LPDDR6: New Beginning of AS: 12GB Standard

[deconstruct60]

If I were to predict where Apple would go with AI training on their hardware, it’s probably gonna be small models on a single machine, or dedicated towards things like image and music generation. Y’know, stuff that helps individual “creative” types.

There is no inherent coupling with 'training' has to be done on exactly the same hardware as 'inference' is done . There is an Nvidia inspired hype-train that it has to be the same hardware, but it not necessarily so. The 'parameters' of the "neural model" have to be figured out in training, but when done you essentially have a piece of 'software'; not hardware.

Now is Apple's training exercises going to be highly skewed to running well on Apple specific inference hardware? Yes. But they don't need to use Apple's inference hardware to do the training.

 

[MRMSFC]

What kind of applications do you think would benefit from it? How would an accelerator card be better than making the GPU more flexible for training? If you go the accelerator route, you need fast interconnect, and if you have a fast interconnect, why not simply allow multiple SoC boards in the system?

Good point, I just had some random thoughts.

 

[Allen_Wentz]

I actually believe it. From another post I wrote:

"RAM Defaults in Macs?

What was the default RAM in Macs, how long did it stay, and when did it change? For simplicity, I’ll start from the Steve Jobs era, and I will only focus on the consumer portables.

32MB. This was used in the iBook and iBook Special Edition in July, 1999 until it was discontinued in September, 2000. It lasted 14 months.

64MB. This was used in the Firewire iBooks from September 2000 until the Dual USB iBooks in October 2001. It lasted 13 months.

128MB. This was used in the late 2001 iBooks released in October 2001 through the iBook G4, discontinued in April, 2004. It lasted 32 months

256MB. This was used in the Early 2004 iBook G4 released in April, 2004 through the late 2004 iBook G4 released in October 2004, sold until July, 2005. It lasted 15 months.

512MB. This was used from the Mid 2005 iBook G4 released in July, 2005, until the Mid 2007 MacBook, which was sold until November, 2007. It lasted 28 months.

1GB. This was used in the Late 2007 MacBook released on November, 2007 until the MacBook Late 2008, discontinued in January, 2009. It lasted 25 months.

2GB. This was used in the Late 2008 AL MacBook released on October, 2008, until the MacBook Air 11”, Mid 2011, which was discontinued n June, 2012. This lasted 44 months.

4GB. This was used in the MacBook Air 13”, Mid 2011 released in July, 2011 until the MacBook Air, 13” Early 2015, which was discontinued in June, 2017. This lasted for 71 months.

8GB. This was used as standard on MacBook Air, 2017, released on June, 2017 until now, in November, 2023. 77 months and counting.

Clearly, the amount of time machines stayed at 1GB was half what the time for 2GB, which was almost half of what 4GB. I expect 16GB to become standard in another 40 months or thereabouts, if trends continue. "

So, 16GB as default should happen in about 60 months, or 5 years. But there's no reason why Apple can't increase it by 50%, or 12GB of RAM, and then do another 4GB increase in another 2 1/2 years.

It'll meet the timeline and past trends.

Like I previously commented, IMO focusing on base RAM is just wrong; who cares? Just folks wanting more for less <yawn>. There are a zillion reasons why Apple may choose its base RAM level.

What really matters is the max RAM Apple offers in laptops. Post that timeline: 16 GB max available 2016 and 128 GB available 2024. That trend IMO tells us something. The base RAM evolution tells us nothing really important.

 

[MayaUser]

So now with the M4 out...people still thinks the iphone pro will start with 12gb ram ?!

 

[hovscorpion12]

So now with the M4 out...people still thinks the iphone pro will start with 12gb ram ?!

Yes. The iPhone is completely separate from the M-series. Currently the 8GB on the 15 Pro is sufficient, however in 2025 when the 17 Pro launches, that’s a whole new story.

Also, every flagship Android phone comes with 12GB. Apple will join that club soon.

 

[MayaUser]

Yes. The iPhone is completely separate from the M-series. Currently the 8GB on the 15 Pro is sufficient, however in 2025 when the 17 Pro launches, that’s a whole new story.

Also, every flagship Android phone comes with 12GB. Apple will join that club soon.

yes, but its about SoC design...so if the M4 is 8gb ram...there are 0% for iphone to have 12gb ram in A17 pro
Both are probably build on N3E
But yes, the iphone 17 pro, will probably have higher Ram amount

 

[Populus]

On the keynote they mentioned that the new M4 has a higher memory bandwidth.

Could it be LPDDR6?

 

[hovscorpion12]

On the keynote they mentioned that the new M4 has a higher memory bandwidth.

Could it be LPDDR6?

Bro. Don't temp me

 

[theorist9]

On the keynote they mentioned that the new M4 has a higher memory bandwidth.

Could it be LPDDR6?

As I mentioned earlier:

But they haven't even gone to LPDDR5X yet. I'm not sure if they're going to bypass LPDDR5X and go directly to LPDDR6.

And according to Ryan Smith of Anandtech, the M4 in the iPad is indeed using LPDDR5X rather than LPDDR6:

"M4 Memory: Adopting Faster LPDDR5X

Last, but certainly not least, the M4 SoC is also getting a notable improvement in its memory capabilities. Given the memory bandwidth figures Apple is quoting for the M4 – 120GB/second – all signs point to them finally adopting LPDDR5X for their new SoC.

The mid-generation update to the LPDDR5 standard, LPDDR5X allows for higher memory clockspeeds than LPDDR5, which topped out at 6400 MT/second. While LPDDR5X is available at speeds up to 8533 MT/second right now (and faster speeds to come), based on Apple’s 120GB/second figure for the M4, this puts the memory clockspeed at roughly LPDDR5X-7700.

Since the M4 is going into an iPad first, for the moment we don’t have proper idea of its maximum memory capacity. The M3 could house up to 24GB of memory, and while it’s highly unlikely Apple has regressed here, there’s also no sign whether they’ve been able to increase it to 32GB, either. In the meantime, the iPads Pro will all either come with 8GB or 16GB of RAM, depending on the specific model."

Apple Announces M4 SoC: Latest and Greatest Starts on 2024 iPad Pro

www.anandtech.com

 

[JPack]

On the keynote they mentioned that the new M4 has a higher memory bandwidth.

Could it be LPDDR6?

No. While this thread was entertaining while we waited for M4, it was completely unrealistic.

Apple Announces M4 SoC: Latest and Greatest Starts on 2024 iPad Pro

www.anandtech.com

 

[Populus]

@theorist9 @JPack Thank you both for the clarification. I just saw the entire keynote a few minutes ago.

So basically we went from 100GB/s to 120GB/s. A bit underwhelming but hey, at least there’s an improvement.

I’ll go to the M4 thread to read and discuss about the new Neural Engine (I think that’s the A17 Pro NPU), the new CPU architecture (now ML-capable) and all the other goodies.

 

[TigeRick]

Apple surprised me by launching M4 SoC with iPad Pro. Are you guys feeling something is not right cause for the first time Apple is launching latest generation of M4 in iPad Pro but not in Mac lineup? I have listed all the M-series launching, MacBook always get to launch first before iPad Pro lineup....

Hmm, I am still underestimating Apple to segment the Apple SoC. With M4 + 8GB 128-bit LPDDR5X, it occurs to me that Apple might be further splitting the M4 SoC into two: one for iPad and another one for MacBook. Please have a look on the second table in the second thread; the first M1 SoC was using LPDDR4X while M1 Pro/Max were using LPDDR5. I think we might see same segmentation, but this round is between iPad and Mac.

I know it sounds weird, but how can you explain the first launching of M4 SoC appears in iPad Pro lineup with starting price $999?

Can you imagine Apple going to launch MBP14 with 8GB RAM for $1599?

Have you wonder why does Apple launch new M4 iPad Pro before WWDC is up?

iPadOS do not need more than 8GB of RAM thus it is making sense for iPad Pro to bundle standard 8GB RAM. However, MacOS requires more RAM for multi-tasking, thus 12GB RAM should be standard. And I have explained the benefits of LPDDR6; from technical and BOM LPDDR6 has clear advantages over LPDDR5X.

I know most of you won't believe my theory. I don't mind at all, after all this is purely my speculation, no one in the whole Internet is believing LPDDR6 as strongly as me.

Let's see whether Apple will launch any Mac with M4-series in WWDC? If yes, forget my thread and move on. If no, then we should see Apple adopts LPDDR6 in upcoming iPhone 16 Pro lineup. Mac lineup should get updated by the end of the year as claimed by Mark Gurman. Now we all know how good is his source...

 

[leman]

Apple surprised me by launching M4 SoC with iPad Pro. Are you guys feeling something is not right cause for the first time Apple is launching latest generation of M4 in iPad Pro but not in Mac lineup? I have listed all the M-series launching, MacBook always get to launch first before iPad Pro lineup....

Hmm, I am still underestimating Apple to segment the Apple SoC. With M4 + 8GB 128-bit LPDDR5X, it occurs to me that Apple might be further splitting the M4 SoC into two: one for iPad and another one for MacBook. Please have a look on the second table in the second thread; the first M1 SoC was using LPDDR4X while M1 Pro/Max were using LPDDR5. I think we might see same segmentation, but this round is between iPad and Mac.

I think we will see desktop Macs with M4 at the WWDC. M4 has introduced some features relevant for high-performance computing (in particular, it's the first ARM CPU to support the SME extensions), so it would make sense to put it into the next Mac Pro. I doubt the MBP will be updated with M4, and I believe Apple will go straight to M5 for the prosumer laptops in late 2024.

 

[TigeRick]

I think we will see desktop Macs with M4 at the WWDC. M4 has introduced some features relevant for high-performance computing (in particular, it's the first ARM CPU to support the SME extensions), so it would make sense to put it into the next Mac Pro. I doubt the MBP will be updated with M4, and I believe Apple will go straight to M5 for the prosumer laptops in late 2024.

Yep, SME seems pretty impressive even for GeekBench.

Apple would most likely breaking 100 billion transistors in upcoming M4 Max. And that's another reason I believe Apple will split the M4 into two. M4 for iPad is fabbed by N3E, power optimised process. Meanwhile, M4 for Mac will be based on N3B, density optimised process.

 

[TigeRick]

Gurman’s report suggests that the biggest picture is what you might expect: There will be an M4 chip (code-named Donan) in the MacBook Air, low-end MacBook Pro, and low-end Mac mini; and an M4 code-named Brava in the high-end MacBook Pros and high-end Mac mini.

Gurman suggests that Apple is testing versions with both a “still-unreleased” M3 chip and a variation of the M4 Brava processor.

Gurman mentioned about Apple is testing still-unreleased M3 chip; it seems this is the SoC used by iPad Pro as M4. If my theory is correct, M4 for iPad SHOULD NOT be code-named "Donan". Anyone could verify the code name of M4 for iPad?

 

[leman]

If my theory is correct, M4 for iPad SHOULD NOT be code-named "Donan". Anyone could verify the code name of M4 for iPad?

"Donan" is an identifier used in the new Xcode, it most likely refers to M4.

 

[DaniTheFox]

Yep, SME seems pretty impressive even for GeekBench.

Apple would most likely breaking 100 billion transistors in upcoming M4 Max. And that's another reason I believe Apple will split the M4 into two. M4 for iPad is fabbed by N3E, power optimsed process. Meanwhile, M4 for Mac will be based on N3B, density optimsed process.

It looks like Apple (TSMC) were faster than the industrie with next generation memory technology. And if new processes comes by TSMC in this fast succession I think Apple will call it M5.
But don’t worry. The time will come with 12 GB as base memory as you predicted. AI needs a lot. Who cares then about a version number of the SoC.

 

[Confused-User]

I think we will see desktop Macs with M4 at the WWDC. M4 has introduced some features relevant for high-performance computing (in particular, it's the first ARM CPU to support the SME extensions), so it would make sense to put it into the next Mac Pro. I doubt the MBP will be updated with M4, and I believe Apple will go straight to M5 for the prosumer laptops in late 2024.

Wow, I thought I was the optimist.

What makes you think they will introduce two generations of M chip in barely more than half a year? That seems very unlikely to me.

I think you might be right about the scheduling, though I also think it's possible we will see M4 prosumer at WWDC, or in July.

 

[Confused-User]

Apple would most likely breaking 100 billion transistors in upcoming M4 Max. And that's another reason I believe Apple will split the M4 into two. M4 for iPad is fabbed by N3E, power optimsed process. Meanwhile, M4 for Mac will be based on N3B, density optimsed process.

That's delusional.

There is no way they're doing M4, or anything else, for N3B. That process was one-and-done, for all the reasons we've discussed previously. I will bet every one of my body parts, against a single dollar, that they will not lay out the entire processor twice, once for N3E and once for N3B. That would be the single dumbest decision made since... probably since someone won a darwin award.

"Density" is the last thing Apple cares about. They care about power and performance, usually in that order, as you can see for example with their display controllers. To the extent that they want better density (really, lower area) - for example, plausibly, with those same controllers - they'll use N3E FinFlex 2-1 or 2-2 for those transistors, where compatible with their performance targets.

Everything will stay on N3E until N3P hits. They will move to that, and then on to N2 afterwards.

 

[JPack]

That's delusional.

There is no way they're doing M4, or anything else, for N3B. That process was one-and-done, for all the reasons we've discussed previously. I will bet every one of my body parts, against a single dollar, that they will not lay out the entire processor twice, once for N3E and once for N3B. That would be the single dumbest decision made since... probably since someone won a darwin award.

"Density" is the last thing Apple cares about. They care about power and performance, usually in that order, as you can see for example with their display controllers. To the extent that they want better density (really, lower area) - for example, plausibly, with those same controllers - they'll use N3E FinFlex 2-1 or 2-2 for those transistors, where compatible with their performance targets.

Everything will stay on N3E until N3P hits. They will move to that, and then on to N2 afterwards.

"Hey, let's propose to our boss Johny Srouji, that instead of working on M5 or M6, let's use 20% of the silicon design team to lay out M4 again on N3B! That's a great way to spend a couple hundred million dollars on masks, time, and materials!"

 

[SBeardsl]

Everything will stay on N3E until N3P hits. They will move to that, and then on to N2 afterwards.

and when N3P is ready, Apple will start releasing chips labeled M5.

The interesting question is will N3P/M5 be in time for the "late 2024/early 2025" window to update the MacBooks or will their next round of updates be M4?

 

[Confused-User]

and when N3P is ready, Apple will start releasing chips labeled M5.

The interesting question is will N3P/M5 be in time for the "late 2024/early 2025" window to update the MacBooks or will their next round of updates be M4?

The next round will almost certainly be M4, starting next month at WWDC, though it's at least as likely that will only be desktops, and we won't see the laptops until this summer.

 

[TigeRick]

It looks like Apple (TSMC) were faster than the industrie with next generation memory technology. And if new processes comes by TSMC in this fast succession I think Apple will call it M5.
But don’t worry. The time will come with 12 GB as base memory as you predicted. AI needs a lot. Who cares then about a version number of the SoC.

Yep, maybe I should just stick to 12GB RAM, people won't argue so much about LPDDR6. But what fun would that be? Plus I want to provide more technical info about LPDDR6 technology instead of pep-talk.

Everyone and Apple knows 8GB is not enough for next generation Mac lineup. Yet people don't question why Apple bundled 8GB LPDDR5X with M4 SoC?

Maybe Apple would bundle 16GB LPDDR5X with upcoming MBP14???

Isn't 16GB LPDDR5X more expensive than 12GB LPDDR6? In case people don't read the frontpage, 16GB LPDDR5X would require 8 pcs of 2GB memory die. 12GB LPDDR6 only requires 3 pcs of 4GB memory die. Which one is making more sense in terms of technical and business POV? I actually omitting memory bus and chips difference.

I bet Apple is working on LPDDR6 right after they launched M1 Pro/Max with 512-bit LPDDR5. It is truly next gen memory technology that is fabbed by EUV machinery. FYI, M1 series are manufactured by TSMC N5 process, the first process to utilize EUV machinery (N6 is lower cost version). Do you think Apple of all OEMs don't know the advantages of memory chip build by EUV?

Anyhow, let's wait WWDC to verify my speculation. Wait a minute, Mark Gurman should leak about upcoming WWDC if there are Mac being launched. Maybe we should wait for Gurman's leaks in the coming weeks? OR NOT...

 

[senttoschool]

I think LPDDR6 is unlikely. But I commend @TigeRick for making the effort on his prediction. It's a far better post than 99% of the stuff written here.

 

[senttoschool]

I think we will see desktop Macs with M4 at the WWDC. M4 has introduced some features relevant for high-performance computing (in particular, it's the first ARM CPU to support the SME extensions), so it would make sense to put it into the next Mac Pro. I doubt the MBP will be updated with M4, and I believe Apple will go straight to M5 for the prosumer laptops in late 2024.

Do you think Apple will release M4 Ultra for Studio/MacPro before an M4 Max for MBPs?