-e lanes? max ram? storage not on die?

      the Mac pro will need stuff like that and maybe even video power to drive 3-4 8K displays.
      And an pro system will need pci-e slots / sata storage ports Not locked to the MB raid 0 setups. As well as ram slots.

      Pretty sure they are going to eventually evolve into a small cluster of a few different functional-block SoCs for their high-end designs.

    • Re: (Score:2)

      by slaker ( 53818 )

      As a guess, Apple will probably push for external disks and expansion slots over Thunderbolt, just to keep as much of the internals sealed as possible. RAM is the biggest reason to avoid Apple PCs right now and presumably those systems that have workstation-grade amounts of it will also be its biggest profit-makers.

      I'm a camera hobbyist, but I (sometimes) shoot video in 8k and work with 45MP stills. I use a Threadripper system with 64GB RAM; I was running out of RAM when I only had 32GB, which really makes

      • I'm a camera hobbyist, but I (sometimes) shoot video in 8k and work with 45MP stills. I use a Threadripper system with 64GB RAM; I was running out of RAM when I only had 32GB, which really makes me wonder how people working from 16GB Macbooks are getting anything done.

        You take a very, very fast hard drive and very, very fast swapping.

    • Re: (Score:2)

      by dfghjk ( 711126 )

      "And an pro system will need pci-e slots / sata storage ports Not locked to the MB raid 0 setups. As well as ram slots."

      Why? I suspect Apple does not agree.

    • Re:pci-e lanes? max ram? storeage not on die? (Score:4, Interesting)

      by Anubis IV ( 1279820 ) on Friday November 05, 2021 @04:23PM (#61961265)

      max ram?

      Today's M1 Max caps out at 64GB on die, which is intended for laptop usage, so it can be safely assumed that they'll offer higher options for something intended to be used in a Mac Pro. Whether it will allow for later expansion remains to be seen, however.

      storage not on die?

      ...is more or less out of Apple's hands by definition. They'll likely bake some storage on die just like they do for the various flavors of M1, but how you choose to expand on that in a Mac Pro will be entirely up to you, not Apple. It seems safe to assume that they will allow for expansion in this area, just as they always have in the Mac Pro.

      pci-e lanes?

      Irrelevant to end users. While system builders need to be aware of those details, end users only need to know what their expansion options are (e.g. how many slots and how fast is each?). And if that feels like a distinction without difference because you think that everyone uses PCIe slots, I'll point out that today's Mac Pro, which has been fairly well received, gets its highest performance expansion options from something other than PCIe slots. Sure, there are 36 lanes provided across four PCIe expansion slots, but the other 56 lanes made available to the user come in the form of MPX modules, many of which take advantage of the additional bandwidth afforded by MPX (I'll mention one of those in a sec, in fact).

      Which is to say, you can probably expect Apple to match or exceed today's throughput, but what form that comes in will be more important to the end user than how many lanes the chip itself supports.

      the Mac pro will need [...] video power to drive 3-4 8K displays

      You've jumped incredibly far into wishful thinking with this one without apparently realizing you've asked for something that's insane by today's standards. If the M-series chip used in the Mac Pro could support "just" two HDR 8K60 displays, it would put them lightyears ahead of everyone else and would represent a tremendous value in today's market.

      To put your demand in perspective, so far as I'm aware there aren't any GPUs on the market capable of natively driving two, let alone three or four HDR 8K60 displays, so I don't know why you'd think the Mac Pro needs to do so. I'll admit I may be wrong or have outdated info, but the closest I can think of is the AMD Radeon Pro W6800X Duo MPX Module. It's two W6800Xs bolted together in a single MPX module that AMD just released three months ago, but it can "only" drive eight 4K displays (i.e. the equivalent horsepower of two 8K displays) and it costs $5000, so you'd need to shell out $10,000 for the horsepower you're talking about. Alternatively, you could get that much horsepower by picking up four RTX 3090s at $1500 apiece (assuming you can actually find them at MSRP), but when your plan relies on purchasing four perpetually-out-of-stock RTX 3090s, you need a new plan.

      What the Mac Pro actually needs are expansion capabilities that cover a wide variety of use cases, enabling it to keep up as this technology continues to rapidly evolve. That said, if Apple can drive "just" two 8K displays out of the box, that'd be a huge value-add that could potentially save affected Mac Pro users a cool $3000-$5000 apiece.

      Even more interesting is that it's actually reasonable to think Apple may be able to do just that. Today's M1 Max MacBook Pro can drive its own display + three 6K + one 4K. I think those are all HDR @60 Hz or better, which by my math puts its aggregate throughput just barely over what it needs to drive two HDR 8K60 displays. With the Mac Pro not having a display of its own, it seems reasonable that it actually could natively support two HDR 8K60 displays out of the box, which would be crazy.

    • The MacBookPro is already shipping with the ability to drive itâ(TM)s built in 4k/120Hz 10 bit display, 4 6k 10 bit displays, and a 4k 10 bit display all at once, so itâ(TM)s pretty close to your metric for running screens already.

  • As I see it, the main reasons to do this as a more complicated and somewhat slower multi-die configuration are Yield (much harder to get decent yield the more cores on a die) and easier and more flexible Performance Tiers (think Pro vs. Max).

    Plus they can more easily horse-race different numbers and configurations of dice in their Prototype Labs to see where the laws of diminishing returns start to put an upper-limit on TDP, etc.

    Like swapping CPUs and Graphics cards in and out; but on a (much) smaller scale

    • Re: (Score:2)

      by dfghjk ( 711126 )

      Multi-die processors are not new, they do them for yield as you say. A company will transition away from them at the earliest opportunity because they are expensive and"slow".

      Apple would only use this kind of approach for high end systems, it would be stupid to think of using a multi-die processor in a laptop, for example. Could Apple introduce a 40 core processor in 2-3 years? Maybe, but the cost of the Mac Pro they put it in would be prohibitive (and SuperKendall would tell us what a bargain it is).

      It

      • It would be hysterical to imagine how big the part would be, with the massive number of discrete DRAMs soldered directly onto it. Apple's M1 design is clever but lacks configurability and doesn't scale well, yet the fanboys are losing their minds because someone wants a paycheck by multiplying Apple recent core counts by 4x. Wildly uninteresting speculation, and one not supported well by processor release histories.

        My prediction is that M2 Pro/Max will either have the ability to work together with one to three identical processors, or the ability to use practically unlimited amounts of DRAM. So they can either build a MacPro with 32+8 cores and up to 256 GB of RAM - at a price - or a MacPro with 8+2 cores and unlimited RAM. Although 1.5TB costs you $25,000 from Apple right now, and I'm told Dell isn't much cheaper either.

        And the M3 Pro/Max will have both.

        And the M4's will all be built on 3nm, so they can fit 50%

      • Multi-die processors are not new, they do them for yield as you say. A company will transition away from them at the earliest opportunity because they are expensive and"slow".

        Apple would only use this kind of approach for high end systems, it would be stupid to think of using a multi-die processor in a laptop, for example. Could Apple introduce a 40 core processor in 2-3 years? Maybe, but the cost of the Mac Pro they put it in would be prohibitive (and SuperKendall would tell us what a bargain it is).

        It would be hysterical to imagine how big the part would be, with the massive number of discrete DRAMs soldered directly onto it. Apple's M1 design is clever but lacks configurability and doesn't scale well, yet the fanboys are losing their minds because someone wants a paycheck by multiplying Apple recent core counts by 4x. Wildly uninteresting speculation, and one not supported well by processor release histories.

        The Mx SoCs as they exist currently are a spectacular solution for high end Mobile and at least respectable laptops, AIOs and slotless desktops. IOW, their current use-cases. But I agree with you that the jury is quite the fuck still out when it comes to how they will adapt their strategy to a general purpose (and fairly robust) slotted design; preferably one that can support third-party cards (including dGPUs), and (dare we hope?) expandable RAM and mass-storage. This is no doubt where they are thinking mu

      • Multi-die processors are not new, they do them for yield as you say. A company will transition away from them at the earliest opportunity because they are expensive and"slow".

        Which is why AMD is putting so much effort into a faster infinity fabric. And yield isn't the only reason. Design flexibility is another.

  • RIP Intel, we hardly knew ye.

    • Yep, Intel is dead. MacOS is about to hit *double digit* market share. I'm sure Intel is quaking in its boots.

      Sarcasm aside, no one in the history of computing has based their decision of Mac vs PC on performance of the CPU. The only people who can make Intel quake in their boots is Lisa Su and her band of merry engineers.

    • Re: (Score:2)

      by gtall ( 79522 )

      Intel has nothing to worry about. China will Jinping Taiwan before these new chips get out.

  • From the TFA:

    ... the chips will apparently contain two dies, which can allow for more cores.

    <grumpy_old_man_mode>

    I thought a die was a chip. It's a very expensive patterned flake of silicon. When did "chip" come to mean "package"?

    </grumpy_old_man_mode>

    • Re: (Score:2)

      by dfghjk ( 711126 )

      A "chip" is something you attach to a "board", a "die" is something inside the chip that is wired to the pins. This is not new terminology.

    • You made me look ;-)

      My MacBook Air is idling at 3.3%

      You might want to check for a mining virus. Or Edge is shipping your hard drive to Microsoft. Or Firefox is being it's usual inefficient self.

Slashdot Top Deals

In the long run, every program becomes rococco, and then rubble. -- Alan Perlis

Close