Apple reportedly working on M1 Max Duo chip for next generation Mac Pro

Cal Jeffrey

Posts: 3,217   +883
Staff member
Rumor mill: Apple's latest silicon, the M1 Pro and M1 Max, have barely come off MacBook Pro assembly lines, and rumors are already circulating about what's next for Apple silicon. Rather than going back to the drawing board and designing chips with even higher specs, why not just knit two Max SoCs together?

As you are probably aware, Apple vowed to completely ween itself off of Intel components by the end of next year. So far, it is right on track. The MacBook Air, 13-inch Macbook Pro, Mac mini, and the 24-inch iMac are already sporting M1 silicon, and Apple just unveiled the redesigned 14- and 16-inch MacBook Pros with their next-generation M1 Pro and M1 Max chips. That only leaves the 27-inch iMac and the Mac Pro workstations still on the Intel nipple. Surely this will change next year.

Noted Apple leaker Mark Gruman says Apple plans to power the Mac Pro with dual M1 Max SoCs (M1 Max Duo) on the base models and four (M1 Max Quadro) on high-end units. That's a minimum of 20 compute cores and 64 graphic cores with up to 128GB of RAM and double all that for the Quadro.

Hector Martin, who is in the process of porting Linux to M1 Pro/Max Macs, indirectly backs up Gruman's claims with evidence he discovered in macOS.

"The macOS drivers have plenty of multi-die references, and the IRQ controller in the M1 Pro/Max is very clearly engineered with a (currently unused) second half for a second die," tweeted Martin. "For the technically minded: it's a second set of config/mask/software-gen/hw-state registers, and the hardware inputs are all idle but you can software-gen IRQs in that block just fine and they get delivered with a die-id of 1 in the top 8 bits of the event register."

As far as cost, speculation is that top configurations will run an eye-watering $50,000 or more, but that is not surprising since buyers can configure the current Intel-based Mac Pro to cost nearly $55,000. That said, low-end units should start within the $6,000-$6500 price range.

Gruman only spoke of the Mac Pro receiving the Duo and Quadro silicon, so that just leaves the 27-inch iMac sporting Intel chips. However, Max Tech host Vadim Yuryev says there is strong evidence that Apple will use an M1 Max Duo in the larger iMac variant (video above). He points out that simply using a single M1 Max would be a downgrade from current Intel-based iMacs.

Designing a whole new chip is a possibility. Rumors are already circulating that a 40-core M2 chip is on track for 2023, but that falls beyond Apple's deadline for ditching Intel. Yuryev is convinced that it will probably use binned and unbinned M1 Max Duos for prosumer and professional configurations, respectively, rather than a whole new die. He suspects Apple to reveal the next iMacs as early as March of next year. Again, this falls in inline with Cupertino's roadmap for silicon independence without sidelining the M2 predictions.

Permalink to story.

 

Tom Yum

Posts: 133   +297
Makes sense, M1 Max is already a very large die (432mm2) despite the advanced 5nm lithography, increasing the core count whilst maintaining a monolithic core isn't really possible unless they reduce the GPU size which would negate the unified architecture they are going for. Will be interesting to see how a M1 Max duo manages inter-die communication, given the memory architecture used.
 

Dimitriid

Posts: 1,337   +2,625
Will it be just the same die M1 max chips on a single dual-quad socket mobo (Well no "socket" but you get what I mean)?

Because while getting the extra GPU part to work shouldn't be an issue, I wasn't aware that most of the workstation grade software or things people might be doing with Premiere, Final Cut and such would be able to just leverage more than 1 CPU: such programs are very specialized for server applications because that's where you see more than one CPU per mobo/rig.

Otherwise it seems like you'd be paying for the full 2nd (and 3rd and 4th) M1 Max chip but only utilize a small portion of it that has dedicated GPU cores or the ML ones but t he CPU ones will be just not utilized much at all unless you do something like virtualization (And on what's supposed to be a pro workstation, why would you? It's not a server)
 

hahahanoobs

Posts: 3,869   +1,920
Still waiting for Intel or AMD to make anything even remotely close in terms of performance per watt.
Enter Alder Lake.
Enter CUDA.

Anyone making BIG money with a PC will choose perf over power more times than not. And they def aren't going to choose a major tool that is really good at some things and worse in other things. Not this early in M1's life. Wake me when M1 has this hype two years from now. CUDA just got better in Blender where M1 is strong and rumours already put ADL above the M1 in perf so no one is resting. M1 GPU is just a joke right now so that will take time.

At the very least wait for the major players to respond before predicting their end. I'll say it again. M1 is no x86 killer from what I've seen, but we'll see what the future brings.
 
Last edited:

Shadowboxer

Posts: 1,884   +1,497
Enter Alder Lake.
Enter CUDA.

Anyone making BIG money with a PC will choose perf over power more times than not. And they def aren't going to choose a major tool that is really good at some things and worse in other things. Not this early in M1's life. Wake me when M1 has this hype two years from now. CUDA just got better in Blender where M1 is strong and rumours already put ADL above the M1 in perf so no one is resting. M1 GPU is just a joke right now so that will take time.

At the very least wait for the major players to respond before predicting their end. I'll say it again. M1 is no x86 killer from what I've seen, but we'll see what the future brings.
You are incorrect, people with “BIG” money will choose the expensive option that suits them most, this isn’t defined by hardware but software. There are a LOT of pro users who prefer Apple software to Windows, this is where Apple are strongest. Alder Lake with CUDA acceleration doesn’t come anywhere near M1 levels of efficiency. They compete on absolute performance but use several times more power to do it.

But the fact is that M1 is absolutely pissing all over Intel and AMD right now and it’s just getting started. We haven’t seen how far this tech can scale at all. Believe me if they make a chip that uses the same power as an Intel or AMD CPU it will humiliate it in performance.

The M1 itself is not an “X86 Killer”. Time is the X86 killer, it’s old, ARM is more efficient. That’s why MS and Apple are both moving to it.
 

Wrinkle

Posts: 91   +68
You are incorrect, people with “BIG” money will choose the expensive option that suits them most, this isn’t defined by hardware but software. There are a LOT of pro users who prefer Apple software to Windows, this is where Apple are strongest. Alder Lake with CUDA acceleration doesn’t come anywhere near M1 levels of efficiency. They compete on absolute performance but use several times more power to do it.

But the fact is that M1 is absolutely pissing all over Intel and AMD right now and it’s just getting started. We haven’t seen how far this tech can scale at all. Believe me if they make a chip that uses the same power as an Intel or AMD CPU it will humiliate it in performance.

The M1 itself is not an “X86 Killer”. Time is the X86 killer, it’s old, ARM is more efficient. That’s why MS and Apple are both moving to it.

I’d suggest watching this space a year or 2 before we declare anything dead. Apple made extremely impressive cpu's but on some points the competition is still ahead. Also, pretty much the entire world is using x86, it won't go away anytime soon.
 

Shadowboxer

Posts: 1,884   +1,497
I’d suggest watching this space a year or 2 before we declare anything dead. Apple made extremely impressive cpu's but on some points the competition is still ahead. Also, pretty much the entire world is using x86, it won't go away anytime soon.
Sure. I’m willing to take the bet that X86 is on its way out and ARM is set to replace it. I don’t think this is really up for debate.

The problem with comparing M1 to Intel/AMD is that they both use completely different software. In absolute performance they can be actually quite close when you can compare them. But there is no denying the fact that Apples chips are using a fraction of the power of the X86 parts to compete at the same level.

But whether it takes 2 years or 10 years. I firmly believe X86 is on the decline. This is a good thing, more efficicent tech means smaller tech, it means lower power bills, it means your performance doesn’t change if your using a battery or you’re plugged into the wall. I don’t see how anyone can have a problem with it. Well, except Intel or AMD shareholders of course…
 

hahahanoobs

Posts: 3,869   +1,920
You are incorrect, people with “BIG” money will choose the expensive option that suits them most, this isn’t defined by hardware but software. There are a LOT of pro users who prefer Apple software to Windows, this is where Apple are strongest. Alder Lake with CUDA acceleration doesn’t come anywhere near M1 levels of efficiency. They compete on absolute performance but use several times more power to do it.

But the fact is that M1 is absolutely pissing all over Intel and AMD right now and it’s just getting started. We haven’t seen how far this tech can scale at all. Believe me if they make a chip that uses the same power as an Intel or AMD CPU it will humiliate it in performance.

The M1 itself is not an “X86 Killer”. Time is the X86 killer, it’s old, ARM is more efficient. That’s why MS and Apple are both moving to it.
Efficiency is all it has. That's why it's the focus of every argument.
We'll see what happens in the next 2 years, but until then, perf still trumps power efficiency.

Think of it like that 8000Hz polling rate mouse from Razer. It's great when it works, but who's going to switch to a mouse that has the wrong feel, weight and software? Not a ton when the mouse is great when it has support in some of your games, but pants in ones that don't resulting in high CPU usage and choppy mouse movements. That's the M1 right now.
 
Last edited: