There won't be an M2 Ultra this year. I would be totally shocked if there were. M2 yes, M2 Ultra definitely not (unless it's somehow utilized in a Mac Pro, but I don't expect that).
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Testing at Least Nine New Macs With Four Different M2 Chip Variants
- Thread starter MacRumors
- Start date
- Sort by reaction score
The Mac Pro has to use something better than the M1 Ultra, which Apple said would be the final iteration of the M1.
“Two neighbours came before Jupiter and prayed him to grant their hearts’ desire. Now the one was full of avarice, and the other eaten up with envy. So to punish them both, Jupiter granted that each might have whatever he wished for himself, but only on condition that his neighbour had twice as much. The Avaricious man prayed to have a room full of gold. No sooner said than done; but all his joy was turned to grief when he found that his neighbour had two rooms full of the precious metal. Then came the turn of the Envious man, who could not bear to think that his neighbour had any joy at all. So he prayed that he might have one of his own eyes put out, by which means his companion would become totally blind.
Vices are their own punishment.”
Well the good news is that we don’t have that long to wait until we find out. I wouldn’t be terribly shocked by seeing it announced in the next 6 months.
Wouldn't work, you need twelve Ultras to achieve 1.5TB of RAM; it's basic multiplication... ;^p
To be fair, they did not specify the generation of Mn Ultra... ;^p
5 pages discussing variants and not a Loki reference to be found.
What happened to this place!! Shame!
That said, I held off getting the M1 iMac, I’ll wait it out for an M2 version.
What happened to this place!! Shame!
That said, I held off getting the M1 iMac, I’ll wait it out for an M2 version.
Any time you see me use the word "variant" or "variants" in my posts (which I do rather than "version" or "versions"), I am literally referencing Loki, y'all just don't know it...! ;^p
that’s always a losing proposition. There will always be a next generation just over the next hill. Get what you need when you need it and don’t worry about the next one. Especially as the next gen is not going to be dramatically faster than this one.
Decoupling the M2 Pro from the M2 Max makes about zero sense. The volume of the M2 Pro would not be high enough to have a almost completely forked off design just to support that.
IF the Pro and Max has the same number of CPU cores then they are probably the same CPU cluster designs. Apple's modus operandi is to reuse design elements across SoC packages. It is not to come up with as many different dies as possible. (e.g., the M1 Ultra is literally just two M1 Maxes 'glued' together. One die for two products. And a minor designed 'chop' for a smaller Pro die that does 98+% reuse. )
More likely Max is same 8+4 than the Pro would be. Ultra 16+8. Extreme gets 32+16. Major issue have is that TSMC N4 isn't some super wonder in density increase. Die space wise and bandwidth wise it is just cheaper to add E cores. CPU core aggregate is still likely provisioned a smaller subset of memory bandwidth. Perhaps toss the P cores some faster clocks and a L2 cache size bump to get to a higher performance number.
If adding two GPU cores works at the M2 level then reusing that 10 cluster over the rest of the line up is more likely. That would likely be where the core count bump goes. [ Given more often bins/segments on GPU cores than CPU cores it is the price ladder they prefer. And like the A12X could have spare GPU core(s). ]
The Kraken thing is just smoke. UltraFusion for just two 32GPU max dies is running at 2TB/s. There little reasonable expectation going to scale that inter die bus bandwidth up 2-4 times on top of the that and still keep "seamless" Unified, uniform memory semantics. If Apple was going to toss "Unified memory" out the window then maybe, but that doesn't look likely at all.
Even getting the the M2 Extreme would be difficult. Another 100 GPU cores past that in terms of bisection bandwidth required to keep them feed... that is getting past the point where will probably run into scaling problems.
My prediction is that the new Mac Pro will announced in less than 2 months.
Ugh, no mention of an iMac 27”. Literally the best computer Apple has ever made and they’ve either killed it or are really dragging their feet.
When Apple said there was only "one more" Mac to be updated - the Mac Pro, i think they made it clear that there won't be a 27" iMac (unfortunately).
So maybe more like...?
| SoC | CPU | Core Types | GPU | RAM |
|---|---|---|---|---|
| M2 | 8-core | 4P/4E | 10-core | 16GB |
| M2 Pro | 12-core | 8P/4E | 20-core | 32GB |
| M2 Max | 12-core | 8P/4E | 40-core | 64GB |
| M2 Ultra | 24-core | 16P/8E | 80-core | 128GB |
| M2 Extreme | 48-core | 32P/16E | 160-core | 256GB |
I was just having fun...!
Opens calendar... "Hmm... Well, the math checks out, with WWDC 2022 on the sixth of June..." ;^p
I’d rather kill the “affordable MacBook Pro”. It’s an asinine idea that doesn’t age well and costs too much for what it offers - barely an increase over the M1 MBP for M1 MBP 13”. 2023 will make this even worse worthy.
I think Apple is going to lean away from any All-In-One iMac beyond the entry level Mn 24" variant...
Anything with a more powerful SoC or larger screen will cost more, and it might be(come) bad optics to have a "double disposable" (meaning a non-upgradable computer & a display you cannot use if the computer dies) product at those higher price points...?
Please do not solely leave this on virus! It's just as much Tim cook's fault to allowing Ar assemblies in since 2008!
Dylankt tweeted that there was a 12 core M1 for the iMac Pro".
There is no iMac. There are no more M1's.
And you declared at the beginning of that thread that it was 10P+2E ; not Dylankt.
If Apple is trying to keep these M2 Pro and Max die sizes approximately the same size then 8P+4E is more likely. Going from N5 to N4 doesn't buy a large increase in realized density. If Apple is throwing 2 more GPU cores at each GPU cluster (10 cores instead of 8 cores ) then probably have used up a decent amount of the incresed transistory budget (and space). Toss in some cache size boosts , and some other incremental improvements , then probably just have a E core budget increase budget left. Additionally, that is less bandwidth pressure to add on top. ( the GPU clusters are soaking up more now with the larger core count increase. )
As for as the DUo M2 ... that doesn't make sense in the context of 12 cores. Neither does it make sense given the UltraFusion connector takes up an entire edge of a vastly larger M1 Max die. M1/M2 die size doesn't have copious empty, unused spaces on its edges. Unless nuked several of the memory controllers there is no space for a much smaller "Fusion" connector. Finally the "redundant" SSD controller, secure elements , uncore function elements that only need one of for a Mac is a substantively higher percentage of a plain M1 class die. So doubling those up is a bit of a die space waster. [ Vast buik of M1 Max die is the GPU , internal bus , Video de/encoders, and memory. The 'wasted' uncore ( extra SSD controller , secure element , etc) space is relatively minor. ]
Can’t believe people people are freaking out and worried about the Mac Pro laptop that was released A few months ago will be updated this year.
as the article mentioned, Apple is testing these chips. Anyone suggesting these will all be released later this year is just hyperbole.
It is expected that this would be tested internally to find issues and to refine the chips way before they are planned for release. Maybe the M2 base chip will be released this year. I can say for certain the M2 Max M2 Ultra chip is not coming out this year, so chill out.
as the article mentioned, Apple is testing these chips. Anyone suggesting these will all be released later this year is just hyperbole.
It is expected that this would be tested internally to find issues and to refine the chips way before they are planned for release. Maybe the M2 base chip will be released this year. I can say for certain the M2 Max M2 Ultra chip is not coming out this year, so chill out.
While Apple’s notebook offerings offer something at any price point from $999 to $6,000, it’s desktop lineup has a huge gap between the $699 and $1,999 price points. The reason? Not a single desktop machine offers an M1 Pro, either in an iMac or a standalone machine.
Making sure that they correct this during the M2 generation should be a priority, because very few people can afford the $1,300 jump to the Mac Studio, but they would shell out hundreds more for a higher end Mac Mini or iMac.
Making sure that they correct this during the M2 generation should be a priority, because very few people can afford the $1,300 jump to the Mac Studio, but they would shell out hundreds more for a higher end Mac Mini or iMac.
I agree, it would make sense for the iMac and Mac mini to offer an MX pro chip, which are now only used in the 14”/16” MacBook Pro
This is the main reason I am holding out for the M2 line before I finally upgrade my aging mid 2012 15". I honestly don't need the Pro, but I like to buy big and keep for many years. I also prefer not to buy first-generation stuff just to avoid some of the growing pains of change. So I hope they do release the M2 Pro's this fall.
Wow. Really? So that means Apple won't be discontinuing existing products and instead keep them updated?!
The M1 has
L1 Cache
192+128 KB per core (performance cores)
128+64 KB per core (efficient cores)
L2 Cache
12 MB (performance cores)
4 MB (efficient cores)
and up to 128GB memory per M1 max chip.
So if they needed more memory could they add them in a hybrid mode. So if you in the M2-Quad would add an interposer for normal DDR5 memory. Then you have the L1, L2 cache then unified memory 128x4 + up to 1TB+ slower DDR5 (ad sticks as you please) to do the memory hungry jobs.
That would leave 128G of memory per die for graphics, ML and high performance IO and use the slow memory for huge memory sets for scientific computing and similar.
We all remember the hybrid hard drives where a 128GB SSD acted as a cache for the large spinning disk.
By adding 4 max chips there will also be enough performance to add additional storage cards.
Each M1 max chip has four thunderbolt 4 controllers. In a quad config there will then be 16 controllers, each with 40GB bandwidth that could be used both for internal and external expansion.
Of course a 16 lane PCI 5.0 is 63GB but maybe it’s possible to do a multi lane TB interfaces.
Remembers connecting high resolution displays that used multiple display ports.
L1 Cache
192+128 KB per core (performance cores)
128+64 KB per core (efficient cores)
L2 Cache
12 MB (performance cores)
4 MB (efficient cores)
and up to 128GB memory per M1 max chip.
So if they needed more memory could they add them in a hybrid mode. So if you in the M2-Quad would add an interposer for normal DDR5 memory. Then you have the L1, L2 cache then unified memory 128x4 + up to 1TB+ slower DDR5 (ad sticks as you please) to do the memory hungry jobs.
That would leave 128G of memory per die for graphics, ML and high performance IO and use the slow memory for huge memory sets for scientific computing and similar.
We all remember the hybrid hard drives where a 128GB SSD acted as a cache for the large spinning disk.
By adding 4 max chips there will also be enough performance to add additional storage cards.
Each M1 max chip has four thunderbolt 4 controllers. In a quad config there will then be 16 controllers, each with 40GB bandwidth that could be used both for internal and external expansion.
Of course a 16 lane PCI 5.0 is 63GB but maybe it’s possible to do a multi lane TB interfaces.
Remembers connecting high resolution displays that used multiple display ports.