Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
×
Apple

Apple Unveils M2 Ultra Processor (venturebeat.com) 94

Apple announced the M2 Ultra processor, a new chip for its Mac Studio workstation for professional users. From a report: The chip has 134 transistors and 24 central processing unit (CPU) cores with 20% faster performance. It has up to 76 graphics processing unit (GPU) cores at up to 30% faster performance. Apple made the announcement at its WWDC event today on the Apple campus in Cupertino, California.

The chip will go into the Mac Studio product, which previously used Intel silicon. These are machines like those used by engineers to deliver Saturday Night Live or create blockbuster movies, said Jennifer Munn at Apple. Apple said this completes the transition to Apple silicon. Developers can build new versions of apps at warp speed, with up to 25% faster performance than in the past, Munn said. The 32-core neural engine is 40% faster at AI calculations. It supports 192 gigabytes of unified memory, which is 50% more than M1 Ultra.

This discussion has been archived. No new comments can be posted.

Apple Unveils M2 Ultra Processor

Comments Filter:
  • how many pci-e lanes does it have?

    as the new mac pro does seam to have an pci-e switch in it

  • by Ragnarok21 ( 413417 ) on Monday June 05, 2023 @01:04PM (#63577613)
    That's pretty cool...only 134 transistors to create 24 cores?!? Me thinks you got the number wrong,
    • by alvinrod ( 889928 ) on Monday June 05, 2023 @01:47PM (#63577737)
      I'm kind of curious about the minimum number you could get away with. The Inte 4004 only had about 2,300 transistors which is pretty crazy. Obviously the ISA was far simpler than what we have today, but it's still impressive.
      • by _merlin ( 160982 )

        The 4004 is a bit odd in that the ROM and RAM chips are involved in instruction decoding. The instructions are visible on the bus, and the RAMs and ROMs detect relevant instructions and handle address decoding/latching, etc. Without that, the 4004 itself would have needed more transistors.

        • The 4004 is a bit odd in that the ROM and RAM chips are involved in instruction decoding. The instructions are visible on the bus, and the RAMs and ROMs detect relevant instructions and handle address decoding/latching, etc. Without that, the 4004 itself would have needed more transistors.

          The 4004 is basically just the logic instruction core. The 4004 by itself was relatively useless - it was sold as part of a chipset - there were 4001 to 4007 chips I believe - consisting of RAM chips, ROM chips (programmable), I/O chips and other things. You needed about 4 or 5 of them in order to create a working processor. The 4004 is the "central computation" part, where it actually handles the basic ALU and other processing tasks. Things like instruction decoding, instruction pointer (PC), branches etc., were handled by companion chips.

          You can't just take a 4004 by itself and create a working system - you do need a bunch of external logic bits (handled by the rest of the 400x chips) to implement a working system.

          • by _merlin ( 160982 ) on Tuesday June 06, 2023 @02:05AM (#63579465) Homepage Journal

            The 4004 includes the register file and ALU. It does a lot of the work, but not all of it. The ROM and RAM chips have I/O ports, too. It's largely to save pin count on any given chip, and make the system scalable. A working system requires:

            • 4201 multi-phase clock generator
            • 4004 CPU (instruction sequencer, register file, ALU)

            • At least one 4001 ROM (mask programmed, 256 bytes, one 4-bit I/O port)
            • At least one 4002 RAM (64 data nybbles, 16 attribute nybbles, one 4-bit output port)

            The 4001 ROMs have their base addresses hard-coded in the mask as well as the contents, so you needed to order them as a set. There was also a 4003 10-bit shift register available, designed to be used as an I/O expander.

            If you wanted to use standard ROM or static RAM chips for program storage, you needed a either the 4008 and 4009, or the later 4289 that provided more functionality in a single chip. These would typically be used with 2102 RAMs or 1702 UV EPROMs. These were originally designed for development systems and prototypes to allow changing the program without ordering new mask ROMs, but quite a few products used them so they could use standard ROMs rather than ordering mask programmed 4001s from Intel (e.g. arcade video games and gambling machines often used a 4289 with standard OTP ROMs, like the 82S126, for program storage).

      • by weberjn ( 771517 )

        The Megaprocessor has 15,300 transistors.

        https://www.megaprocessor.com/ [megaprocessor.com]

    • by sxpert ( 139117 )

      also, last I checked, the mac studio always features some sort of M1 apple silicon processor

    • by Anonymous Coward
      Me thinks venturebeat got the number wrong. The correct number (134 billion) is even in the slide at the top of their article. Maybe the /. editors are moonlighting for VB?
    • That's pretty cool...only 134 transistors to create 24 cores?!? Me thinks you got the number wrong,

      Ha ha ha.

      Um, yeah, the word "billion" is missing, as in it should be "134 billion transistors".

      Here's a better summary:
      "The new M2 Ultra, built using 5nm technology and featuring 134 billion transistors, will feature 24 CPU cores, up to 76 GPU cores (there’s also a 60-core option) and a 32-core Neural Engine. The CPU consists of 16 next-gen high-performance cores and eight high-efficiency cores.
      One major change, too, here is that the M2 Ultra supports up to 192GB of unified memory, backed by 800GB/s o

    • The most important question is... Can it run Crysis?
    • by Tetch ( 534754 )
      How utterly baffling that despite 36 hours having now gone by since this obviously laughably idiotic braino in the writing of this article summary, *still* nobody from Slashdot has seen fit to correct the damn summary. Asleep at the wheel hardly even begins to cover it :(
  • 134 transistors? (Score:5, Informative)

    by DrMrLordX ( 559371 ) on Monday June 05, 2023 @01:04PM (#63577617)

    How's that for area efficiency?

    (read the slide, it's 134 billion)

  • 192GB is to low for an pro workstation with vampire video

  • I had no idea that so much could be done with just 134 transistors.
  • The chip has 134 transistors

    I know a processor with 1,000 times as many transistors. I give you the Intel 80286!

  • Will the MK Ultra processor be able to read your mind.
  • There's no link in TFS, so I can't verify, but doing 24 CPU cores and 76 GPU Cores using 134 transistors is a feat I didn't think could be done. Slashdot must have switched servers to use this new CPU because I tried to post this seven times before it worked, but that's what you get for 134 transistors.
  • Omg! That transistor count! How does this new thing stack up to the Apollo Flight Controller? Please, don't obfuscate your information behind 'core count.' /. was always a nudist colony so far as tech goes. Why do we even put up with such information-free articles anyway? Was it something I did? Bare your tech. We'll all appreciate it. If you only have 134 transistors to work with, that's little better than a few guitar pedals strung together.
    • Replying to myself, since /. still doesn't know what editing means. I needed a laugh. Thanks for giving me one, half-chewed Apple.
  • Did they ever solve that problem with running Intel binaries? Just askin'. I was issued an M1 mac for work a few months ago and they had to take it back and go find an Intel Mac somewhere, as the software I was supposed to be developing on was Intel only. Side note, trying to get the IT guy to understand that M1 is a COMPLETELY DIFFERENT ARCHITECTURE was a real adventure. They do understand that now.

    I'm perfectly happy with my Intel mac, but I know this is only an interim solution.

    I'm aware that vendors

    • by Bert64 ( 520050 ) <.moc.eeznerif.todhsals. .ta. .treb.> on Monday June 05, 2023 @01:39PM (#63577707) Homepage

      The M1 macs have had the ability to emulate intel code right from the start, it's called rosetta 2 and it works pretty transparently. I still have one or two apps which run under emulation.

      • This was June 2022, was it available then?

        It might be moot. At the time, the only solutions that IT would support were VMWare and Parallels, both had versions that would run, (vmware I think was beta) but in both cases, they'd only allow ARM instances -- no code translation was supported. And the software I needed to run was not compiled for ARM. As an added complication, it wasn't compiled for Mac -- I was expected to run Linux virtually and then run the software on the Linux instance. But there was no

        • Rosetta isnâ(TM)t installed by default. You have to explicitly install it. And yes, it was available last summer.

          Iâ(TM)m happy with the Intel MBP I got two years ago. By the time itâ(TM)s up for a refresh (five years at our company), Iâ(TM)d hope that the ARM ecosystem has matured including better support for Windows ARM from Microsoft ⦠but weâ(TM)ll see.

          • by dgatwood ( 11270 ) on Monday June 05, 2023 @02:58PM (#63578045) Homepage Journal

            Rosetta isnâ(TM)t installed by default. You have to explicitly install it. And yes, it was available last summer.

            Either way, it is moot. The original poster needed to run Intel VMs, which can't be usably done on ARM-based Macs currently, because Rosetta emulates only enough instructions for user-space applications, not to run a full VM with synchronization across multiple CPU cores.

            • Thanks, that answered my original question, could it be done today, and the answer seems to be, not yet.

              • by dgatwood ( 11270 )

                Thanks, that answered my original question, could it be done today, and the answer seems to be, not yet.

                You can, of course, run the ARM versions of Linux or Windows, and at least in the latter case, run x86-64 binaries using Microsoft's emulator. So the main impact is that you can't run existing VMs or virtualize older versions of macOS. The latter is what bit me.

              • by Bert64 ( 520050 )

                It can be done, with UTM (which is a frontend to qemu).. UTM also runs on iOS.
                You can emulate a full x86 machine (you can even emulate SPARC or PowerPC etc too).
                Emulation carries a performance hit, and the performance hit is greater if you're emulating the full system instead of just userland, so expect it to be significantly slower.

                • by dgatwood ( 11270 )

                  It can be done, with UTM (which is a frontend to qemu).. UTM also runs on iOS. You can emulate a full x86 machine (you can even emulate SPARC or PowerPC etc too). Emulation carries a performance hit, and the performance hit is greater if you're emulating the full system instead of just userland, so expect it to be significantly slower.

                  I remember trying that. Unfortunately, I think you're somewhat understating things when you say "performance hit". QEMU runs single-threaded on a single ARM64 core [github.com] when emulating x86-64. So for an 8-core CPU, you'll take an 8x performance hit from losing all but one core, times probably a least a 2x performance hit from extra thread scheduler contention in QEMU itself as it switches back and forth emulating the various cores and has to switch execution contexts with timing that likely isn't as efficient

            • Rosetta isnâ(TM)t installed by default. You have to explicitly install it. And yes, it was available last summer.

              Either way, it is moot. The original poster needed to run Intel VMs, which can't be usably done on ARM-based Macs currently, because Rosetta emulates only enough instructions for user-space applications, not to run a full VM with synchronization across multiple CPU cores.

              Rosetta 2 doesn't Emulate anything, unless it has no other choice. It is primarily an On-Load or On-Initial-Launch Translator.

            • by Malc ( 1751 )

              Either way, it is moot. The original poster needed to run Intel VMs

              He didn't actually say that. He said he needed to run Intel software and he's IT dept only supported VMs. That's not enough information to jump to any conclusions, including whether or not he has to use solutions proposed by his IT dept. I think the requirements became clearer in later replies in this thread though.

          • by cfalcon ( 779563 ) on Monday June 05, 2023 @04:02PM (#63578331)

            Gods if you just go ahead and disable translation from ' to (pile of crap) under general keyboards smart punctuation, that would be GREAT

            • Gods if you just go ahead and disable translation from ' to (pile of crap) under general keyboards smart punctuation, that would be GREAT

              Gods if Slashdot would just join the rest of the world and properly support Unicode, that would be GREAT

            • by Malc ( 1751 )

              How will that fix the fuck-up /. makes of my country's currency symbol? Oh, it won't, will it? Why should I configure my devices to make just some of the broken characters work with /.? I won't; they should fix their bugs. Save your breath, your whinging at me won't make a difference.

        • by Bert64 ( 520050 )

          Rosetta 2 was available at launch (late 2020) and even on the developer machines which came earlier than that.

          Linux has always had the ability to run non native binaries via qemu-user, so you install linux for arm and then qemu-user to run x86 binaries or whatever architecture your binaries were compiled for.
          It's rarely used because most linux software is open source and can already run natively on ARM.

          There is also windows for arm, which has its own facility to run non native binaries.

          • At the time I was working on this, my assignment was to develop in a proprietary software package that ran on Linux for which source was not available. Furthermore, running it on the ARM version of Linux under a translator, even if possible, was not supported. So no help there.

            I'm aware that there's a Windows version on ARM. It's not generally available, and you have to join the Microsoft developer's group to get the bits, but that's not difficult to do.

            The problem, again, was that the software I needed

            • I should also say, CC is the only thing I still need Windows for. Everything else runs in a browser and could run anywhere. The moment Adobe ports CC to ANY version of Linux, (Android doesn't count -- Lightroom and Photoshop on Android are toys, not for serious work) will be the time I abandon Windows for good.

              Howabout 99.9% of CC running Natively on Apple Silicon? :

              https://helpx.adobe.com/downlo... [adobe.com]

        • Virtualization software was one of the few types that couldn't run under Rosetta 2 at first. Which is understandable, given how deep it goes into the system architecture. But I think both VMWare and Parallels now support Intel virtual machines, and the whole bundle (including the virtualization software) gets run under Rosetta 2.

          • Ok, cool. Maybe it's time to give the M1 another try.

            • Sorry, I was wrong. VMware and Parallels now run on ARM, but they do not (and seemingly will not) support a virtual machine running an OS compiled for Intel. They say that Rosetta 2 doesn't support this arrangement. However, they can run ARM-compatible OS's that have Intel emulation built in, e.g., Windows 11 for ARM.

              So it sounds like your best option is either to run an ARM version of Linux, possibly with emulation support for Intel binaries (if that exists).

              Some more info available here [vmware.com] and here [parallels.com].

              • Since you're trying to run Linux software, not Windows, another option might be to use a thinner virtualization environment like Docker [docker.com] or Karton [github.io]. (Also see here [docker.com] and here [frai.dev].) It looks like they will use Rosetta 2 to translate Intel code to run on the Mac.

      • The M1 macs have had the ability to emulate intel code right from the start, it's called rosetta 2 and it works pretty transparently. I still have one or two apps which run under emulation.

        There are limitations on what Rosetta 2 can do, though. You can't, for example, run an x86-based operating system as a VM (Windows or macOS).

      • by dfghjk ( 711126 )

        SuperKendall is about to tell you how stupid you are. Didn't you get the memo? Apple doesn't allow the term "emulate" to be used when describing Rosetta. It "translates" the x86 instructions, don't ya know, totally different!

    • Did they ever solve that problem with running Intel binaries? Just askin'. I was issued an M1 mac for work a few months ago and they had to take it back and go find an Intel Mac somewhere, as the software I was supposed to be developing on was Intel only. .

      That makes no sense. Rosetta 2 has shipped on Apple Silicon Macs from day 1, allowing you to run Intel binaries. Maybe you mean Windows Intel apps? For that, use Parallels and run Windows ARM, which includes an emulation layer to let you run Windows Intel apps.

      • I covered this in another reply. The software ran on Linux, so the plan was to run Linux in an emulator, then install the software on that instance. Except there was no version of the software compiled for Linux on ARM, and even if it ran in a emulator on a Linux instance (that makes my head hurt thinking about it) that solution was not supported by the vendor.

        Similarly, I could get ARM Windows running as an instance, but the software components I needed to run under Windows were also not compiled for ARM

    • docker on mac is not that good
      In summary, running Intel-based containers on Arm-based machines should be regarded as “best effort” only

    • by Spansh ( 219937 )

      I had this exact issue. I spent some time experimenting with UTM, QEMU and Vagrant. This was a little over a year ago (March 2022 according to the timestamp on the Vagrantfile).

      I was hoping that Parallels would eventually release something that would make this a lot more seamless, but apparently they're nervous about Apple bashing them overt the head for providing x86 on Arm compatibility and as such aren't going to release something that allows it.

      I managed to get x86 vagrant images working fine, here's

  • 134 transistors

    Sweet. Even the 8085 from 1976 needed 6,500. What are they doing with all the extra space in the case? Government spying?

  • That's over 5 transistors per core. Surely they can do better.

  • The Mac Pro is mostly the same as the Studio, only with PCI slots. No upgradable memory, reduced to 192GB from 1.5TB and the Intel Mac pro is discontiuned. Hope you like Hackintoshing, as that's the only way to get a new Intel Mac now.
    • by djb ( 19374 )

      Yeah I wonder what happened to the Jade 4C-Die version of the M processors? That was ment to be double the Ultra and give you 384GB of RAM. Which is far more usable on a Workstation class machine these day.

      That extra 3k also gets you much better cooling and a big PSU, so in theory they could clock the M2 Ultra a fair bit higher. Will be interesting to see benchmarks

      • That extra 3k also gets you much better cooling and a big PSU

        Okay, so what is the other $2,900 for? Some PCI-E slots they called PCI slots in their literature? That's pretty fucking expensive for slots and air.

      • by jsepeta ( 412566 )

        Apple made an announcement several months ago that they weren't going to produce the super-duper M2 with 384gb RAM because the market was too small. I have no idea how large the Mac Pro market is, or what type of expansion cards someone might want to put in an expandable Mac.

  • The original road map had a chip that doubled the Ultra and would have allowed 384gb of ram. Which a lot of Workstation users will want these days.

    The first gen Studio had to under clock the GPU on the Ultra due to the PSU running too hot and not being able to deliver enough juice. The Mac Pro had much much better cooling and larger PSU, so hopefully Apple let the chip run to its full potential in that machine.

    Will be looking out for the bench marks.

  • ... but what will Apple be leaving out of the machines to make them yet thinner? The RAM? The PCB? Electricity?

  • ..transistors than a cheap dollar store digital watch. I am quite impressed.

    I think the poster meant "134 billion(?) transistors spread out across 24 cores."

  • Site: https://fullukdrivinglicense.c... [fullukdrivinglicense.com] Apply for your first provisional driving licence - Full UK Driving . Discription:---Buy UK driving licence online without exams required. We are here to help people for authentic, original & DVLA verified UK driving licences. DVLA Registered
    • by 0xG ( 712423 )

      Hgdgbjjmnvcdwwxvbnillkmmkokii;..,,,klllooo0-987ygffdxfcxaqqdfgghhjfuckfuckfuckkfhggftyfdesdvhjkmllplmmni?

No spitting on the Bus! Thank you, The Mgt.

Working...