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.
M1X worst case scenario
- Thread starter JMacHack
- Start date
- Sort by reaction score
You forgot the "doomsday" side effect that it may spit locust out occasionally or something.
This isn't as 'worst" a you make it out to be. This exact combination is a lower and lower probability the more you tie them together in a total conjunction.
First the context of what talking about here. The "X" stands for a substantively bigger die with more stuff on it. In that context. ( Let's stay 2x P cores , drop 2E cores , and add 2-4x GPU cores, one more TB port/controller, and about 2x increase in cache sizes. So end up with a 2.20-2.5c larger die. ). For example move from a 120mm^2 die to a range around 280-300mm^2 die.
1. No IPC or clock improvements or losses is a pretty much a 'win'. Generally when double cores the clock speed drops. Power dissipation problems , clock propagation issues , internal network latencies/'traffic jams' , etc.
If can get no single thread performance drop offs and linearly scale up then that there would be there is a very substantively better "uncore" improvements here. That is one of the principle things Apple would need for a substanitally bigger die.
[ In contrast, If Apple has to lean heavily on a sub-node process shrink to get scale. That can be indicative that the "uncore" isn't really up to the task of scaling. . ]
2. Moving to a 290mm^2 die would be around a 60% increase of circumference for the die. If all of that increase had to be thrown at more RAM channels to keep the GPU+Core feed with enough bandwidth then yeah that would be a scaling issue.
Frankly, more "powerful" A15 cores that have even higher bandwidth needs may not solve that issue.
The easier out would be to just 300mm^2 for the die. It is still relatively midsize ( not really a "large die" in 2021 terms. Skylake Xeon Low-core-count die was about 320mm^2 the XCC die (i.e., large ) was about 697mm^2 ).
Apple's SoC black hole asbsorbtion of the dGPU here of the MBP 16" is going to require some hefty size increase over an iPad Pro sized SoC. Not like a real black hole where the new matter is crushed to a singularity. The mass is going to be bigger if "suck it inside ". I guess more like the 'Blob'.
This can be coupled to issue 4. (more on that later ).
3. Issue 3 and 2 happening at the same time is a bit of an oxymoron. Both problems at the same time really don't happen with the very large increase in circumference and die size. Either threw the whole circumference at more memory channels ( e.g., gone from 2 RAM die stacks per die to 4 (or more) die stacks per die ) or you didn't. More capacity will come with more die stacks. And if you are 'stuck' with just 2 die stackes there is gobs and gobs and gobs of space for a raster display controllers . If give some of the increase to both size then get some movement in both ( smaller capacity and displacy deltas. )
so pick one. both is unlikely. ( Apple isn't likely throwing tons of off die edge space to some huge PCI-e complex or mainstream I/O . If this is a chiplet componet then would need some for interconnection . but a very short range interchip connection would not 'pad out' as much as the other because the interfaces are narrower and less power consuming. )
4. If have same speed but also got 50-100% more memory channlesl( went from 8 to 12 or 16 ) the same speed is far, far, far, from the end of world. You would have far more aggregate bandiwidth which is highly needed for the incrased core count.
if the die size was almost constant ( 130 versus 120 area ) and same number of die stacks attached..... yeah that would be a problem. But the whole poiint of a "X" is to be a bigger one. Far bigger die and still clutching at just 2 die stacks ? Why?
the aggregate bandwidth to the die stacks counts ... not the speed of the individual dies.
[ Optionally the cores don't talk to the memory directly. The memory controllers do (and are part of the "uncore"). Just because the cores haven't changed doesn't mean the "uncore" can't move. The internal central bus has moved. The memory controllers could too. Or had already had a broader range of memory could interface with if willing to sacrifice a high power draw. The use of the M1 of 8 memory controllers/channels is very LDDR5-ish in design. It isn't presenting as strictly 4 design. ]
5. More P cores means more multithreaded performance.
Single threaded performance is not some kind of huge problem for the current P cores. If go back to point one above and there is zero drop off of getting the same single thread outcome with all this "extra stuff" attached to the die then what is the problem there. that the chip could seamless shift between the two different types of workloads would be a signficant win. Usually to max out on one have to contribute some scracfice for the other.
6. This goes back to 2, 3. 4. If grew the chip by a substantive amount but attached an iPad Pro memory subsystem to it with just two RAM stacks and some deficient, overtasked raster display engine .... what in the world did you spend all the extra die space on???????????????
Again this is one of those oxymoron situations. You stuff a doubling or quadrupling of GPU cores (and associated cache allocations) into the middle of the die to grow it much, much bigger. That cause the perimeter to grow and then you stick next to nothing along that increased perimeter. Why? If the die grows bigger than typically means you get more I/O . Which I/O class you allocate to the increase to is flexible but the notion that is is zero I/O and just more dense compute/cache with a bridge to "no where" doesn't make much sense.
If the die stacks when from 2 to 3 because had to make some other I/O tradeoffs then yeah probably going to get sublinear performance increase with a 2x increase of cores. 4x is more likely to be sublinear.
If they only jumped to 8 GPU cores then pretty likely it would fail at the mission of matching the current MBP 16" dGPU performance envelope. if the point was to just cover the MBP 13" four point model then that would be a win. But would be losing for the MBP 16".
There are rumors of a JadeCut were the die probably is reduced (hence 'cut down' ) and likely takes a loss on memory and/or display output. But there 'M1X' would be two SKUs. The smaller one probably costs less too. If get less and pay less that isn't "end of the world" either. That is the other issue whether 'M1X' is a grouping or single die.
One single die to cover the space that the ucrrent MBP 13" and MBP 16" fill with all the associated BTO options on GPU. Doing all that with one single die is probably and issue if Apple wants to keep the costs the same for similar configuration. Especially since Apple charges such a high mark up for RAM. A four RAM die stack SoC will cost more in part solely on the RAM markup inflation that is soldered on to the SoC.
So yeah the SoC to primarily cover the MBP 13" might get capped at GPUs ... grow less ...and get capped at 3 RAM stacks instead of 4. GPU performance will be higher than the M1 but less than the 4 stack variant, but also cost substantively less.
Last edited:
I could conceivably see an "MxZ" for "Jade2C"* and "Jade4C"** said to be going into the Apple Silicon model Mac Pro. And I think we will have just one model, still a tower, and likely with at least two PCIe slots and maybe four to allow for an MPX module card (at the expense of the module taking the space of all four slots).
I do not think we will see an "iMac Pro" at this stage. And I do not believe Apple will want to put Jade2C/Jade4C into the 16" MacBook Pro for thermal and battery life reasons.
* - The 20 CPU (16P/4E) and 64 GPU core Apple Silicon SoC
** - The 40 CPU (32P/8E) and 128 GPU core Apple Silicon SoC
I do not think we will see an "iMac Pro" at this stage. And I do not believe Apple will want to put Jade2C/Jade4C into the 16" MacBook Pro for thermal and battery life reasons.
* - The 20 CPU (16P/4E) and 64 GPU core Apple Silicon SoC
** - The 40 CPU (32P/8E) and 128 GPU core Apple Silicon SoC
Depends upon the original timeline for these MBP 16" replacement was. If it was suppose to be May June and the
1. pandemic slowed case design.
2. the microLED panel schedule ran off the rails.
3. Apple scarficed larger M1x dies for smaller die A14 and M1 dies on a limited access number of wafers
Then M1X and MBP 16" are 4-6 months late.
The whole premise that "oh but it would look back after an iPhone A15" intro is fundamentally flawed because the original schedule would have been before. Due to circumstances out of Apple's control it is now after.
If Apple's original schedule was always mid Fall 2021 then is more credence to that "posturing" premise (overly concerned about comparing iPhones to Macs. IMHO that is kind of goofy. They are in totally difference performance spaces and workloads. ). Also contributing could have been risk mitigation by Apple just keeping things simpler with just 1 chip to do over the first approximately 12 months and just largely dealing with shifting allocation problems between the wide set of production with variety demand curves ( iPad Pro , MBA , MBA 13" , Mini, and iMac 24" ... how many folks want to buy each of those at what times was likely a minor joggling act. ).
I'll be shocked if we get a M2X in these machines. They were supposed to be out earlier this year. They have apparently been delayed and as such they were always going to be M1X kit I reckon.
They should be M2X though at this point, but how often does Apple change it plans like this? They pretty much don't when it comes to hardware.
Plus there have been zero rumors around M2X in these machines. Doesn't mean it won't happen, but it's highly unlikely.
Love for you to be correct, but I think the central theme of this thread of temper expectations should be applied in this situation.
But apple does change their hardware plans, there is precedent, see my “PowerBook g5” quip. I find it hard to believe apple was happy to let the PowerBook languish with the G4, Motorola simply couldn’t provide the chips. Same with the layer MacBook pros. There’s no reason a 2014 MacBook Pro should have been able to keep up with a 2017, but here we are. That’s because intel couldn’t keep up with their promises. Now apple controls the design of the chips, so people thought that would solve the stagnation problem. Enter COVID-19 and TSMC capacity and mini-LED production problems.
PowerBook G5? Ancient history to be studied only!
Not sure what a layer MBP is.
But anything is possible!
It was a typo, it was supposed to say “later”. History tends to repeat itself.
Fair enough, but I don't think you really believe there is going to be M2X in these machines.
No, as I stated above. I assume that while I’d think if everything had gone well, we’d be halfway through a mbp run of m1x by now, I suspect that this fall will see the delayed release of the m1x machine. I hope.
When Quanta was hacked back on 4/21 of this year, and the J314 and J316 logic board schematics were leaked (reported by Macrumors) - we found out the 2021 MBP was in Engineering Validation Test prototype stage at that moment. This meant a summer launch was a near impossibility based on engineering timelines. If the MBP was ever supposed to launch over the summer, Apple internally knew this deadline (if it was ever an actual deadline) wouldn't be hit a long time ago. Probably back in 2020.
Speaking of the logic board schematics - they told us what the I/O looked like, but did any of the 21 leaked images show more of the board that would indicate processor changes? None of the articles I can find show all the leaked images, but rather showcase just a few of I/O and keyboard/case dimensions. Perhaps examining these timelines can help us determine if the new processor is based off A14 or A15. I'm leaning toward A15, but could be wrong!
Speaking of the logic board schematics - they told us what the I/O looked like, but did any of the 21 leaked images show more of the board that would indicate processor changes? None of the articles I can find show all the leaked images, but rather showcase just a few of I/O and keyboard/case dimensions. Perhaps examining these timelines can help us determine if the new processor is based off A14 or A15. I'm leaning toward A15, but could be wrong!
And yet AMD do that "year old" thing and their solution will probably sell more than quite well.
October-November 2020 Ryzen 5000 series
October-November 2021 Threadripper 5000 series.
It really isn't about old and microarchectures turning into pumpkins when the age clock strikes 12 months. If the wineskins are of vastly different sizes then then probably aren't going t be on exactly same iteration schedule. What matters more is that there is a regular pattern. Threadripper is somewhat on a 12-14 month iteration schedule. there was a previous update late 2019 - early 2020. (and the panedemic has somewhat thrown a curveball at this years update. )
In the primary target market versus Intel's Xeon W-3300 (Ice Lake) it is probably going to be OK. Versus a M-series that caps out at 10 cores ... probably will do just fine in that target market as well ( since has up to 64 cores ).
Where it is being used and for what targeted workloads matter.
Once the M-series line becomes much broader than the A-series in terma of which models on are being iternated onto the next generation, the M-series has more to update. That will likley update over a broader window than something where there is just one version to pop out every 12 months that gets used across the whole "new" iPhone line up.
already leaked that the MBP 16" has more ports. three TB ports (versus 2 ). Probably necessary to be able to put video out all three ports. Even more complete to be able to walk and chew gum at the same time and also put video out the HDMI port without it collapsing from the adjacent TB port. ( might even be able to qualify for TBv4 certification. )
Drive two 6K displays ( which the MBA can't and likely still won't with the 'plain' M2 ) . Unless want to backslide from the Intel MBP 16" , that's the minimal bar the M-series power one needs to hit.
The intel version has a AMD 5700M with 8GB of HBM2 memory. Is the plain M2 going to beat that? Probably not.
If I would have to guess…
It will be an M1x with 10 cores and 10% faster single core speed, up to 32 GB and up to 32 GPU cores.
I also expect the GPU to be equal or better than a RTX 3070, the 32 core version. I don’t expect hard ware raytracing however.
It will be an M1x with 10 cores and 10% faster single core speed, up to 32 GB and up to 32 GPU cores.
I also expect the GPU to be equal or better than a RTX 3070, the 32 core version. I don’t expect hard ware raytracing however.
You are focusing too much on these mnemonics. Let's not forget that these are not official names, just something that the internet has been using based on some old iPad chips naming. We don't know whether "M1X" is a thing. Technically, M1 is a "A14X", that's it.
The new prosumer chip could be called M1X, or it could be called M1 Pro or it could be called P1 or it could be called MP1 or something completely different. None of this means that it has to be based on exactly the same tech as M1. They could go for "M1 Pro" but base it on the A15 technology or in fact on a completely different, desktop-specific technology branch. It's just names. Don't interpret too much into them.
Oh I agree. I'm just saying that if they decide they don't want to use 'X' branding any longer, that's how they'd differentiate them ¯\_(ツ)_/¯
You know exactly what M1X means. It's the beefed up M1. It's pretty simple. The community here is running with this name until announcement otherwise it'd be chaos.
Sure it could be called something else. But that's not what I was talking about. I even floated the name P1 before WWDC on this forum.
More power, more heat, more fan noise. I love the M1 silent, yet powerful performance, have the Mac mini M1 which keeps cool and silent. But with the MacBook Air M1, it can get really hot if pushed hard, and something which worries me about M1X in the upcoming Pro’s. Hope apple has better cooling in these machines to keep perform without throttle and noise under pressure 😅
The hypothetical June 2021 M1X launch wasn't necessarily abandoned at the 11th hour. If the M1X Macs were due to launch in Spring/Summer 2021 the roadmap would probably have been drawn up sometime in 2019, if not before. The pandemic delays would have started to bite in spring/summer 2020. Any delay could easily have crept in long before the date of the leaked schematics.
I'd speculate that the higher-end 24" iMac was supposed to be M1X - which would explain the otherwise bizarre decision to have two slightly different models, one with better cooling and more ports. Just a guess, though. Even without that, though, launching the 24" iMac before the flagship 16" MacBook Pro, which you'd expect to be #2 priority after the Air, was a bit of a surprise.
From what I've seen, all we know is that it has three USB type C sockets (of unknown capability) and a HDMI socket (of unknown version) and some sort of Magsafe connector.
Best case:
3 x independent TB4 ports (supporting 2 displays) + 1 x Ethernet-over MagSafe + independent HDMI 2.1
Worst case:
2 x TB3/USB4 + 1 x USB-C 3.1 + HDMI 1.4 or 2.0 (good enough for Keynote presentations)
Best case needs a substantially expanded M1X with more I/O and display support... the "worst case" scenario could probably be implemented with a warmed-over M1 by including an on-board TB hub to share the single external display output with the HDMI. That would suck and be a huge disappointment, but the leak doesn't rule it out.
That, of course, assumes that the leaked schematics were/are ever on course to become an actual shipping product.
Except that would mean the M2X appearing before the M2 - which currently isn't expected until the new Air shows up - rumoured to be next spring.
Fair point, but why would it matter if the M2X comes out before the M2? It's just the high end version in significantly more expensive computers. If it's about consumer expectation, then yeah, that could be an issue. But the way around that is to rebadge them as M and P or something like that.
You're stating this as fact when no-one knows what the outcome will be.
All signs point to an iteration of M1, for the simple reason of how M1 was designed. It's setup perfectly for scaling linearly, meaning Apple could double the existing core counts and increase the die size by only a small percentage. This has the bonus effect of not requiring an entirely new architecture just to prioritise performance.
Remember, as has been the case in the past, Apple's A-series chips have always been numbered by generation. Thus, the M2 will be designed as the general purpose SoC just as the M1 is, but with minor improvements to energy efficiency and performance. It would make no business sense to use this architecture for their next pro machines when it's effectively designed for the entry level machines (that will, in 2022, gain this chip).
Furthermore, a chiplet-designed Mac Pro would benefit far more from an M1X since the chiplets would offer more performance.
(a) simple progression - first design the cores & make any tweaks to the fabrication process, then make the A15 and get that reliably into production, then soup that up a bit to make the M2, then use more of the same cores to make the "pro" version (which is going to be harder to manufacture if only because it has a larger area).
plus
(b) Apple's priority - the iPhone is Apple's #1 moneyspinner, so the A15 will always be first in line for new tech. The M2 is likely going in the Air and the iPad Pro which are probably Apple's next biggest selling items after the iPhone. The 14/16" MBP and 5k iMac will always be further down the list (...and might not always get annual updates anyway). Also, the top priority for the A15/M2 is likely to be power consumption - M2X can afford to suck a bit more juice.
Sure, it would be great if Apple had one big heave and refreshed their entire Mac range, in one go, with a coherent line based on the same core technology - but it is a long, long time since they've done that and it probably wasn't just Intel preventing it.
Was there a date on those schematics? I have looked at the two or three that were published by media outlets and I don't see any dates on them. So reEVIL might have leaked older documents to prove they had them, but kept the most recent documents in reserve as an incentive for Apple to settle.