M4+ Chip Generation - Speculation Megathread [MERGED]

[tenthousandthings]

If they still use monolithic dies and reuse the one generation older Max design (as they did with M3 Ultra) they could redesign it for the A16 node while M7, M7 Pro, and M7 Max would get N2P.
TSMC is remarkably reluctant when it comes to A16 and mass production...

TSMC: "The logic layout migration from N2P to A16 is actually quite straightforward because the cell structure and most of the layout patterns are quite the same. So, besides keeping the same front side structure, the beauty of A16 is that it inherits the NanoFlex feature from N2 ..."

The progression from N2P to A16 is built into the 2nm architecture, by design. So Apple could use N2P for M7 and probably M7 Pro/Max, but for the server-grade silicon project Apple could use A16 and back side power delivery ("Super Power Rail"), starting from the M7-family designs. There's zero chance they would use an older design. M3 Ultra didn't change process nodes -- it is built using first-generation N3, just like the rest of the M3 family.

The unknowns are M5 Ultra and M6 Ultra. I get what you're thinking, that maybe M6 Ultra will lag behind the rest of M6 and there would be no M7 Ultra, but that doesn't fit with TSMC A16.

TSMC A16 is aimed at AI and HPC -- you're not going to see it in consumer products, but *if* Apple is building its own server/workstation silicon, it would be the time and place for a major stride forward.

 

[Antony Newman]

If TSMC lowers the (estimate) $45k/wafer price of A16 - it could kill the chance that the 2 x potential interested parties sign onto Intels competing 14A process (link) - leaving Intel stranded in 18A with a transition (I presume) to move state of the art foundry work to TSMC (and perhaps Rapidus).

I don’t know if modern HPC implicitly means they can handle AI workloads efficiently - or if there is a growing proportion of consumer products that are expected to have natural language processing. It is probably just a question of when commodity products are on A16.

If Intel believe that 18A will suffice until at least 2030 - perhaps A16 chiplets will make an appearance in consumer products not long after.

 

[smalm]

There's zero chance they would use an older design. M3 Ultra didn't change process nodes -- it is built using first-generation N3, just like the rest of the M3 family.

And I once thought there is zero chance Apple will come out with a M3 Ultra 16 months after introducing the M3 Max and 4 months after introducing M4 Max and retiring the M3 Max.

 

[tenthousandthings]

And I once thought there is zero chance Apple will come out with a M3 Ultra 16 months after introducing the M3 Max and 4 months after introducing M4 Max and retiring the M3 Max.

No doubt, but it doesn’t change the odds that M6 Ultra, if it were to launch in the Mac Studio alongside M7 Max, would still be on TSMC N2, not N2P/A16. Just like M3 Ultra is on TSMC N3, not N3E.

The only way M6 Ultra might be on TSMC A16 would be if Apple skips over N2 and the M6 Pro/Max/Ultra launches in 2027 on N2P/A16. There’s a slim chance of that — it looks like M5 Pro/Max/Ultra may not launch until early 2026 (on N3P).

There is some recent precedent — Apple A16 was on TSMC N4P and the M-series skipped over it — something like that could happen for Apple A20 on TSMC N2.

 

[tenthousandthings]

Coming this Fall:

J700 = Mac17,1 :: A18 Pro MacBook 12" WiFi
J701 = Mac17,2 :: A18 Pro MacBook 12" WiFi + Cellular

It will launch along with the iPads:

J510 = iPad17,1 :: A18 Pro iPad mini WiFi
J511 = iPad17,2 :: A18 Pro iPad mini WiFi + Cellular

J817 = iPad17,3 :: M5 iPad Pro 11" WiFi
J818 = iPad17,4 :: M5 iPad Pro 11" WiFi + Cellular
J820 = iPad17,5 :: M5 iPad Pro 13" WiFi
J821 = iPad17,6 :: M5 iPad Pro 13" WiFi + Cellular

Coming in H1 2026:

J704 = Mac17,3 :: M5 MacBook Pro 14"

J723 = Mac17,4 :: M5 iMac 24" (Two ports)
J724 = Mac17,5 :: M5 iMac 24" (Four ports)

J714s = Mac17,6 :: M5 Pro MacBook Pro 14"
J716s = Mac17,7 :: M5 Pro MacBook Pro 16"
J714c = Mac17,8 :: M5 Max MacBook Pro 14"
J716c = Mac17,9 :: M5 Max MacBook Pro 16"

J873g = Mac17,10 :: M5 Mac mini
J873s = Mac17,11 :: M5 Pro Mac mini

J775c = Mac17,12 :: M5 Max Mac Studio
J775d = Mac17,13 :: M5 Ultra Mac Studio

J813 = Mac17,14 :: M5 MacBook Air 13"
J815 = Mac17,15 :: M5 MacBook Air 15"

J480d = Mac17,16 :: M5 Ultra Mac Pro
J480e = Mac17,17 :: M5 Ultra+ Mac Pro

 

[MayaUser]

A basic Macbook Air to have Cellular option and not an Pro Macbook..that would be strange
Hopefully they all get cellular options

 

[crazy dave]

Coming this Fall:

J700 = Mac17,1 :: A18 Pro MacBook 12" WiFi
J701 = Mac17,2 :: A18 Pro MacBook 12" WiFi + Cellular

It will launch along with the iPads:

J510 = iPad17,1 :: A18 Pro iPad mini WiFi
J511 = iPad17,2 :: A18 Pro iPad mini WiFi + Cellular

J817 = iPad17,3 :: M5 iPad Pro 11" WiFi
J818 = iPad17,4 :: M5 iPad Pro 11" WiFi + Cellular
J820 = iPad17,5 :: M5 iPad Pro 13" WiFi
J821 = iPad17,6 :: M5 iPad Pro 13" WiFi + Cellular

Coming in H1 2026:

J704 = Mac17,3 :: M5 MacBook Pro 14"

J723 = Mac17,4 :: M5 iMac 24" (Two ports)
J724 = Mac17,5 :: M5 iMac 24" (Four ports)

J714s = Mac17,6 :: M5 Pro MacBook Pro 14"
J716s = Mac17,7 :: M5 Pro MacBook Pro 16"
J714c = Mac17,8 :: M5 Max MacBook Pro 14"
J716c = Mac17,9 :: M5 Max MacBook Pro 16"

J873g = Mac17,10 :: M5 Mac mini
J873s = Mac17,11 :: M5 Pro Mac mini

J775c = Mac17,12 :: M5 Max Mac Studio
J775d = Mac17,13 :: M5 Ultra Mac Studio

J813 = Mac17,14 :: M5 MacBook Air 13"
J815 = Mac17,15 :: M5 MacBook Air 15"

J480d = Mac17,16 :: M5 Ultra Mac Pro
J480e = Mac17,17 :: M5 Ultra+ Mac Pro

M5 Ultra+?

 

[tenthousandthings]

M5 Ultra+?

My fantasy of SoIC allowing for a GPU-heavy variant, using whatever the Apple PCC AI server silicon is called. Too big for the Studio chassis, Mac Pro only. Maybe combining one Max with one of the AI units?

In my dream, it will launch alongside a new Pro Display XDR.

 

[crazy dave]

My fantasy of SoIC allowing for a GPU-heavy variant, using whatever the Apple PCC AI server silicon is called. Too big for the Studio chassis, Mac Pro only. Maybe combining one Max with one of the AI units?

In my dream, it will launch alongside a new Pro Display XDR.

Got my hopes up that I missed a new rumor!

 

[dgdosen]

If an A18Pro is on par with an M1 (with faster single core perf) - what would you speculate the first A16/14 apple silicon phone chips will provide? Assuming they'll have GAA transistors and use the SPR, two huge advances, what will that version's iPhone chip be equivalent of? An M1Max (with faster single core perf)?

 

[tenthousandthings]

If an A18Pro is on par with an M1 (with faster single core perf) - what would you speculate the first A16/14 apple silicon phone chips will provide? Assuming they'll have GAA transistors and use the SPR, two huge advances, what will that version's iPhone chip be equivalent of? An M1Max (with faster single core perf)?

TSMC A14 won’t have SPR. The second generation of SPR will be related to A14P. Not much else is known, because it depends on what they learn from N2P/A16. It’s not likely that iPhone Pro would get it before A14P in 2029. Even then, perhaps not.

Apple A20/M6 will use N2, so GAA transistors will debut next year.

I don’t know the answer to your question, but my guess would be the equivalence would be closer to M2 than to any Pro/Max silicon. I could be wrong.

 

[Icelus]

Dougall (@dougall@mastodon.social)

Apple Silicon CPU Optimization Guide version 4.0 has been released, with fantastic details about the SME Engine microarchitecture (including some fancy operation fusion pairs), and new information for the A17 Pro, A18 and M4 CPUs...

mastodon.social