Rumor: new iPads will have M4

[Cmaier]

Usually there is a flurry of new patents before new SoC release. Nothing like that happened on the last few months.

There were a bunch of new NPU-related patents last year, so again, I can imagine that the iPad gets some sort of M3 AI edition, but nothing beyond that.

I think you may be reading too much into what is likely coincidence. I assume you are referring to patent applications, not patents - there is no way to control when patents actually issue (i mean, you can delay issuance if you are willing to play games and risk issuance, but you can’t speed it up), but you can control when you file an application.

But patent applications aren’t generally published until 18 months after they are filed (and in some cases aren’t published until they issue as patents). Until they are published, patent applications cannot be seen by the public. So if you see a flurry of patent applications in the months before an SoC becomes available, that generally means those patents were filed a year and a half earlier.

I don’t believe Apple is on an 18-month SoC design cycle at this point - I think it‘s closer to 12 months - so it’s unlikely that patent applications becoming available to the public today correspond to work that was done on SoC’s becoming available to the public today.

1120-Eighteen-Month Publication of Patent Applications

www.uspto.gov

 

[throAU]

Usually there is a flurry of new patents before new SoC release. Nothing like that happened on the last few months.

There were a bunch of new NPU-related patents last year, so again, I can imagine that the iPad gets some sort of M3 AI edition, but nothing beyond that.

It could be just as simple as a die shrink with minor tweaks / improved process to get more chips out of a wafer or better power consumption. if they can get more chips out of a die with similar yields, maybe that's enough to justify M4.

 

[Jimmyjames]

Gurman still sticking with his M4 prediction. At least it spices up a potentially mundane event!

 

[Cmaier]

Gurman still sticking with his M4 prediction. At least it spices up a potentially mundane event!

This is going to be a big pain for me at work - I just got finished analyzing the M3, and now already I have to deal with the M4!

 

[Yoused]

I believe that the M4 is all in Gurman's head. Clearly they have to work on the processes underlying the large model, because he is seriously glitching.

 

[casperes1996]

I believe that the M4 is all in Gurman's head. Clearly they have to work on the processes underlying the large model, because he is seriously glitching.

I’m inclined to agree. I at least won’t be disappointed to not see it tomorrow. And I’d be pleasantly surprised if it does happen but my expectation is firmly that it won’t.

 

[Altaic]

I’m inclined to agree. I at least won’t be disappointed to not see it tomorrow. And I’d be pleasantly surprised if it does happen but my expectation is firmly that it won’t.

I think the M4 may debut tomorrow as an appetizer for WWDC wrt machine learning. I have a sneaking suspicion that the ANE will transform (npi) on both the hardware and software sides, particularly for more capable devices.

 

[leman]

I think you may be reading too much into what is likely coincidence. I assume you are referring to patent applications, not patents - there is no way to control when patents actually issue (i mean, you can delay issuance if you are willing to play games and risk issuance, but you can’t speed it up), but you can control when you file an application.

But patent applications aren’t generally published until 18 months after they are filed (and in some cases aren’t published until they issue as patents). Until they are published, patent applications cannot be seen by the public. So if you see a flurry of patent applications in the months before an SoC becomes available, that generally means those patents were filed a year and a half earlier.

I don’t believe Apple is on an 18-month SoC design cycle at this point - I think it‘s closer to 12 months - so it’s unlikely that patent applications becoming available to the public today correspond to work that was done on SoC’s becoming available to the public today.

1120-Eighteen-Month Publication of Patent Applications

www.uspto.gov

What I was trying to say is that until now we had a very consistent pattern of relevant patents appearing close to the hardware release. That they submit them years in advance is obvious, these kind of architectural changes are not developed overnight. I do not know anything about patent publication process (that's your expertise), at the same time it does not seem like a coincidence either. Even more, the patent trail aligns with the rumor mill. For example, remember the rumors that the new GPU with raytracing was supposed to be part of the A16, but Apple canned that in the last minute because of TSMC delays and issue back porting the GPU to 5nm? Well, the GPU RT patents were initially published early 2022, supporting the idea that the tech was delayed. And so far all functionality-related patents appear to be timed like this (other patents like the packaging technology instead do not appear to be timed). I fully agree with you that

So far pretty much all SoC-related features Apple has patented have been implemented. The only big omission is the new NPU architecture which was described in multiple patents last year.

Of course, all of this is conjecture and supposition at best.

 

[Altaic]

The only big omission is the new NPU architecture which was described in multiple patents last year.

I’m really curious about the NPU patents that’s you’re talking about from last year. Would you mind elaborating?

 

[jido]

Here it is doubters, the new iPad has a M4.

 

[exoticspice1]

iPad16,6 - Geekbench

Benchmark results for an iPad16,6 with an ARM processor.

browser.geekbench.com

 

[Cmaier]

iPad16,6 - Geekbench

Benchmark results for an iPad16,6 with an ARM processor.

browser.geekbench.com

so around 20% improvement?

Not too shabby. The cores have to be new and different microarchitecture than M3 cores.

 

[jido]

Over at MR they think it’s partly due to SME support on the M4 (new).