Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M4 Family discussion here:

Page 263 - Discussion - Apple Silicon SoC thread

Page 263 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[mvprod123]

https://twitter.com/x/status/1896633519692280248

 

[mvprod123]

Apple introduces iPad Air with powerful M3 chip and new Magic Keyboard

Apple today introduced the new iPad Air, powered by the M3 chip and built for Apple Intelligence.

www.apple.com

The new M3 iPad Air (9-core GPU) and iPad with A16

 

[mvprod123]

https://twitter.com/x/status/1896633519692280248

I tried to lighten the image. It looks like the M4 Max doesn't have an UltraFusion connection again.

 

[Eug]

Apple introduces iPad Air with powerful M3 chip and new Magic Keyboard

Apple today introduced the new iPad Air, powered by the M3 chip and built for Apple Intelligence.

www.apple.com

The new M3 iPad Air (9-core GPU) and iPad with A16

The AI pundits had me almost convinced that the base iPad would get A17 and 8 GB RAM. Apple actually went with my original prediction of A16. Along with A16 I was thinking 6 GB but we don't have confirmation of that yet.

If 6 GB that would be great, because that would be a 50% increase in RAM. Also more tangible to most people is they doubled the storage to 128 GB, which is something I wasn't 100% convinced they'd do this generation.

BTW, they are using a binned A16 with 5-core CPU (and 4-core GPU). I was not expecting that, but it will have almost no impact to most users in this price class. Certainly, I don't care about that. I'm much more happy to see the doubling of storage and hopefully 50% increase in RAM.

This will be a perfect replacement for my kid's 2 GB / 64 GB iPad Air 2.

 

[mvprod123]

View attachment 118855

I tried to lighten the image. It looks like the M4 Max doesn't have an UltraFusion connection again.

Apple M4 Pro/Max is codename Brava.

M4 Pro (T6040) = Brava S
M4 Max (T6041) = Brava C
M4 (T8132) = Donan.

Mark Gurman mentioned that a high-end Hidra chip is also in development. The lack of UltraFusion in the M4 Max confirms that the M4 Ultra is a separate die.

 

[mvprod123]

https://twitter.com/x/status/1896972159731556579

LOL. M3 Ultra?

 

[okoroezenwa]

https://twitter.com/x/status/1896972159731556579

LOL. M3 Ultra?

Having a very hard time believing this.

 

[jdubs03]

Having a very hard time believing this.

Yeah.The definitely seems suspect. The M4 itself has been out 9 months, and the M3 for what 15? I’d be disappointed if the Studio just wasn’t a binned M4 Ultra.

 

[Eug]

https://twitter.com/x/status/1896972159731556579

LOL. M3 Ultra?

If it's a separate chip as some have suggested (and not 2 x Mx Max), then perhaps it's a design in between the M3 and M4 generations.

 

[Doug S]

Now if Apple and Google can get carriers to coordinate more it may aid in simplifying the cellular mess. Qualcomm has a vested interest in keeping their IP in the mix and that could be a complicating factor. Apple and Google have some cellular IP from Infineon and Motorola but I doubt they want to complicate connectivity. Google makes money on ads, Apple makes money on hardware and services and neither make money on cellular IP. In ten years the baseband chip may be relatively simple and work globally.

A lot of the problem is the different bands in different countries, and different laws about stuff like power levels. That's not something within Apple or Google's power.

Besides, Apple doesn't care now that they've done the heavy lifting of implementing their own cellular baseband. Why make it easier for others to follow in their footsteps, now that they have (or rather soon will have) freed themselves from Qualcomm they don't care if Qualcomm is terrible to deal with and charges a lot of money - every additional dollar Qualcomm sucks into its maw is one less dollar per phone that the Android competition is able to invest.

 

[The Hardcard]

If it's a separate chip as some have suggested (and not 2 x Mx Max), then perhaps it's a design in between the M3 and M4 generations.

Not to mention the M5 is also already in production. The Ultra might be a stopgap chip. Apple has been working on the M5 Ultra for a while, but it involves a lot of new technology and was projected to possibly not appear until sometime in 2026.

That was going to make for a huge lost opportunity since there is a rapidly growing demand for high capacity RAM Macs for running large language, models, and other generative AI. There is strong evidence that Nvidia’s Project Digits will only have 273 GB/s memory bandwidth just like AMD’s Strix Halo, half that of the M4 Max and down to 1/4 of an Ultra if the Ultra how’s the same memory speed.

Dylan Patel from SemiAnalysis made a post on X on January 5 that included several Apple code names supposedly all coming an N3P - Theras, Tilos, Hidra, Sotra, Baltra, Isonoe to be N3P. Hidra and Baltra have been identified as the big chips, the others are likely the M5 set, base, Pro, and Max.

 

[poke01]

Not to mention the M5 is also already in production. The Ultra might be a stopgap chip. Apple has been working on the M5 Ultra for a while, but it involves a lot of new technology and was projected to possibly not appear until sometime in 2026.

That was going to make for a huge lost opportunity since there is a rapidly growing demand for high capacity RAM Macs for running large language, models, and other generative AI. There is strong evidence that Nvidia’s Project Digits will only have 273 GB/s memory bandwidth just like AMD’s Strix Halo, half that of the M4 Max and down to 1/4 of an Ultra if the Ultra how’s the same memory speed.

Dylan Patel from SemiAnalysis made a post on X on January 5 that included several Apple code names supposedly all coming an N3P - Theras, Tilos, Hidra, Sotra, Baltra, Isonoe to be N3P. Hidra and Baltra have been identified as the big chips, the others are likely the M5 set, base, Pro, and Max.

Theras, Tilos, and Isonoe are M5 generation since all relate to greek/greece city or myth
Sotra is a Norway island - So A19?
Hidra is a Norway island
Baltra is for AI server

 

[The Hardcard]

Theras, Tilos, and Isonoe are M5 generation since all relate to greek/greece city or myth
Sotra is a Norway island - So A19?
Hidra is a Norway island
Baltra is for AI server

An earlier Digitimes rumor/report stated that Apple planned to use the same chips for the upcoming server and consumer devices just as the current servers use the same M2 Ultra as the Mac Studios (though likely with a different PCB and case). While the Baltra information just talked about server, I didn’t see anything indicating they had changed plans on the dual use.

Combined with the claim that the expensive server-grade packaging is going to be used on the lower end Pro and Max variants, I don’t yet see any reason to presume that consumers won’t get some variation of Baltra. The Apple servers are a sea of small machines handling not many concurrent users each, at least that was the earlier plan. The Baltra announcement didn’t say anything about that changing.

The design is similar to Google’s TPU infrastructure which is also a giant volume of small servers that can’t maintain the concurrency of traditional servers. Google decided that this setup was more effective for serving clioud inference users.

The designer of that Google TPU cloud was Sumit Gupta. He has been at Apple for the last few years heading up the Private Cloud Compute rollout. Of course, companies can change directions any time they choose, but until such change is revealed, the likelihood is that Baltra will be able to also be a consumer chip as well, either an Ultra or maybe even an Extreme.

- - -

The main problem I have with an M3 Ultra is that top Studio has to have Thunderbolt 5. If they put TB5 in it, why wouldn’t they also just use the rest of the nesrly year old M4 IP as well. Plus, the M5 is already in production. They are going to release a new Ultra which would again very quickly have 2 generations old CPU and GPU cores?

 

[sallymander]

I stumbled on this forum post about passmark on M3/M4 - curiously passmark (unlike other benchmarks I've seen) gets better single thread results for the M3. They suggest it may be something to do with throttling (although with only a single core load I would be quite surprised).

Apple M4 chips - PassMark Support Forums

Hello. They should be at the top in single-threaded rating. Why they don't? Thanks.

forums.passmark.com

PassMark CPU Benchmarks - Single Thread Performance

Benchmarks of the single thread performance of CPUs. This chart comparing CPUs single thread performance is made using thousands of PerformanceTest benchmark results and is updated daily.

www.cpubenchmark.net

They also have the M4 Max lower than the regular M4, which again is usually the reverse of what we see (higher memory bandwidth available to the Max CPUs tends to give them a little advantage in some single core tests).

 

[Nothingness]

I stumbled on this forum post about passmark on M3/M4 - curiously passmark (unlike other benchmarks I've seen) gets better single thread results for the M3. They suggest it may be something to do with throttling (although with only a single core load I would be quite surprised).

They are compiling with XCode 12.5 which dates back to 2021 which might explain the odd result.

 

[mvprod123]

M3 Ultra with Thunderbolt 5 and up to 512 GB unified memory!!!

Apple reveals M3 Ultra, taking Apple silicon to a new extreme

Apple today announced M3 Ultra, offering the most powerful CPU and GPU in a Mac.

www.apple.com

 

[mvprod123]

Mac Studio with M4 Max and M3 Ultra

Apple unveils new Mac Studio, the most powerful Mac ever

Apple today announced the new Mac Studio, the most powerful Mac ever made, featuring M4 Max and the new M3 Ultra chip.

www.apple.com

 

[mvprod123]

M4 MacBook Air

Apple introduces the new MacBook Air with the M4 chip and a sky blue color

Apple announced the new MacBook Air, featuring the M4 chip, up to 18 hours of battery life, a 12MP Center Stage camera, and a lower starting price.

www.apple.com

 

[okoroezenwa]

Damn, it was actually an M3 Ultra. That’s so bizarre.

 

[mvprod123]

M4 MacBook Air

Apple introduces the new MacBook Air with the M4 chip and a sky blue color

Apple announced the new MacBook Air, featuring the M4 chip, up to 18 hours of battery life, a 12MP Center Stage camera, and a lower starting price.

www.apple.com

Starting from $999. Great value.

 

[okoroezenwa]

Starting from $999. Great value.

Yeah it’s good to see the latest Air get down to 999 again. It’s been a while. Also the Studio remaining at 1999/3999 is good too, and highlights how dumb the 1399 M4 Pro’s pricing is.

 

[mvprod123]

It seems that M3 Ultra and M4 Ultra (?) can indeed be considered separate chips. M3 Max doesn't have TB5, but M3 Ultra does. I think the M3 Ultra is two M3 Max dies with TB5, but made on the N3E process. This explains the lack of UltraFusion in the M3 Max chip for MacBook Pro. Does this mean that we will get M4 Ultra on the N3P process?

 

[jdubs03]

It seems that M3 Ultra and M4 Ultra (?) can indeed be considered separate chips. M3 Max doesn't have TB5, but M3 Ultra does. I think the M3 Ultra is two M3 Max dies with TB5, but made on the N3E process. This explains the lack of UltraFusion in the M3 Max chip for MacBook Pro. Does this mean that we will get M4 Ultra on the N3P process?

If they’re going to bifurcate their CPU line up like this going forward then yup.

 

[Eug]

It seems that M3 Ultra and M4 Ultra (?) can indeed be considered separate chips. M3 Max doesn't have TB5, but M3 Ultra does. I think the M3 Ultra is two M3 Max dies with TB5, but made on the N3E process. This explains the lack of UltraFusion in the M3 Max chip for MacBook Pro. Does this mean that we will get M4 Ultra on the N3P process?

M3 Ultra
The most powerful Apple chip ever made, M3 Ultra fuses two chips to deliver twice the CPU, GPU, Neural Engine, and memory bandwidth. Experience breakthrough performance for the most extreme workflows — from crunching massive datasets to simultaneously animating and rendering 3D scenes. And with up to half a terabyte of unified memory, you can run LLMs entirely in memory.

 

[Eug]

Apple says the Ultra will skip certain generations.

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