Rhodes is a Greek island — the US island was Rhode, from the Dutch “Roode” meaning red (from its red clay), later anglicized as “Rhode.”
[Sorry, I couldn’t resist. I looked it up and then tried to stop myself from posting it, to no avail.]
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.
M4+ Chip Generation - Speculation Megathread [MERGED]
- Thread starter GoetzPhil
- Start date
- Sort by reaction score
[Sorry, I couldn’t resist. I looked it up and then tried to stop myself from posting it, to no avail.]
Apple announces M3 Ultra—and says not every generation will see an “Ultra” chip
It could explain why we’re getting an M3 Ultra this deep into the M4 rollout.
arstechnica.com
M4 Ultra might not see the lights of the day. Apple has chosen M3 Ultra with 512GB LPDDR5 due to high density of LPDDR5 (819 GB/s) is available. M4 Max is using LPDDR5x: higher bandwidth (546 GB/s) but lower capacity (128GB Max).
Based on article above, Apple most likely jump to M5 Ultra with 512GB LPDDR6.
Last edited:
As always, you have very strange ideas about these topics.Apple announces M3 Ultra—and says not every generation will see an “Ultra” chip
It could explain why we’re getting an M3 Ultra this deep into the M4 rollout.
M4 Ultra might not see the lights of the day. Apple has chosen M3 Ultra with 512GB LPDDR5 due to high density of LPDDR5 (819 GB/s) is available. M4 Max is using LPDDR5x: higher bandwidth (546 GB/s) but lower capacity (128GB Max).
Based on article above, Apple most likely jump to M5 Ultra with 512GB LPDDR6.
"Ultra" has a built-in 2x capacity potential advantage over "Max" of the same generation simply because an "Ultra" is two "Max" chips glued together, and therefore has twice as many memory channels.
But I was careful to say potential because there's no rule that forces Apple to choose to use all of that potential in every case. The fact that there's no 256GB M4 Max can just mean that Apple decided 128GB was the biggest memory config they'd build in a Max. It doesn't hint that capacity is linked to memory channel speed.
Well, it is called analytics mind with technical example:
- Clearly Apple has taken priority of memory size than speed (512GB LPDDR5 over LPDDR5x). Why do you think Apple rather respined M3 Max die to enable Ultra Fusion support than using M4 Max? Does it look weird that Mac Studio lineup to have M4 Max in lower tier and M3 Ultra in higher tier? And yet, Apple goes for M3 Ultra..And since memory size is the top priority due to AI rush, why don't Apple enable 256GB in M4 Max if Apple could?
- If you are familiar with memory technology, capacity and speed are not always growing linearly especially in server platform. Let me show you the best example in the picture below:
As for 512GB LPDDR6: I managed to download up-to-date JEDEC LPDDR6 documents dated April'25 to confirm my theory about 96-bit LPDDR6 memory interface. I am just waiting for JEDEC to officially announce the standard.
I was not able to find any technical documentation stating that LPDDR5 has higher density compared to LPDDR5x. RAM manufacturers like Samsuns seems to specify the same density for both (which makes sense given that the differences between specs are minimal).
As to why Apple chose to skip M4 Ultra, I suspect has much more to do with business operations and strategy than RAM technology. It's possible that they had a stockpile of M3 Max chips they wanted to use, or maybe it takes a while to accumulate enough Max dies to make enough Ultras, and using M4 for that purpose was not financially viable. Either way, I would not read too much into it.
I do not follow. The example you bring clearly shows linear improvement in bth size and speed (2x capacity, 2x speed).
P.S. I would be very surprised to see LPDDR6 any time soon on Apple devices. Historically Apple lags behind others in adopting RAM tech — not because they are conservative, but because they need these chips in volumes that absolutely draw other clients. It's one thing to procure cutting edge RAM chips to ship in your flagship Android phone of which you will only sell a couple of thousands units, and a very different one to get enough chips for dozens of billions devices.
Do you remember how wrong you were when you tried to make big claims about the future of Apple Silicon based on your misunderstandings of memory technology? You can't claim an "analytics mind" for a topic you don't understand well enough to analyze.
Can you tell me when memory speed decreases with increased capacity, when it does not, and why? This is really the key issue. I think you just assumed there must be a tradeoff between speed and capacity here, but based on technical understanding of memory interfaces and observing how Apple uses them, I don't think that's a good assumption.
Can you tell me why, if the "AI" push for more memory is Apple's top priority, Apple failed to put a 256GB M3 Max Studio on sale a year ago? After all, M3 Max uses LPDDR5 at the same clock speed. They had the chip, why no product launch?
I think the real explanation is simple: schedule slip. M1 Ultra and M2 Ultra both launched about 0.5yr after the Max chips of their generations, but for M3 Ultra, that delay increased to 1.5yr. +1yr is a lot of time! Hints at a problem. It's plausible that Apple's attempt to get back on track was to cancel M4 Ultra, freeing resources to assure M5 Ultra (and other future chips) launch on time.
Alternatively, perhaps they never planned on a M4 Ultra at all. Ultra chips require extra effort to validate, and when you combine that with low sales volume, they probably aren't Apple's top priority. One of their execs admitted this (more or less) at the M3 Ultra launch when he mentioned not every M chip generation will have an Ultra.
Finally, as to why no 256GB M4 Max - that's simple. The downside of not using modularized RAM is that if Apple offers every physically possible config to customers, the number of different SoC packages and motherboards Apple has to manufacture and distribute in its internal supply chain explodes. Combine that with low sales volume of these products and it's a recipe for Apple having a pretty big incentive to keep the number of extremely high end Mac configs down to a minimum. Since there will be some overlap between "people who want a Mac with a lot of CPU cores" and "people who want a Mac with a lot of memory", they have chosen configs that are an intersection between those requirements. Some will have to pay extra for CPU and GPU cores they don't want, but customers tend to be less price sensitive in the high end workstation market.
The M3 Max version used in M3 Ultra Mac Studio is a newer version of the M3 Max used in prior Macs. Those do not incorporate the UltraFusion interface and the new version has TB5 instead of TB4.
The M4 Max used right now is lacking UltraFusion too.
~
There’s a lot we don’t know about M3 Max — the Apple graphics from 2023 are not definitive. There could even be another chapter in the saga, possibly tomorrow but more likely later on.
There's nothing PREVENTING Apple from using more DRAM chips if that's required when using LPDDR5x...
The simplest way to do this is to mount the DRAM chips on the reverse side of the packaging and treat the two sides as two ranks. There's nothing especially novel in this, it's basically the same idea as a standard DIMM.
I would be not be surprised if Apple skip the M4 Ultra, but I don't think it's especially for the reason you give.
Note that he also updated the first two posts in his old (April 2024) M4 = LPDDR6 thread, so it is now M5 = LPDDR6. Few (if any) of the basic lessons that emerged as his original claims (preserved in responses on the second page) collided with reality as M4 and M4 Pro/Max launched last year have been learned.
So, again, while A20/M6 could use it next year, there is little chance A19/M5 will use it this year. It’s true that Samsung is rumored to be using it in the Galaxy S26 Ultra, shipping early next year. For Apple to ship it this year, it would have to be in volume production right now, and that just isn’t the case.
Another of his unlikely claims also persists, his apparent belief that TSMC N3 (N3B) will be used for M4 Pro/Max and M5 Pro/Max, in addition to M3 Pro/Max. We know that M4 Pro/Max is N3E, so, again, a collision with reality doesn’t seem to have had any effect.
FWIW, I think there may be a language issue here, so carefully measured, detailed corrective responses like yours in May 2024 often fail to make an impression. I think he means well, but it feels like things are getting “lost in translation.”
So no mention of M5 yet. We're over one year since the M4 was released.
Might that lack of cadence point to the M5 being destined for TSMCs N2 node and some early version of GAA? Or might Apple be slowing any cadence to focus efforts on chips in sever farms?
Yeah Rosetta 2 is almost certainly here for the long haul. Another key difference is that Apple had to pay licensing for Rosetta 1, while Rosetta 2 is in-house. Forever is indeed a long time, but there is little reason for Apple to get rid of it and for the foreseeable future Rosetta 2 is a pretty integral part of their strategy for both virtualization and especially game development. In short: as things stand now, the only circumstance where I could see Apple dropping Rosetta 2 entirely is because they switch ISAs again and need Rosetta 3(and even then Rosetta 2 would stick around until ARM-Macs lost support entirely).
About the Rosetta translation environment | Apple Developer Documentation
Learn how Rosetta translates executables, and understand what Rosetta can’t translate.
Apple have said they'll stop supporting building new Mac apps for Intel machines after the next two OSes, but Rosetta 2 will stay around for unmaintained Intel-macOS titles (and presumably GPTK-Wine purposes). While this statement is a bit vague, it seems Rosetta 2 and Intel frameworks will stay around for awhile, but Apple will not support building macOS-Intel apps for those future macOSes.
Neither. M5 will probably be released in the fall on the latest N3 node with the Macs a year after M4 Macs - we won't be seeing N2 products until next year. The M4 iPad being released only 7 months after the M3 Macs was likely an aberration. A yearly average cadence with some wiggle room given specific product development needs will likely be the norm (which is indeed what the rough average cadence since the M2 will be if the M5 launches in the fall).
Last edited:
So with no new Mac Pro announcement from WWDC, I’m not sure what to think of the Mac17,1 + Mac17,2 any longer.
The only new Mac announcement that we can be almost 100% sure of each year is new MacBook Pro’s in late October/November and those two IDs doesn’t cover enough SKUs. So maybe Mac17,1 + Mac17,2 is for some lower tier M5 MacBooks for back to school in August?
The only new Mac announcement that we can be almost 100% sure of each year is new MacBook Pro’s in late October/November and those two IDs doesn’t cover enough SKUs. So maybe Mac17,1 + Mac17,2 is for some lower tier M5 MacBooks for back to school in August?
That's certainly a possibility - it's also possible that the Mac Pro or another (new) M5 device is still coming before the next OS release. It's also possible that Apple's timetables have slipped and Macs 17,1 and 2 will be coming with the rest of the Macs after the new OS release ... or well if I remember correctly there were a couple of M2s (Mac 14s) listed in the original data dump that never made an appearance. Maybe these M5s won't either. Obviously we'll find out one or the other eventually, but for now? 'tis a mystery.
Maybe M5 is on N3P for most machines, but the Mac Pro (and the all-new Mac Pro Cube) are on N3X, allowing higher performance chips for Apple's personal workstations...?
That's certainly a possibility - it's also possible that the Mac Pro or another (new) M5 device is still coming before the next OS release. It's also possible that Apple's timetables have slipped and Macs 17,1 and 2 will be coming with the rest of the Macs after the new OS release ... or well if I remember correctly there were a couple of M2s (Mac 14s) listed in the original data dump that never made an appearance. Maybe these M5s won't either. Obviously we'll find out one or the other eventually, but for now? 'tis a mystery.
Maybe the all-new Mac Pro Cube...?!? ;^p
Maybe if there is an “extreme” or at least a special edition the M5 Ultra for the Mac Pro, it may be built on that N3X node. Or maybe they use the N3X for the M5 Ultra in general?
I think that’s the most likely possibility, if Apple decides to use that node. The other most likely possibility is Apple just using the N3P for all the M5 family and jumping straight into the N2 process node.
I don’t think Apple is going to use N3X for lower power chips, not even the M5 Max or M5 Pro, because from what I’ve read, it draws quite a lot power, and the thermal management could be a problem, especially with the beefier GPU we’re expecting for the M5 generation, which is the main heat generator.
If I had to bet, I’d bet Apple using just the N3P process for the M5 family and not using the N3X at all, for two reasons: heat and economy of scale.
And the M6 will be N2 already, if everything goes according to plan.
PS: Wait, an all new Mac Pro Cube??? Is that a rumor? Anyway, if they go that way, more reason to stick with the N3P process node, as an M5 Ultra built on the N3X process could risk having overheating problems, if it draws as much power as I’ve read.
With Apple being all for efficiency (I guess they learned the lesson back in the PowerPC era), honestly I don’t see them adopting the N3X process node.
No lol, that’s just an idea Boil is obsessed with.
My guess:
N3 -> N3E -> N3P -> N2 -> Second gen A14
M5 = N3P : Reduction in power (wattage) : 2.5D Packaging
M6 = N2 : Complete chip redesign around GAA : ‘3D’ : higher clock speed : back to front bonding
M7 = A14 (second gen) : drastically lower power with class leading backside power : face to face bonding
N3 -> N3E -> N3P -> N2 -> Second gen A14
M5 = N3P : Reduction in power (wattage) : 2.5D Packaging
M6 = N2 : Complete chip redesign around GAA : ‘3D’ : higher clock speed : back to front bonding
M7 = A14 (second gen) : drastically lower power with class leading backside power : face to face bonding
Keeping in mind that TSMC has been producing N3P chips reliably since this past winter, and that the most likely first customer to be supplied is Apple… if they finally release the M5 family in the fall -like I, too, expect-, that would mean they’ve been producing M5 family chips for at least 9 months at the moment of launch. This could probably result in a high supply of M5 family chips, have very good yields, and they will be able to simultaneously refresh most of the mac lineup or even the iPad Pro. In time for Tahoe launch.
Also, if the M5 is also intended for the next Vision Pro as it’s been rumoured, they will need exec more chips.
If all of this ends up being true, the M5 could be something like the A15 or the M2: a highly refined chip meant to be used in most of the devices as a “new baseline”, although the A18 is almost that.
For instance, I know the Vision Pro has an M2. Several iPads also have the M2. The M2 is based on the A15 architecture and cores, and the A15 is a widely used SoC across many iPhones, iPads and even AppleTV.
This kinda grants me that this family of chips (A15/M2) will be supported by Apple and many developers for many years because there is likely a high base of devices running this chip architecture. Also, remember, the M2 chips have been the only SoC mounted on the Mac mini and Mac Studio up until the release of the M4 Macs…
This is pure speculation, mind your.
If the M5 is launched simultaneously on most devices, that would be this is the last, most refined and tested, and powerful SoC of this “era”.
The next family, the M6… that will be a whole new thing, I feel. Much better, of course, but with new transistor technology, new nodes, new architecture… it will have so many new things that, for many years, the M2/M3/M4/M5 architecture will be the most widely supported, known, and used. And among those, the M5 will be the most powerful chip, implanted in most of the next product updates.
Then, the M6 will probably adopt a new architecture in order to take advantage of 1) new transistor technology, 2) new substrate technology, 3) new demands for the upcoming computing needs, such as a bigger implementation of LLMs across the operating system, hopefully in local.
But that’s still too far away.
I concur… I mean I mostly agree, but for the most immediate release, the M5, I’m not so sure it will be focused in power efficiency, or reduced wattage.
From what I’ve read the N3P offers small improvements, both in efficiency andand performance, compared to the current N3E node. If that’s the case, I see Apple pushing the clock speed a bit further (just like they’ve done with previous generations), to gain a more significant performance improvement. That, along with what I expect to be a significantly improved GPU, with at least 2 more GPU cores, offering the base M5 with at least 12 GPU cores, would mean a hotter SoC.
That would be at least the case of the desktop models. Maybe they’re preparing a binned version with just 10 or 11 cores and fewer CPU cores for the MacBook Air which, with the M4, it already runs hotter than the previous models.
We’ll see, soon, what’s the strategy with N3P and M5 chips, but I think they are going to opt in favour of performance gains instead of improving the efficiency.
Another hint at this, is the fact that the A19, the full, non binned version of the A19 aimed at the iPhone Pro lineup, is going to come equipped with a vapor chamber.
Could be M5 slippage, like you say, but we’re still on track for Gurman’s summer 2025 Mac Pro with silicon codenamed Hidra. He has insisted repeatedly, as recently as October 2024, that Hidra is M4.That's certainly a possibility - it's also possible that the Mac Pro or another (new) M5 device is still coming before the next OS release. It's also possible that Apple's timetables have slipped and Macs 17,1 and 2 will be coming with the rest of the Macs after the new OS release ... or well if I remember correctly there were a couple of M2s (Mac 14s) listed in the original data dump that never made an appearance. Maybe these M5s won't either. Obviously we'll find out one or the other eventually, but for now? 'tis a mystery.
So the first M5 Macs (Mac17,1 and Mac17,2) could have slipped, while M4 Hidra could still have been cancelled — the Apple comment about not every M generation will have Ultra may be less about cadence and more about Hidra being cancelled.
Here’s the thing, though. There’s still something else big going on. There is tremendous movement in the display industry right now. 5K will finally gain a foothold in gaming, as “dual mode” displays prove their worth. Both Samsung and LG have tandem OLED technology capable of both high resolution and high refresh rates at the same time. It’s coming. Samsung appears to be slightly ahead, they already have a working prototype of a QD-OLED Retina 5K 120Hz panel.
I will be very surprised if Apple does nothing this year as the industry finally catches up. There’s at least three new Retina 6K displays coming (ASUS ProArt, Acer ProCreator, both of which also have Retina 5K models), including the LG Thunderbolt 5 display. Apple should respond.
So even if M4 Hidra wasn’t built because it didn’t perform the way the Mac Pro needed it to (or because Gurman was wrong and it was M3 Ultra all along), the Thunderbolt 5 Pro Display to go alongside it must still be on the way.
So here’s what I think, all due this summer:
- Mac17,1 M5 iMac 32" Retina 6K
- Mac17,2 M5 Pro iMac 32" Retina 6K
- Studio Display 32" Retina 6K
- M3 Ultra [?] Mac Pro
- 34" Liquid Retina 8K Pro Display XDR Max (Just kidding, I have no idea what they will do, there’s an old “7K” rumor, but I would buy the above in a heartbeat.)
It's also possible Gurman is simply wrong about Hidra being M4, according to the leaked data sheet there are no new M4s due before the next OS. Of course it is also possible that an M4 Hidra will launch with the next OS not this one (and thus wouldn't have been in the leaked data), after all the M3 Ultra launched after the M4s. But unless Apple's statement that "not every generation will get an Ultra" is misdirection (very possible - i.e. it's possible M4 will have an Ultra but M5 won't and thus the statement is true but still a misdirect, who knows?), it seems like there is no M4 Ultra tier. Or maybe Hidra, whatever it is, is solely for PCC. Seems wasteful, but maybe.
Bigger iMacs with more powerful chips would be cool, but if they were launching this summer, shouldn't we be seeing display supply chain leaks already? I haven't been paying super close attention to this, so I may have missed any, but the last rumors I remember is that "Apple is still exploring it but don't expect anything soon here" and maybe some work on new Studio displays (which could provide cover for a new iMac?).
If the Mac Pro is just going to get the M3 Ultra, that doesn't seem like they needed to wait, but then Apple's decisions don't always make sense (to me, from the outside looking in through a glass, darkly).
Well, I’m almost always wrong, so who knows?
On the iMac and the supply chain, with regard to 6K display panels, I think there is a lot of activity right now. I mean, LG is producing their own Thunderbolt 5 display (which I will buy, I’m not waiting, sorry Apple), so the retooling for that new product could mask a large Apple order. I don’t know who is supplying the 6K panels ASUS and Acer are using, it could be LG.
Rhode Island in the USA is a State; not an island . You can drive/walk into/from various parts of Connecticut or Massachusetts to RI without crossing any significant body of water. There are islands inside the State’s boundaries. Originally those ( and the large bay) were key to the State’s economy.
There are two Rhodes Island in Canada; one on either side of Atlantic and Pacific coasts. But no where near as populated as Staten Island. So likely Greek one . 😀 ( but maybe someone had a nice , very remote vacation to one of those )
P.S. Ellis was the A15 . The relation is coupled to the CPU microarthecture implementation baselines more so than direct physical location ( although sometimes looking at same map location to pick names .
Ellis and Statrn are islands around New York City area. Something like Govenours , Roosevelt , Manhattan, Liberty , etc. ( probably not Rikers island though …. Although Apple did at one point use Alcatraz. Alcatraz is more a Pier 39 side trip these days so more an outing than a prison. )
Process nodes N2 and A14 are not likely design rule compatible . Pretty likely there will be a N2-suffix ( b , e , p , etc something besides ‘X’ ) . Very likely Apple iterates on another cycle with more optimized design tools they already worked with before they drift off to the lunatic fringe again.
There is a A16 before A14 . A16 may be closer to N2 rules ( derived from N2P ) there may be a non- hyper optimized path from N2P to A16 that is cheaper to do .
It is pretty likely some other chip designers get to A16 before Apple does. ( someone skipping ahead for relatively expensive product. )