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

 

[Nothingness]

No, I didn’t say it was impossible, I think it’s unlikely based on the notable recent trends, talent loss.

They indeed lost some talented people, but I guess they also attracted some new ones (they have attractive salaries and retention plans). Of course they likely lost time, and it remains to be seen how long it will take them to recover from the loss.

 

[FlameTail]

512 GB of Unified Memory in the next Mac Studio (M4 Ultra) is going to be dream for consumers who want to run AI models.

 

[SiliconFly]

...Apple could make a comeback in the next two years...

Should they? Afaik, they've already made the come back. Apple Silicon is the reference today (esp. the M series).

 

[FlameTail]

Should they? Afaik, they've already made the come back. Apple Silicon is the reference today (esp. the M series).

I am talking about the IPC of their P-cores.

Look how PPC gains slowed to a crawl in recent years.

 

[SiliconFly]

I am talking about the IPC of their P-cores.
View attachment 97202
Look how PPC gains slowed to a crawl in recent years.

Lower than expected node PPA gains might also be a significant factor.

 

[FlameTail]

Do the binned Apple M chips have the same cache capacity as the unbanned M chips?

 

[FlameTail]

Do you guys think the Apple M5 will exceed 200 mm²?

M1 = N5. = 118 mm²
M2 = N5P = 151 mm²
M3 = N3B. = 146 mm²
M4 = N3E. (?) = 165-175 mm² (?)
M5 = N3P (?) = 200-210 mm² (?)

If it is on N3P and not N2, I think it certainly will.

 

[smalM]

It depends on what Apple considers a sufficient GPU for the base M4/M5.
If the answer is "buy a M4 Pro/M5 Pro" then...

 

[FlameTail]

Also the CPU. From M1->M3 generation, the CPU has stuck to a 4P+4E configuration.

I have a feeling that M4 or M5, will upgrade that to 6P+4E or 4P+6E.

 

[Doug S]

Also the CPU. From M1->M3 generation, the CPU has stuck to a 4P+4E configuration.

I have a feeling that M4 or M5, will upgrade that to 6P+4E or 4P+6E.

Why? Base models are base models. Apple doesn't offer hundreds of SKUs like Intel/AMD, so they want to maintain reasonable separation between each step in their lineup. It is more likely Apple will devote any additional area to a larger NPU or GPU.

There is already plenty of CPU power in the base Apple Silicon CPUs, the only reason to have more is for pointless pissing contests like Cinebench. If they bump that up too not only do they make their chip larger/more expensive, they have to bump up Pro and by extension Max to maintain useful separation between the product families.

 

[smalM]

Maybe we will see more P cores in the base modell when Apple changes the P cluster size from 6 to 8 P cores on N2.

 

[Doug S]

Maybe we will see more P cores in the base modell when Apple changes the P cluster size from 6 to 8 P cores on N2.

I think the driver of cluster size is more likely to be L2 cache contention. It probably doesn't have much to do with transistor density aside from an ability to make L2 larger if the process allows smaller SRAM cells - though that's a balancing act since a larger cache (and higher clock rate) implies higher latency.

 

[Saylick]

I found this Chinese analysis of the A17 Pro, which is pretty dang impressive. In the video, he analyses the die area and was able to directly measure the density of the parts of the chip via electron microscope.

 

[ashFTW]

I found this Chinese analysis of the A17 Pro, which is pretty dang impressive. In the video, he analyses the die area and was able to directly measure the density of the parts of the chip via electron microscope.

Finflex seems to be already in use

 

[Saylick]

Finflex seems to be already in use

Correct.

 

[FlameTail]

@Doug S

You said N3B doesn't have FinFlex?

 

[Henry swagger]

I found this Chinese analysis of the A17 Pro, which is pretty dang impressive. In the video, he analyses the die area and was able to directly measure the density of the parts of the chip via electron microscope.

Beautiful scientific breakdown.. i.m sure tsmc will make him take down the video. Exposing all they secrets lol

 

[SiliconFly]

Beautiful scientific breakdown.. i.m sure tsmc will make him take down the video. Exposing all they secrets lol

Not necessary for TSMC to take it down. Apple A17 Pro density was already known. This video revises it down by just 5%. No biggie. Also, the estimated (theoretical) peak density of TSMC N3 has already been estimated long back (~200), which tracks well with this. So, nothing to worry.

 

[Henry swagger]

Not necessary for TSMC to take it down. Apple A17 Pro density was already known. This video revises it down by just 5%. No biggie. Also, the estimated (theoretical) peak density of TSMC N3 has already been estimated long back (~200), which tracks well with this. So, nothing to worry.

Still companies want to save face if they lied about density values

 

[SiliconFly]

Still companies want to save face if they lied about density values

No. I don't think they lied. They haven't released N3 node characteristics yet. It's just that the tech websites might have speculated wrong.

 

[FlameTail]

According to that video, A17 Pro is 113 mm²

Which contradicts the 103 mm² figure from this:

 

[FlameTail]

I think the driver of cluster size is more likely to be L2 cache contention. It probably doesn't have much to do with transistor density aside from an ability to make L2 larger if the process allows smaller SRAM cells - though that's a balancing act since a larger cache (and higher clock rate) implies higher latency.

There are 2 ways to increase L2 size:

(1) Increasing the amount of L2 within a cluster, while keeping core count per cluster constant
+ L2 per core, is increased
+ Improves ST performance
+ Improves MT performance
- Increases die size

(2) Increasing the amount of L2, by increasing the core count per cluster, by the same factor
- L2 per core, is constant.
+ Improves ST performance
- MT performance is unchanged/regresses
+ Same die area

 

[FlameTail]

Apple might unveil next generation iPads with M3 chips, on May 7th:

https://twitter.com/x/status/1782773176621002852

 

[Doug S]

@Doug S

You said N3B doesn't have FinFlex?

I am pretty sure TSMC claimed at one point that N3B did not include Finflex, but assuming that guy's analysis is correct it is used on A17P. I was confused though by his tables always showing N3E, not N3B, despite looking at A17P.

So I guess one of two things happened. Either my recall is incorrect and N3B always included FinFlex, or Apple got a slightly modified version of N3B. Apple is TSMC's largest customer and works closely with them on process development, and the official "start of mass production" for N3B was months in advance of when Apple's first wafers would be run. They are slotting BSPDN into N2 six months after its start of mass production, so it is at least possible the same was true with FinFlex and N3B for Apple.

I'll note the same gap between start of mass production and start of mass production for Apple's wafers also exists with N3E, so the possibility exists some elements of N3P could find their way into A18/M4.

 

[FlameTail]

I am pretty sure TSMC claimed at one point that N3B did not include Finflex, but assuming that guy's analysis is correct it is used on A17P. I was confused though by his tables always showing N3E, not N3B, despite looking at A17P.

He said that he was too lazy to change it from N3E to N3B, in the comments section of the video.