M1 Ultra is the last chip based on the M1

[crazy dave]

I don't dispute that. Would it still be based on M1? I would say so, just as M1 Pro is a new die but is based on M1.

You were:

Why would it be entirely new dies? They can use current MCM to duplicate M1 Ultra. It's not ideal, but it works. Based on what we know, something call Jade 4C-Die is coming. That doesn't sound like a whole new chip to me.

This won't work for all the reasons stated above. A new "W1" die with icestorm/firestorm to make a x2/x4 MCM could not be the M1 Max which maxes out at x2 in the M1 Ultra. The only reference we’ve seen is a 12-core iMac and I think there’s a chip designation we haven’t seen but no Jade code name that I know of for a new individual die.

 

[LonestarOne]

Hector Martin actually predicted this some time ago. No 4-die support on M1 Max.

The chip layout shown at today's event supports that idea as well. Connector on only one edge.

 

[LonestarOne]

Yes, but the M1 Max 24 to 32 core upgrade is $200, so 2 x M1 Max upgrades should be $400, not $1000.

If we assume the M1 Ultra has two M1 Max dies that are then "joined", it looks the majority of the cost is actually the "silicon interposer" that does the joining, rather than the GPU core upgrades.

However, you are already paying this cost with the jump from the M1 Max to even the binned (48-core GPU) Ultra, so I don't know where the $600 "mark-up" for the GPU cores comes from.

You're confusing cost with price. In business, the idea is to charge more for products than what they cost you to make.

 

[Adult80HD]

You're confusing cost with price. In business, the idea is to charge more for products than what they cost you to make.

He's also confusing yield--a non-linear thing--with a desire for linear scaling in price. The two are incompatible.

 

[Fomalhaut]

You're confusing cost with price. In business, the idea is to charge more for products than what they cost you to make.

Yes indeed....but the differential is notable.

I think you are right though; people who really need a 64-core GPU are almost certainly generating income from the machine, and the "entry price" is just another business cost.

Apple may even have to produce the two M1 Max dies with some common process that requires both to be "matched", and the yields on creating 2 x 32-core matched dies may be lower, driving up the price.

 

[ultrakyo]

There is still hope for Quad Max chip for Mac Pro lineup by connecting the Max to I/O die (M1 Extreme?). Perhaps technically they can scale more than 4 Max?

 

[Gotfredsen]

There is still hope for Quad Max chip for Mac Pro lineup by connecting the Max to I/O die (M1 Extreme?). Perhaps technically they can scale more than 4 Max?

this is what I am talking about . Apple M1 Ultra Extreme with UltraFusion for a Quad M1 Max config . That will bring 40 Core, 128 Core GPU, 1.6TB/sec bandwith with up to 512 GB Unified Memory and 16 TB of storage

 

[senttoschool]

There is still hope for Quad Max chip for Mac Pro lineup by connecting the Max to I/O die (M1 Extreme?). Perhaps technically they can scale more than 4 Max?

They could do this, yes.

But they're more likely to do this with the M2 Ultra than the M1 Ultra.

I don't know nearly enough about MCM/interposers to understand whether transparent multi-GPU (OS sees one GPU instead of 2 or 4) would work with a I/O die. As far as I know, the connect in the M1 Ultra is 2.5 tb/s which is enough to make 2 GPUs work as one. But how would introducing an I/O die affect the bandwidth required to make 4 GPUs look like 1?

 

[Andropov]

No, they can't do that. Jade C-die (M1 Max) supports up to 2 chips. If they do release a 4-die, it won't be based on the M1 Max die. And after yesterday's keynote, Apple has been very clear on their opinion on having any kind of I/O between their dies that makes each die to appear as a separate device to software developers.

If there's a M1-based SoP with 4 dies, it won't be using the M1 Max cores. There have been clues in the firmware pointing to another M1-based SoP that is not some version of the M1 Max, but at this point I doubt that whatever that was is coming to market.

 

[Kardinal1911]

In case you didn’t catch this little tidbit in the presentation.

This highly suggests that the 4x version will be based on the M2 instead.

It makes sense if they plan to release the M2 Air before converting Intel Mac Pros over to the M2 GigaUltra.

Also, a bit disappointed that we didn’t see the M2 today. But to me, it never made sense to see a chip as important as the M2 launch in a 6 year old design. Made zero sense. It should launch with the brand new MacBook Air and the new iPad Pro.

As an Apple shareholder, I would like to see a new MacBook Air 14” and a brand new MacBook Air 16” instead of rehashing the 6 year old MBP 13”.

And most importantly of all, release a $750 MacBook SE in the shell of the current Air. This will be the most important Mac of all.

Timestamp of Apple saying the M1 Ultra is the last chip in the M1 family:

I hate to tell you but Apple is not jumping into the M2 space until at least 2023. MacPro may have its own unique Apple Silicon not based on any particular M1 Architecture. I believe that if the M2 were coming we would've seen it yesterday. Now that all ipads have M1 chips, and Macbooks have either M1, Pro, Max, and the studio is now on Max/Ultra Apple will sit on these chips for at least another year. Intel has no real plan to catch up so they can squeeze as much life as possible out of the current generation. I honestly think this makes sense and gives other forms of technology a chance to catch up. For instance Face ID components, screen tech, etc... I think we won't hear solid rumors of M2 until at least Fall... with products launching next spring .

 

[TechnoLawyer]

the m2 has no purpose in ipad pro currently. apple continues to gimp iPadOS such that the M1 itself already stands as overkill.

The only real purpose seems that the ipad should stay fast and fluid running ipados for a good 5+ years. Android ruins fluidly on the latest qualcomm chips which are on par with what an A13, and if I had to guess, I'd bet ios/ipados is way more optimized.

It would be nice if they'd un-gimp ipados, but $750 for a 256gb M1 tablet that will run fast and fluid for many years isn't too bad.

 

[quarkysg]

The people who have been reverse engineering M1, M1 Pro, and M1 Max to port Linux to them found, several months ago, that (a) the interrupt controller in M1 Pro/Max supports multi-die configs and (b) the possible configurations are one die and two die. No support for four die.

Actually, if the interrupt controller only returns 1-bit of information when an interrupt occurs, it's not enough to tell the OS what happened, because knowing which die has generated an interrupt is not enough for the interrupt controller to do anything. The OS would need to know which component generated the interrupt so that it can then run the interrupt routine attached to that component.

When an interrupt occurs and the component information is delivered via the data bus of the M1 SoC, that means that the interrupt controller basically only need 1 bit of information (i.e. interrupt on or off).

In this case, basically many M1 Max dies can actually be connected to the UltraFusion fabric.

But it is probably the M2 Quad Ultra (for a 4 die combo) for the Mac Pro then, since John Ternus said the M1 Ultra is the last M1 family SoC.

 

[JPack]

But it is probably the M2 Quad Ultra (for a 4 die combo) for the Mac Pro then, since John Ternus said the M1 Ultra is the last M1 family SoC.

The biggest question is: What is an M1? What's the definition of M1?

The combination of Firestorm and Icestorm cores? If Apple makes something with 20 P-cores and no E-cores, is that considered an M1? Apple might say that's a workstation W1 chip. Some of us might say that's still an M1 because it uses Firestorm cores.

 

[quarkysg]

The biggest question is: What is an M1? What's the definition of M1?

My take would be that as long as they are based off the M1 SoC IP cores, they are in the M1 family.

Edit: It make sense from an OS point of view for the existing crops of Apple Silicon Macs to be based off the M1 SoCs as it means not much effort will be required (I think) to support these SoCs.

Going to M2 will likely happen post WWDC with a new version of macOS, so I think M2 based SoCs Macs will be announced in WWDC.

 

[synonys]

M2 will likely be based on ARMv9.

 

[Juuro]

I'm pretty sure there will be no new die unless Apple decides to go straight to the M2 family/architecture for the Mac Pro. I think the most logical thing would be to have a (truly) modular Mac Pro which you can configure with 1, 2 and maybe even 3 or 4 M1 Ultras. Maybe even user exchangeable. The would look like these large GPU modules with huge thermal systems on top. And yes, that would result in efficiency loss compared to the UltraFusion method but it would still be ok for a machine like the Mac Pro. So with 2 M1 Ultras the power would not increase by 100% but maybe by ~80%. WWDC would be a great place and time to at least preview this system.

I think we don't need to overthink John Ternus' comment about "one last chip for the M1 family". I think for Apple family and architecture ist interchangeable. Ans also it doesn't make sense to design a specialised die for an extreme niche product.

It also differs from AMD and Intel in who they market their server/workstation chips to. AMDs and Intels server/workstation chips are marketed for companies which build servers. Apple uses their server/workstation chips only by themselves so they have to market them towards the end customer in this case professionals. If they now call the Mac Pro chip a completely different name that is not relatable to one of the existing chips it would even hurt Apple. Because by now many Pros own a MacBook Pro with M1 Max or at least heard praises of colleagues of some M1 device. So it makes sense to use this well known name with a positive conotation also für the highest end Mac.
And the other way around: Apple doesn't make much money with the Mac Pro. It sells in low quantities. The Mac Pro for Apple is an image device. A device where they can flex their muscles and put all of their technology know how into it almost regardless of cost. It has some kind of positive effect when one hears Apple builds one of the most powerful computers a normal person can buy even for some people who are only in for a Mac mini.

 

[crazy dave]

Actually, if the interrupt controller only returns 1-bit of information when an interrupt occurs, it's not enough to tell the OS what happened, because knowing which die has generated an interrupt is not enough for the interrupt controller to do anything. The OS would need to know which component generated the interrupt so that it can then run the interrupt routine attached to that component.

When an interrupt occurs and the component information is delivered via the data bus of the M1 SoC, that means that the interrupt controller basically only need 1 bit of information (i.e. interrupt on or off).

In this case, basically many M1 Max dies can actually be connected to the UltraFusion fabric.

But it is probably the M2 Quad Ultra (for a 4 die combo) for the Mac Pro then, since John Ternus said the M1 Ultra is the last M1 family SoC.

Hector’s analysis came to a different conclusion. Basically Apple has an interrupt controller designed to accommodate up to 8 dies in theory but the ones on the Max is only able to handle two. In his opinion, the Max cannot be extended to more than two dies.

https://twitter.com/i/web/status/1466582101890060289

Edit better one:

https://twitter.com/i/web/status/1498388636706156544

 

[iDron]

Because they literally said that the M1 Ultra is the "last chip in the M1 family" during the keynote.

Well what if they call the next thing then M1 Ultra Duo? Still an M1 Ultra, but two of them? Maybe even visible in the system as two chips.

 

[crazy dave]

Well what if they call the next thing then M1 Ultra Duo? Still an M1 Ultra, but two of them? Maybe even visible in the system as two chips.

That would still be a member of the M1 family.

For what it’s worth of the known code numbers for the Mac chips (sans M1 which is actually an iPad chip) have gone from 6000 to 6002 so far … the remaining 6000 chip? 6500. That indicates something different is coming if it is for the Pro.

https://twitter.com/i/web/status/1498317101245034502

 

[Waragainstsleep]

I think its worth looking at the entire product lineup for clues.

The iPhones all get the latest A1X CPU with the SE lagging a little behind and updated less often;
The base iPad is also A1X
iPad Air is now M1 and will stay with the base MX CPU but lagging the Pro
iPad Pro will probably go to M2 next time;
MacBook Air will go to M2
I'd expect the 13" MBP to be discontinued;
MacBook SE maybe stays M1 and updates less often like the iPhone (Not sure about this even happening as it will cannibalise other items but would definitely bring new users in with a lower price point. Great for offices with basic needs but laptops are different to phones which are subsidised by carriers which means more consumers buying overpowered devices);
14" & 16" MBP goes M2 Pro/Max;
Mac Mini could also keep an SE version with M1 but otherwise goes M2+;
iMac 24" goes M2, do we ever get a 30" now? Not sure we do;


The Mac Studio is an interesting beast. Does it replace the 27" iMac? Maybe it does. What it definitely replaces is an important subset of Mac Pro users. Now that video pros can run 18 simultaneous streams of 8K video on a Mac Studio with 4 or 5 displays, what need do they have for a Mac Pro? Not much I'd say. I imagine its the same story for audio producers who could run crazy simultaneous Logic tracks on a base M1 Mac Mini so a Studio is going to be a multi-year luxury for them too.
So who is actually going to need the next Mac Pro? To me it looks like people doing high performance work, some devs, scientists maybe? Whoever it is, its going to be aimed at a different market than before I think. This was always a big point of Apple Silicon. Versatility. So if its aimed at a small niche, a quad-die M2 would make more sense. On the other hand, your new architecture is going to take time ramp up the yield so it makes sense to leave the top level stuff until last. But they said its the last M1 so if its going to be based on M1 cores, then its going to have to be significantly different in order to do quad die, or more CPU cores to GPU, or whatever it is they have in mind. One would imagine that this new Pro system is going to need PCI-E5 in order to compete with the opposition too. M1 doesn't have that yet.

Whatever its based on, it will be different enough to warrant its own name I think.
I'm having more trouble deciding who they are going to aim it at. The Mac Pro has a crucial difference to other models: They don't have to worry about power consumption and heat. But thats the exact opposite of what M1 was designed for. And with video, audio, print and a number of others more than satisfied by the Mac Studio, are Apple really going to go to the lengths required to design something so different and so complicated for such a tiny niche market?
They could look into getting back into rack servers but the advantage there is again low power consumption and heat. Not the Mac Pro at all.
Apple has increasingly moved to services in recent years. Maybe they are planning to be their own best customer for a new MacServe that sort of doubles as a Mac Pro for high performance needs. Something like the old Xserve Cluster Nodes so hardcore devs can compile rapidly on their own racks? Or on Apples cloud based ones?
If thats the plan then they need to have some disruptive new enterprise services in the offing. Can't see what else they will offer consumer services except more storage. Still they have a few data centres full of HP servers they'd maybe like to save on upgrading. I've been rambling for several paragraphs now....

 

[JMacHack]

Why? Because the marketing team can't use the word "Extreme"?

We've already seen how accurate this prediction is, not sure why there's suddenly doubt about 4x Max.

Jade C-Chop
Jade C-Die
Jade 2C-Die
Jade 4C-Die

Is it possible that Jade 4C just didn’t work out? There’s no real reason to discount Apple’s direct statements in favor of a rumor.

 

[planteater]

((one more product to move to AS)+(M1 Ultra is the last M1 SoC))=(M2 Quad will be the first M2, and it will be released in the Mac Pro)

 

[senttoschool]

Is it possible that Jade 4C just didn’t work out? There’s no real reason to discount Apple’s direct statements in favor of a rumor.

Or that Jade 4C was just an internal test and was never intended to be released?

At this point, there is no reason to think that Apple will use the M1 design for 4x M1 Max.

 

[JPack]

Is it possible that Jade 4C just didn’t work out? There’s no real reason to discount Apple’s direct statements in favor of a rumor.

Based on Apple’s execution performance, there’s no reason to think that.

M1 is a only marketing name with no clear definition. At what point does the design stop being an M1? Different cache, different ratio of P and E cores? If the workstation chip lacks any efficiency cores, Apple may brand it as a W1.

 

[dieselm]

In this case, basically many M1 Max dies can actually be connected to the UltraFusion fabric.

It's not a fabric.