> I wonder if they'll just do 4x M1 Max for that.

Unlikely, M1 Ultra is the last chip in the M1 family according to Apple [1].

"M1 Ultra completes the M1 family as the world’s most powerful and capable chip for a personal computer.”"

[1] https://www.apple.com/newsroom/2022/03/apple-unveils-m1-ultr...

I've been saying 4x M1 Max is not a thing and never will be a thing ever since the week I got my M1 Max and saw that the IRQ controller was only instantiated to support 2 dies, but everyone kept parroting that nonsense the Bloomberg reporter said about a 4-die version regardless...

Turns out I was right.

The Mac Pro chip will be a different thing/die.

Plus they are running out of M1 superlatives. They’ll have to go to M2 to avoid launching M1 Plaid.

They could do M1 More Thing.

m1 hyper turbo deluxe

or they could take a page out of microsofts book and just call the next one "m one"

M1 One: The Second One!

M1 Series One

M1 Ultra²

M1 Mark II ala Sony

Bloomberg brought the Supermicro hit pieces. I personally can't take them seriously anymore. Not after the second article with 0 fact checking and sad attempt at an irrelevant die shot. And their word is certainly irrelevant against one of people who are working (and succeeding) at running linux on M1.

Could they do a multi-socket board for the Mac Pro?

I expect this for CPU side. Multiple SoC introduces NUMA but it's already done on Dual Xeon based Mac. I wonder how their GPU work for the configuration.

In a world where latency:storage size tradeoffs need to be made for practical reasons (and will, at some point, be required for fundamental physical reasons), we should just embrace NUMA anyway. Death to the lie of uniform access! NUMA is the future!

Ehrm, anyway.

It actually isn't clear to me whether designing a two socket motherboard is fundamentally an easier task than jamming more of the things into a single package (given that they have already embraced some sort of chiplette paradigm).

I believe designing/manufacturing a 5nm chip only used for Mac Pro can't be profitable. That's why they designed M1 Ultra as MCM of M1 Max.

They would never do that

The Mac Pro (and the high-end Powermacs that preceded it) were always available in a dual socket incarnation, right up to the trashcan Mac.

The trashcan Mac Pro was announced almost 9 years ago. Apple today is a very different company. They've moved onto chiplet designed SoC's with little to no care about upgradability.

They have done this previously for dual socket Xeons. Historical precedence doesn’t necessarily hold here, but in fact, it’s been done on the “cheese graters” previously

They've moved on to chiplet design. Don't disregard their clear direction to SoC's with non-upgradeable RAM, GPU, CPU.

They also water-cooled them but I’d bet a lot of money we never see that again either.

The Mac Pro chip will be a 4x M2 Max

In a previous Apple press release[1] they said:

> The Mac is now one year into its two-year transition to Apple silicon, and M1 Pro and M1 Max represent another huge step forward. These are the most powerful and capable chips Apple has ever created, and together with M1, they form a family of chips that lead the industry in performance, custom technologies, and power efficiency.

I think it is just as likely that they mean "completes the family [as it stands today]" as they do "completes the family [permanently]."

[1] https://www.apple.com/newsroom/2021/10/introducing-m1-pro-an...

edit: This comment around SoC code names is worth a look too: https://news.ycombinator.com/item?id=30605713

What is M2 really going to be difference wise?

I think M2 is going to improve on single thread performance.

Judging from the geekbench scores[0], M1, M1 Pro, and M1 Max perform identically in single threaded tasks. And the newly leaked Mac Studio benchmark[1] shows essentially identical single thread performance.

[0]: https://browser.geekbench.com/mac-benchmarks [1]: https://browser.geekbench.com/v5/cpu/13330272

~15-20% faster if releases start this year, plus whatever optimizations learned from M1 in wide release such as perhaps tuning the silicon allocation given the various system. If next year, M2 or M3 (get it) will use Taiwan Semi's so-called 3nm, which should be a significant jump just like 7-5nm several years ago for the phones and iPads.

Hopefully one of the changes of the M2 design will be a better decorrelation of RAM and cores count.

They’d need that anyway for a Mac Pro replacement (128GB wouldn’t cut it for everyone), but even for smaller config it’s frustrating being limited to 16G on the M1 and 32 on the Pro. Just because I need more RAM doesn’t mean I want the extra size and heat or whatever.

For my purposes, the biggest drawback of using an SoC is being constrained to just the unified memory.

Since I run a lot of memory intensive tasks but few CPU or GPU bound tasks, a regular m1 with way more memory would be ideal.

I doubt there will be much learned after actually shipping M1. Developing silicon takes a long time. I wouldn’t be surprised if the design was more or less fixed by the time the M1 released.

They also said that the Mac Pro is still yet to transition. So they'll have to come up with something for that. My suspicion is that it won't be M branded. Perhaps P1 for pro?

I have been thinking about this, and one theory I came up with was a switch chip between M1 Max chiplets... you have 4 (or more?) M1 Maxes connected to the same switch chip... it might add latency for some tasks but it would be one way to scale without going to a new M2 processor... Then again, M2 could come out in June at the WWDC, doubling everything and adding 3 more ultra connect things to each M2 Max, allowing unlimited (ish) upgradability... but, probably not...

It may be M branded, just not M1. It could be based on an M2 Ultra or M2 Mega or whatever and what they said would still hold true.

That doesn't necessarily rule out more powerful iterations that also launch under the M1 Ultra branding though.

(edit: per a sibling comment, if the internals like IRQ only really scale to 2 chiplets that pretty much would rule it out though.)

Probably not on the same design as the current M1 series, at least not for the Mac Pro. The current x86 pro supports up to 1.5TB of RAM. I don’t think they will be able to match that using a SoC with integrated RAM. There will probably be a different CPU design for the Pro with an external memory bus.

M2 family Mac Pro upcoming…

>Looks like all the people saying "just start fusing those M1 CPU's into bigger ones" were right,

Well they were only correct that Apple managed to hide a whole section of Die Image. ( Which is actually genius ) Otherwise it wouldn't have made any sense.

Likely to be using CoWoS from TSMC [1] since the bandwidth numbers fits. But needs further confirmation.

[1] https://en.wikichip.org/wiki/tsmc/cowos

The interconnect area was already known from third-party die photos of the M1 Max https://twitter.com/vadimyuryev/status/1466526403331952644

They were not known when people keep rampling about the the Mark Gurman report on "just start fusing those M1 CPU's into bigger ones".

I wrote about it three months ago.

https://news.ycombinator.com/item?id=29430817

I've been using a Vega-M for some time which I think follows this model. It's really great.

Right. Still riding gains from the node shrink and on package memory.

Could AMD/Intel follow suit and package memory as an additional layer of cache? I worry that we are being dazzled by the performance at the cost of more integration and less freedom.

The next CPU coming from AMD will be the 5800X3D with 96MB cache. They stack 64MB L3 on top. Rumours say it comes out 20th of April.

edit: typo + stacking + rumoured date

They might have to have the unified memory more dense to get to 1.5 TB max of RAM on the machine (also since this would be originally shared with a GPU). Maybe they could stack the RAM on the SoC or just get the RAM at a lower process node.

The M1 Max/Ultra is already extremely dense design for that approach, it's really almost as dense as you can make it. There's packages stacked on top, and around, etc. I guess you could put more memory on the backside but that's not going to do more than double it, assuming it even has the pinout for that (let's say you could run it in clamshell mode like GDDR, no idea if that's actually possible, but just hypothetically).

The thing is they're at 128GB which is way way far from 1.5TB. You're not going to find a way to get 12x the memory while still doing the embedded memory packages.

Maybe I'll be pleasantly surprised but it seems like they're either going to switch to (R/LR)DIMMs for the Mac Pro or else it's going to be a "down" generation. And to be fair that's fine, they'll be making Intel Mac Pros for a while longer (just like with the other product segments), they don't have to have every single metric be better, they can put out something that only does 256GB or 512GB or whatever and that would be fine for a lot of people.

> You're not going to find a way to get 12x the memory while still doing the embedded memory packages.

https://www.anandtech.com/show/17058/samsung-announces-lpddr...

> It’s also possible to allow for 64GB memory modules of a single package, which would correspond to 32 dies.

It is possible, and I guess that NVIDIA’s Grace server CPU will use those massive capacity LPDDR5X modules too.

The M1 Ultra has 8 memory packages today, and Apple could also use 32-bit wide ones (instead of 64-bit) if they want more chips.

Do people really need 1.5 TB of unified memory? If you had 128 GB of unified memory and another pool of 1.5 TB or whatever of “slower” memory (more like normal speed on the Intel/AMD side) would that work?

You (or the OS or the chip) could page things in and out if the unified memory. Treat unified memory as a MEGA L3 cache.

Depending on how it’s done it may not be transparent if you want the best performance. But would it work?

>I wonder if they'll just do 4x M1 Max for that.

They'll be running out of names for that thing. M1 Ultra II would be lame, so M1 Extreme? M1 Steve?

Just need to increment the 1 instead... Eventually moving into using letters instead of numbers, until we end up with the MK-ULTRA chip.

I suspect they will have a different naming convention after they get to M4.

There might be some hesitance installing an M5. You should stay out of the way if the machine learning core needs more power.

I guess by the time they get to M5, anyone old enough to get the reference will have retired.

That would really be a trip!

> M1 Steve

That would be the funniest thing Apple has done in years. I totally support the idea.

Pro < Max < Ultra < Ne Plus Ultra < Steve

And Steve < Woz, perhaps?

I'm pretty sure all their messaging is preparing us for "M1 Outrageous"

Super M1 Turbo HD Remix

M1 Ludicrous the IV

Maximum Plaid

"M1 More", would show Apple is fun again!

It seems kind of strange to have the "A" line go from smartphones to... iPads, and then have the "M" line go all the way from thin-and-lights proper workstations. Maybe they need a new letter. Call it the C1 -- "C" for compute, but also for Cupertino.

M1 Max Pro... :P

"iPhone 14 Pro Max, powered by the M1 Max Pro".

M1 Hyper or M1 Ludicrous:)

M1 Houndstooth

M1 God

Just get the guys who came up with the new name for the iPhone SE working on it. Oh, wait.

I like "X1" way more than "M2 Extreme".

Plus Ultra would at have historical precedent.

I like M1 Steve, as it can honor two people.

M1 Plaid

lol, good one! I hope marketing sees this.

Steve would be awesome, but a deal with Tesla to use "Plaid" would be perfection.

I would think they could just go to Mel Brooks instead of dealing with Tesla.

Never happen.

Elon is somewhat toxic these days...

iM1 Pro.

M1 Magnum XL

Epic M1 fit good

M1 Greta in 2030 (when it is "carbon neutral")

You don't just "fuse" two chips together willnilly. that was designed in from the beginning for the architecture for future implementation.

My guess would be an M1 Ultra with additional modular expansions for specific purposes. I.e. instead of having a GPU, you could add a tensor processing module if you need to do machine learning, or a graphics processing chip for video production, and so on.

its a chiplet design. whenever people ask what we going to do after 1nm...well, we can combine two chips into one

The CPU trend nowadays seem to be combining chiplets and having performance and efficiency cores.

Throwing more silicon at it, like this, sounds extremely expensive or price-inefficient.

It's at least two separate chips combined together. That makes more sense, mitigates the problem.