Consolidated Apple Silicon Mac Pro rumors.

[theorist9]

Well, that’s the big question. I have no idea. My best guess is no. Porting Everest/sawtooth to 3nm would likely be no less effort than porting M2. I don’t *think* they would “skip” M2 in the high end products.

But who knows.

Just found this from Nikkei Asia (a sister organization to the London-based Financial Times). Consistent with your guess, they say we won't see M3 until 2H 2023, and that it's is going to be on N3E (the successor to N3)

I gather Nikkei Asia is a reputable publication, so it seems notable that they're reporting this as fact rather than rumor.

They do get this wrong—the A16 is on N4P, which is a 5 nm process—but then that's a common mistake caused by TSMC's misleading nomenclature:

"In 2022, only the premium iPhone 14 Pro range has adopted the latest A16 core processor, which is produced by TSMC's 4-nm process technologies, the most advanced currently available."

Apple to use TSMC's next 3-nm chip tech in iPhones, Macs next year

Move will mark industry's first adoption of updated production technology

asia.nikkei.com

 

[Colstan]

So that you no longer need a "members only" jacket to access the details through his newsletter, Gurman has released an article that expounds upon his previous rumors about new Macs launching in March. Most of this has been covered ad nauseam, except the change of release date. That includes the 14-inch and 16-inch MacBook Pro featuring M2 Pro/Max, the new Mac mini with a standard M2 or M2 Pro option, and the Apple Silicon Mac Pro itself.

Of course, he made certain to cover his ass about release dates and how he couldn't possibly be wrong:

When I first reported on the new MacBook Pros in June, I said that Apple was aiming for a release between the end of 2022 and early 2023, so I wouldn’t consider a March release to be late by any means.

Gurman rarely offers a direct opinion, usually tap dancing around the issue, but he was unusually clear in addressing his personal thoughts about the next Mac Pro:

But the first Apple Silicon Mac Pro is clearly running behind the company’s own self-imposed timeline. When it announced the transition to homegrown chips in 2020, Apple said the move would take about two years. The revamped Mac Pro, coming next year, will clearly miss that schedule.

But I think we have a pretty clear reason why, and it’s not a bad thing: The machine will be superior to what Apple originally intended to offer.

As I wrote recently, my belief is that the first non-Intel Mac Pro will have options for 24 and 48 CPU cores and 76 and 152 graphics cores—along with up to 256 gigabytes of memory.

I believe that Apple had originally planned to use a variation of the M1 chip, but at some point made the decision to hold off until versions of the M2 with more CPU and graphics cores are available.

That seems like the right decision. The core M1 architecture is based on the A14 chip from 2020, while the basis of the M2 is quite a bit newer. Apple also may be waiting until it can build chips using the 3-nanometer process (the first M2 chips and the M1 are based on 5-nanometer technology). So it’s probably worth the wait.

I'm glad Gurman is here to tell us these things, because we would have never been able to figure this out on our own.

 

[B01L]

A few alternative thoughts I had over at "the other place"...

Six slot chassis, but one slot is for the SoC daughter card...?

(1) SoC slot - proprietary ultra high-speed / low latency connection to MPX & PCIe slots
(2) MPX slots - x16 Gen5
(3) PCIE slots - one x16, one x8, one x4, all Gen4 (production loadout - M.2 RAID, 8K video I/O, audio DSP)

Maybe Apple takes a page from the old Silicon Graphics playbook...

SoC daughter cards, upgrade cards available but you have to send back the SoC daughter card that was replaced; maybe the same for mobos, which would mean Apple keeps a "universal layout" sort of chassis (in regards to the mobo securing within the chassis)...?

So one gets a shiny new ASi Mac Pro, a few years later the daughter card gets replaced by one with a newer SoC, or maybe just by one with more RAM...?

That happens a few times, but then the mobo is the bottleneck, so time for a new mobo...!

Would this be "cheap"...? Most likely not at all, but it would allow a modular upgrade path of sorts...?

 

[dada_dave]

Also, even though the AMD graphics card is apparently recognized inside macOS, it doesn't work. I question the "In nutshell he tell me that if Apple want to support 3rd party GPU on AS, it can available in just matter of days." statement.

Some posters over there jumped on this immediately as being an indication that AMD graphics cards would be supported, but I pointed out that if you plug an eGPU into any Apple Silicon Mac, it will be recognized, but it won't function without drivers. Other than obviously putting the engineering resources into those drivers, I'm not sure whether or not Apple's design model supports third-party graphics cards in that way.

There's a certain subset of users over at MR that were certain that there would be another Intel Mac Pro. Then they pinned their hopes on external DIMMs. When that apparently fell through, they've now decided that Apple will definitely have support for third-party graphics cards. I've contended that the 2019 Mac Pro is the result of Intel's design philosophy, not Apple's. All Apple did was to design the pretty case and the oddball MPX modules, but everything else is a result of Intel's high-volume Xeon platform. For Apple, the Mac Pro is a niche of a niche, and I don't foresee them putting the engineering resources into anything other than their own silicon. I think there will be PCIe slots, that's a given, but they will be for non-GPU additions. I also believe that Apple is using the transition to Apple Silicon to cut out the PC guys from their product lines, wherever possible, including GPU providers. Heck, Apple even eliminated Intel's retimer for Thunderbolt support and replaced it with a custom solution.

So somewhat ancillary to this discussion we know that at least for the M1, eGPUs are pretty much an impossibility at the hardware level based on Hector's RE for Linux.

https://www.twitter.com/i/web/status/1538426240922963968/

https://www.twitter.com/i/web/status/1534826447596310529/

It's possible the same will be true for internal PCIe and third party graphics card. It's also possible that Apple will fix this in the upcoming chips for Mac Pros specifically to allow for their own discrete (compute) GPU cards (@leman has several good arguments why they wouldn't do so for driving graphics on macOS). If they do, at the very least 3rd party GPUs will become available under Linux, which for professional uses may be good enough for those who need it (and for gaming if Steam had its way ).

 

[Cmaier]

So somewhat ancillary to this discussion we know that at least for the M1, eGPUs are pretty much an impossibility at the hardware level based on Hector's RE for Linux.

https://www.twitter.com/i/web/status/1538426240922963968/

https://www.twitter.com/i/web/status/1534826447596310529/

It's possible the same will be true for internal PCIe and third party graphics card. It's also possible that Apple will fix this in the upcoming chips for Mac Pros specifically to allow for their own discrete (compute) GPU cards (@leman has several good arguments why they wouldn't do so for driving graphics on macOS). If they do, at the very least 3rd party GPUs will become available under Linux, which for professional uses may be good enough for those who need it (and for gaming if Steam had its way ).

I suspect that Apple may provide their own “compute“ cards, but as I, too, have previously said, I doubt they will allow third party GPUs for graphics. I wouldn’t be surprised if they have a card or two in the ML/GPGPU/encode accelerator realm.

 

[B01L]

I suspect that Apple may provide their own “compute“ cards, but as I, too, have previously said, I doubt they will allow third party GPUs for graphics. I wouldn’t be surprised if they have a card or two in the ML/GPGPU/encode accelerator realm.

I could see ASi GPGPU cards working in an (Apple-blessed) TB->PCIe chassis, like the dealio from Sonnet, but for whatever format Apple ends up on for their add-in card, MPX or "standard"...?

So optional graphics horsepower available for all the non-PCIe slot having ASi Macs, and maybe even iPad Pros...?!?

 

[Cmaier]

I could see ASi GPGPU cards working in an (Apple-blessed) TB->PCIe chassis, like the dealio from Sonnet, but for whatever format Apple ends up on for their add-in card, MPX or "standard"...?

So optional graphics horsepower available for all the non-PCIe slot having ASi Macs, and maybe even iPad Pros...?!?

I tend to doubt it, at least in the near term? Did the afterburner card work in external enclosures? I’m not sure. I think their strategy to differentiate vs Mac Studio will be to provide modularity to Mac Pro and limit it for Studio.

 

[B01L]

I tend to doubt it, at least in the near term? Did the afterburner card work in external enclosures? I’m not sure. I think their strategy to differentiate vs Mac Studio will be to provide modularity to Mac Pro and limit it for Studio.

I was more looking at it from the point of having more ASi GPGPUs out there, which might bring the cost of the cards themselves down somewhat...?

One would still need to buy powered housings of course, whether that was eGPU-style or something like the Sonnet rackmount; the former targeted at laptop & desktop (mini & Studio) users, the latter at mini & Studio users looking to rackmount their gear...?

 

[Cmaier]

I was more looking at it from the point of having more ASi GPGPUs out there, which might bring the cost of the cards themselves down somewhat...?

One would still need to buy powered housings of course, whether that was eGPU-style or something like the Sonnet rackmount; the former targeted at laptop & desktop (mini & Studio) users, the latter at mini & Studio users looking to rackmount their gear...?

I don’t think apple would be motivated to reduce the cost of the cards

 

[throAU]

Not very exotic, but what would thrill me, just noodling/brainstorming a bit:

1. Rack mount with slides (similar to the current Mac Pro rack option). Perhaps 5U or 4U height,
2. PCIE with a bunch of slots for custom designed SDR/signal processing/bulk memory cards.
3. A decent power supply. Doesn't need to be 1.4 kW. And room for an analog supply for data converters.
4. 10 Gbit ethernet
5. A back panel that can be modified for various interfaces.
6. Quiet.
7. Somewhat affordable.

Multiple 10 gig ethernet wil probably be onboard, no need to use a slot for that. Upgrade via slots for 25/40/100 gig ethernet - that is where its at for high speed networking in 2022

I'd expect to see 25/40/100 gig networking, maybe fibre channel HBAs and maybe PCIe->M.2 backplane cards for the slots.

This machine will be all about high speed throughput, so high speed storage and networking available to add outside the box makes sense. Doesn't matter how fast you can grind through media/rendering jobs without being able to get work into or out of the box and onto larger network storage - e.g., a multi tera/petabyte array for holding your high resolution media assets.

That sort of storage capacity won't fit inside the Mac Pro, not unless it is several racks large. So you need high speed network connectivity to it.

256 GB max ram capacity seems low to me for a machine with this many high speed cores. I've had 32 core servers with 512 GB for ram for a couple of years now and they aren't even fully populated.

Maybe it will have slots for secondary lower speed RAM, maybe Apple are just planning to use swap - flash is maybe fast enough if you have enough of it these days, otherwise there's optane available.

 

[dada_dave]

So some interesting news with respect to eGPUs on the M1:

https://www.twitter.com/i/web/status/1589193324954804225/

In short, it’s possible but performance may not always be pretty and the solution is definitely ugly and may never be upstreamable but may be good enough to be kept downstream if it can be modularized and made toggleable.

Also Hector says he hasn’t tested the M2 yet to see if it’s fixed and confirms that internal PCIe would be affected if it has not been.