M4+ Chip Generation - Speculation Megathread [MERGED]

[tenthousandthings]

How reliable is this information? Like does this person Ilikeiphone123 have a track record? It seems he's saying he found the codenames in Playgrounds? Someone can confirm?

But let's go on a limb and say it's accurate (and not missing anything), it could be that Pro and Max chips are being pushed back this generation and there will be a gap next generation, but maybe a quicker update with the M7? So let's say Mar 2026 and then October 2027? Rather than releasing October 2025 and October 2027?

I have to admit I'm not sure I believe this - the rumor mill is that Apple will massively updating the M6 MBPs with all new chassis and screens, but if there are literally no Pro and Max chips then there is no 16" M6 MBP at all (and of course no Pro minis nor Studios that generation). So the only upgraded MBPs is the 14" with the base M6 chip. Hmmmm ... possible of course, but seems unlikely.

The only way it makes sense to me is if they use TSMC A16 and introduce SPR (Super Power Rail) — I could see a June 2027 announcement with Fall 2027 availability. I also think they could call it M6 Pro/Max/Ultra, not M7.

 

[deconstruct60]

How reliable is this information? Like does this person Ilikeiphone123 have a track record? It seems he's saying he found the codenames in Playgrounds? Someone can confirm?

But let's go on a limb and say it's accurate (and not missing anything), it could be that Pro and Max chips are being pushed back this generation and there will be a gap next generation,

Is this a spreadsheet of chips or of codenames for the chips? In the latter, case it is matter of what leaks sprung on the codenames. 'Hidra' leaked early was early on linked to magical top end solution coming out ( Hildra and 'ultra' appearing in conversations together. ). And yet is this is correct it is the plain M5.

The M7 entries all have question marks next to their name. More than a decent chance that the stuff with '?' is at least as much as "guess" as it is a documented and confirmed leak. Those entries also follow the same 'S' , 'C', 'D' sequence. If the '?" turns into M6 Pro/Max/Ultra then mystery solved.

 

[l0stl0rd]

View attachment 2568551

Still in the box (opening and installation awaits the completion of other priorities).

IIRC, my M2 Max w/64GB clocks ~400GB/s mem-band. Awaiting M4U/M5U before upping.

Fun times ahead!

Curious too but when I saw like 1.2 it/s on an sd 1.5 model in Comfy UI, I thought where like well guess I am saving that money.

Definitely not made for interference, training yes maybe.

 

[splifingate]

Nice! I wish I could justify getting one, not for AI but even just doing CUDA work. Let us know how it performs!

I will.

Currently have nothing with which to compare, so eyes-wide-shut.

 

[splifingate]

Curious too but when I saw like 1.2 it/s on an sd 1.5 model in Comfy UI, I thought where like well guess I am saving that money.

Definitely not made for interference, training yes maybe.

$ is just $

The prospect of creating unlimited imagery locally--free from API rate limits (and cloud costs)--coupled with the expectation of maintaining (I am hoping to assume) personal privacy is a ++ 😉

I've already dropped USD2K on a base AM5 EPYC system; it really didn't make a lot of immediate sense to drop another USD9K on a Tier-1 GPU. The cost-effective (albeit gimp'ed) Blackwells (USD2K-USD6K) are non-existent from a purchasing standpoint 🤷‍♂️

I'm anticipating finding a cohesive blend of inference vs training...and a greater understanding of what I can do when the tech improves.

 

[Tdude96]

$ is just $

The prospect of creating unlimited imagery locally--free from API rate limits (and cloud costs)--coupled with the expectation of maintaining (I am hoping to assume) personal privacy is a ++ 😉

Take a look at Draw Things if you haven't. It has a native app you can find in the App Store, and a discord with a lot of helpful people. Apple actually name-dropped the app in their M5 press release. That app has always been designed around local generation specifically for Apple Silicon, but they have now added a cloud-based choice as an option for those who want it.

They have a few things on technical details of what they do at engineering.drawthings.ai

 

[wmy5]

How reliable is this information? Like does this person Ilikeiphone123 have a track record? It seems he's saying he found the codenames in Playgrounds? Someone can confirm?

But let's go on a limb and say it's accurate (and not missing anything), it could be that Pro and Max chips are being pushed back this generation and there will be a gap next generation, but maybe a quicker update with the M7? So let's say Mar 2026 and then October 2027? Rather than releasing October 2025 and October 2027?

I have to admit I'm not sure I believe this - the rumor mill is that Apple will massively updating the M6 MBPs with all new chassis and screens, but if there are literally no Pro and Max chips then there is no 16" M6 MBP at all (and of course no Pro minis nor Studios that generation). So the only upgraded MBPs is the 14" with the base M6 chip. Hmmmm ... possible of course, but seems unlikely.

I believe this is to be pretty accurate. Codenames usually leaked in software betas. As for how we know which chip each codename corresponds to, it is relatively easy to guess by the pattern of CPID. For example, since we know T6030 is M3 Pro and T6040 is M4 Pro, now T6050 is Sotra, we can be fairly confident that Sotra is M5 Pro.

Regarding the redesigned MacBook Pro, note that M5 Pro/Max were not released together with M5, which was an unusual move. This could mean the rumored redesigned MacBook Pro will be equipped with M5 Pro/Max and be released in 2026. It will be similar to the situation in 2021: old design MacBook Pro 13 with M1 were selling together with MacBook Pro 14"/16" with M1 Pro/Max.

 

[wmy5]

Yeah this rumour is pretty weird. If it were true what I’d expect is that Apple skip the M5 Pro/Max and update with M6 where the redesign is expected. But the Gurman tier of the rumour mill has already said there’s an M5 Pro/Max early next year *and* M6 redesign updates late next year or early 2027.

Someone is definitely wrong.

Gurman said Hidra was a Ultra-tier chip but it turned out to be very wrong. Regarding silicon codenames Gurman's track record is not as good as his record on final products.

 

[wmy5]

Is this a spreadsheet of chips or of codenames for the chips? In the latter, case it is matter of what leaks sprung on the codenames. 'Hidra' leaked early was early on linked to magical top end solution coming out ( Hildra and 'ultra' appearing in conversations together. ). And yet is this is correct it is the plain M5.

The M7 entries all have question marks next to their name. More than a decent chance that the stuff with '?' is at least as much as "guess" as it is a documented and confirmed leak. Those entries also follow the same 'S' , 'C', 'D' sequence. If the '?" turns into M6 Pro/Max/Ultra then mystery solved.

The moment we know Hidra is M5 not an Ultra is when we have its CPID T8142, since T8132 is M4 and T8122 is M3. Similarly, Komodo being M6 and Delos being M7 is fairly certain now.

 

[crazy dave]

The only way it makes sense to me is if they use TSMC A16 and introduce SPR (Super Power Rail) — I could see a June 2027 announcement with Fall 2027 availability. I also think they could call it M6 Pro/Max/Ultra, not M7.

Is this a spreadsheet of chips or of codenames for the chips? In the latter, case it is matter of what leaks sprung on the codenames. 'Hidra' leaked early was early on linked to magical top end solution coming out ( Hildra and 'ultra' appearing in conversations together. ). And yet is this is correct it is the plain M5.

The M7 entries all have question marks next to their name. More than a decent chance that the stuff with '?' is at least as much as "guess" as it is a documented and confirmed leak. Those entries also follow the same 'S' , 'C', 'D' sequence. If the '?" turns into M6 Pro/Max/Ultra then mystery solved.

Yeah I dunno - the original link isn't specific as to what the original user was looking at but they seem to be able to relate code names to chip IDs, but looking at those chip IDs and @wmy5's list again, I think you are correct, those are almost certainly M6 Pro/Max/Ultra.

I believe this is to be pretty accurate. I cannot say why I feel confident.

At this stage M6 generation is likely to be testing stage. Nobody other than Apple knows when it will be released in final products. I also doubt whether Apple has firmed up its plan about M6 series Macs yet or not.

Regarding the redesigned MacBook Pro, note that M5 Pro/Max were not released together with M5, which was an unusual move. This could mean the rumored redesigned MacBook Pro will be equipped with M5 Pro/Max and be released in 2026. It will be similar to the situation in 2021: old design MacBook Pro 13 with M1 were selling together with MacBook Pro 14"/16" with M1 Pro/Max.

Looking at the pattern though I'm pretty sure those are M6 Pro/Max/Ultra. He may have placed them after M6 in the chart, but look at the numbers:

M3 Pro - Ultra: T6030s
M4 Pro - Max: T6040s
M5 Pro - Ultra: T6050s

So it would stand to reason that T6060s would be M6 Pro-Ultra would it not?

The moment we know Hidra is M5 not an Ultra is when we have its CPID T8142, since T8132 is M4 and T8122 is M3. Similarly, Komodo being M6 and Delos being M7 is fairly certain now.

Aye but similar logic should dictate that T6060s would be M6, yes?

Gurman said Hidra was a Ultra-tier chip and it turned out to be very wrong. Regarding silicon codenames Gurman's track record is not as good as his record on final products.

That's true.

I will.

Currently have nothing with which to compare, so eyes-wide-shut.

Aye the closest thing would be an M5 Pro/Max box when it finally arrives. I'm actually impressed they managed to get a GPU with its size into that enclosure. When you get a chance would you mind running something like Geekbench GPU OpenCL on it (I don't think it has support for Vulkan)?

 

[wmy5]

Looking at the pattern though I'm pretty sure those are M6 Pro/Max/Ultra. He may have placed them after M6 in the chart, but look at the numbers:

M3 Pro - Ultra: T6030s
M4 Pro - Max: T6040s
M5 Pro - Ultra: T6050s

So it would stand to reason that T6060s would be M6 Pro-Ultra would it not?

Other than CPID there is PlatformName, like H17G. The number indicates the generation, and suffix indicates the chip type. This has been fairly consistent in the past.

G = base Mac or iPad Pro chip (Mx)
A = base iPhone chip (Axx)
P = Pro iPhone chip (Axx Pro)
S = Max "chop" chip (Mx Pro)
C = full Max chip (Mx Max)
D = Ultrafusion'd Max chip (Mx Ultra)

T6060, T6061, T6062 happens to be H19S, H19C, H19D respectively. If they were M6 Pro/Max/Ultra they would have been H18S, H18C and H18D.

But of course I could be totally wrong. Everything on my list is an educated guess at best.

 

[crazy dave]

Other than CPID there is PlatformName, like H17G. The number indicates the generation, and suffix indicates the chip type. This has been fairly consistent in the past.

G = base Mac or iPad Pro chip (Mx)
A = base iPhone chip (Axx)
P = Pro iPhone chip (Axx Pro)
S = Max "chop" chip (Mx Pro)
C = full Max chip (Mx Max)
D = Ultrafusion'd Max chip (Mx Ultra)

T6060, T6061, T6062 happens to be H19S, H19C, H19D respectively. If they were M6 Pro/Max/Ultra they would have been H18.

But of course I could be totally wrong. Everything on my list is an educated guess at best.

That's a good point. I still would like to see the original document Ilikeiphone saw though - this isn't quite "an extraordinary claim requires extraordinary evidence" but it's certainly an eyebrow raising claim that I'd want to see confirmation of from another source because, to me, the M6 would've been one of the least likely chip generations to skip over Pro/Max SOCs given what everything else we know, or at least think, is coming.

 

[splifingate]

When you get a chance would you mind running something like Geekbench GPU OpenCL on it (I don't think it has support for Vulkan)?

Certainly.

 

[deconstruct60]

Other than CPID there is PlatformName, like H17G. The number indicates the generation, and suffix indicates the chip type. This has been fairly consistent in the past.

T6060, T6061, T6062 happens to be H19S, H19C, H19D respectively. If they were M6 Pro/Max/Ultra they would have been H18S, H18C and H18D.

But of course I could be totally wrong. Everything on my list is an educated guess at best.

The MBP 14/16 , Mni Pro , and Mac Studio could just be leaving the more mainstream stuff behind on the 'platform' side. Way better I/O ( PCI-e v5 ? , display out upgrade , or something not CPU or GPU core centric. ) . Apple's SDHC offerings are not all that top end . The Mac Pro is definitely dragging behind on PCI-e slot backhaul bandwidth relative to competitive offerings. There is a fair amount of stuff on the dies besides just more general compute 'cores' . All really need is some I/O that doesn't need to trickle down to the iPads and the entry MBA and basically shift the 'platform' technology.

Similarly Apple weaving some aspects of N-family ( follow on to N1) onto the main SoC package. Bigger package more places could stuff that if did something 'exotic' with the packaing. Again a bigger platform shift for those than for the 'plain' Mn on that generation.


P.S. if the Mn + Rm combo is going to be around long term. The iPad and the Vision reuirements should get some drift from the stuff that is gong to be Mac deployed only. The 'cores' are reusable , but the focus points on I/O should get substantively different.

 

[crazy dave]

The MBP 14/16 , Mni Pro , and Mac Studio could just be leaving the more mainstream stuff behind on the 'platform' side. Way better I/O ( PCI-e v5 ? , display out upgrade , or something not CPU or GPU core centric. ) . Apple's SDHC offerings are not all that top end . The Mac Pro is definitely dragging behind on PCI-e slot backhaul bandwidth relative to competitive offerings. There is a fair amount of stuff on the dies besides just more general compute 'cores' . All really need is some I/O that doesn't need to trickle down to the iPads and the entry MBA and basically shift the 'platform' technology.

Similarly Apple weaving some aspects of N-family ( follow on to N1) onto the main SoC package. Bigger package more places could stuff that if did something 'exotic' with the packaing. Again a bigger platform shift for those than for the 'plain' Mn on that generation.


P.S. if the Mn + Rm combo is going to be around long term. The iPad and the Vision reuirements should get some drift from the stuff that is gong to be Mac deployed only. The 'cores' are reusable , but the focus points on I/O should get substantively different.

That's an interesting idea. I like it. In addition, I've always found it odd that the platform IDs for A-series chips was one ahead of the M-series chips (e.g. A19/Pro is H18 while the M5s are H17). Apple updating Pro/Max, but not the base M chip, in such a way that they have a higher platform ID (and one that matches the A-series) could definitely work as a solution here.

 

[wmy5]

I've always found it odd that the platform IDs for A-series chips was one ahead of the M-series chips (e.g. A19/Pro is H18 while the M5s are H17).

This is indeed quite odd. Also M series chip's NPU is one generation behind the contemporary A series chip but the CPU and GPU are on par. For example:

	PlatformID	CPU	GPU	NPU
M5	H17G	Everest V4 + Sawtooth V4	Apple10 family G17G	H17_Theia_J73y
A18 Pro	H17P	Everest V3 + Sawtooth V3	Apple9 family G17P	H17_Theia_D9x
A19 Pro	H18P	Everest V4 + Sawtooth V4	Apple10 family G18P	H18_Apollo_V5x

 

[crazy dave]

This is indeed quite odd. Also M series chip's NPU is one generation behind the contemporary A series chip but the CPU and GPU are on par. For example:

	PlatformID	CPU	GPU	NPU
M5	H17G	Everest V4 + Sawtooth V4	Apple10 family G17G	H17_Theia_J73y
A18 Pro	H17P	Everest V3 + Sawtooth V3	Apple9 family G17P	H17_Theia_D9x
A19 Pro	H18P	Everest V4 + Sawtooth V4	Apple10 family G18P	H18_Apollo_V5x

Aye, I think I'm going with @deconstruct60's idea as being more likely then - that this is evidence that Apple will be upgrading the platform for the Pro/Max in the M6 generation rather than skipping the M6 generation for those SOCs. Will be interesting to see what that means in practice if so.

What does GXXY mean at the end of the GPU family? M5 is Apple 10 family like A19 Pro but G17 like the A18 Pro?

 

[Gnattu]

One M5 GB6 in the MacBook Pro is online:

Mac17,2 - Geekbench

Benchmark results for a Mac17,2 with an Apple M5 processor.

browser.geekbench.com

The one in the MacBook Pro has a slightly higher frequency(4.6GHz) compared with the one leaked earlier in an iPad Pro(4.4GHz)

 

[Pressure]

One M5 GB6 in the MacBook Pro is online:

Mac17,2 - Geekbench

Benchmark results for a Mac17,2 with an Apple M5 processor.

browser.geekbench.com

The one in the MacBook Pro has a slightly higher frequency(4.6GHz) compared with the one leaked earlier in an iPad Pro(4.4GHz)

That's faster than the 12-core M2 Max Mac Studio I still use 😅

That's quite an improvement for 18 months.

 

[smalm]

Apple could do what they did with the 2018 15" to 2019 16" MBP: drop the same Coffee Lake 9th gen, betting on a chassis redesign with various hardware upgrades is enough to sell.

It is completely irrelevant what Apple did 2018/2019.

 

[cbum]

That's faster than the 12-core M2 Max Mac Studio I still use 😅

View attachment 2568771

That's quite an improvement for 18 months.

Those % difference numbers may not mean what you think they mean.

 

[Pressure]

Those % difference numbers may not mean what you think they mean.

I can see the individual test scores.

They mean exactly what the score indicates 🤔

 

[wmy5]

What does GXXY mean at the end of the GPU family? M5 is Apple 10 family like A19 Pro but G17 like the A18 Pro?

No idea. Those are the identifiers found in Metal API.

Another possibility is Apple will skip M7-M11 names, because those two are already used by motion co-processor since iPhone 5S 🙂

 

[AshesXY]

No idea. Those are the identifiers found in Metal API.

Another possibility is Apple will skip M7-M11 names, because those two are already used by motion co-processor since iPhone 5S 🙂

M2027 🙃

 

[Antony Newman]

If the M3 Ultra had roughly circa 35% GPU performance uplift compared to the M3 Max,
would M5 Max perform comparably to an M4 Ultra (if it existed) in the GPU department?

And if the M5 Max is a jump in performance over the M3 Ultra - what is the MacOs App that
is pushing Apple to create Ultra versions of Apple Silicon going forward?

If the M6 Max represents another 25% jump in GPU performance - are we hitting a point where 98% of power users feel they have enough performance for their software needs?