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

 

[Meteor Late]

Apple chips aren't binned for clocks - so worst silicon limits clocks. With such binning x86 cpus wouldn't have 5.7GHz boost clocks but something closer 5GHz. Just binning silicon for clock would give Apple that 200-300 MHz more with same 10W single core consumption. Giving those best binned parts 20-25W single core power would increase clocks much more.

But it's not just binning, x86 CPU consume much more in single core on Desktop compared to Laptop, they are allowed much more power, it's deliberate, of course there is also binning but it's not all of the story.

 

[mvprod123]

One of the Chinese reviews measured the maximum memory bandwidth that the CPU and GPU can take separately.
The upper part of the table shows the single-core memory bandwidth, while the lower part shows the multi-core bandwidth. Switching to LPDDR5X proved to be very beneficial.



M4 Max GPU can receive a maximum of 408 GB/sec, M2 Ultra - 680 GB/sec

 

[poke01]

very impressed with M4 series

 

[mvprod123]

One of the Chinese reviews measured the maximum memory bandwidth that the CPU and GPU can take separately.
The upper part of the table shows the single-core memory bandwidth, while the lower part shows the multi-core bandwidth. Switching to LPDDR5X proved to be very beneficial.

View attachment 111187

M4 Max GPU can receive a maximum of 408 GB/sec, M2 Ultra - 680 GB/sec

View attachment 111188

View attachment 111189

In this reviewer's Blender test, we saw a 2x improvement in the M4 Max but a minor one for the regular M4. Memory bandwidth matters.

 

[igor_kavinski]

Blender
View attachment 111185

So kind of like a 4070 but with almost 128GB VRAM. That's the real killer feature until Strix Halo comes out but I'm thinking max configuration will have 96GB only.

 

[poke01]

So kind of like a 4070 but with almost 128GB VRAM. That's the real killer feature until Strix Halo comes out but I'm thinking max configuration will have 96GB only.

Problem is in Blender it won’t reach desktop 4070 levels because RDNA3.5 is bad at Ray tracing.

Strix Halo will have 128GB SKU but you can only assign 96 video memory just like on a 128GB Mac.

 

[mvprod123]

So kind of like a 4070 but with almost 128GB VRAM. That's the real killer feature until Strix Halo comes out but I'm thinking max configuration will have 96GB only.

Strix Halo will have worse memory bandwidth

 

[igor_kavinski]

Problem is in Blender it won’t reach desktop 4070 levels because RDNA3.5 is bad at Ray tracing.

Let's hope AMD's software geniuses come up with a solution for that

 

[Eug]

The smaller size didn't make the M4 louder than the M2. In fact, perhaps the opposite.

 

[The Hardcard]

View attachment 111190

In this reviewer's Blender test, we saw a 2x improvement in the M4 Max but a minor one for the regular M4. Memory bandwidth matters.

Are those numbers related to the Open Data website? The M4 Max numbers sum up close to the 40 core on the website. But the M3 Max numbers sum up close to the 30 core average on the website. Was the reviewer explicit about which M3 Max he had?

The website also shows the M2 Ultra average about equal to the 30 core M3 Max average, which would align with this chart.

The website only has one M4 Max 40 core, but it is 20 percent higher than the 40 core M3 Max average.

 

[The Hardcard]

Problem is in Blender it won’t reach desktop 4070 levels because RDNA3.5 is bad at Ray tracing.

Strix Halo will have 128GB SKU but you can only assign 96 video memory just like on a 128GB Mac.

You can asign all 128 GB to the GPU ( - 4 to 8 GB to protect OS stability) on Macs. 75 percent is just the OS default limit. There is no hardware limit.

 

[Doug S]

Apple doesn't bin based on frequency. If they did, I think there might be M4 chips that hit 4.7-4.8 GHz.

I think there's no doubt M4 could easily exceed 5 GHz if Apple binned for frequency. The problem is that Apple hates having a proliferation of SKUs, plus there is the inherent unpredictability of yields in different processes. You'd get yourself sort of stuck if you had say M2 yielding well enough to have a halo model at 18% above nominal but M3 yielding only 6% above nominal and then the halo M2 would beat the halo M3! One might suggest Intel is having this exact issue with ARL right now - they are not only using a problematic process but one targeted towards low power which may not work quite as well as Intel 7 for the 250W that modern desktop CPUs can draw.

Over on RWT we've recently been discussing the possibility that Apple Silicon is binned on power. It would make sense, for example, to select M4/M4P dies able to operate on less power for Macbook Pro and use ones that require more in Mac Mini since it isn't running on battery and has (or at least should have, I guess we haven't seen the inside of the new one yet) better cooling.

I have no idea if they do this, but it would be interesting if someone had access to a Mac Mini and Macbook Pro with the same model of M4 and compare power usage when running the same tasks.

 

[Glo.]

You can asign all 128 GB to the GPU ( - 4 to 8 GB to protect OS stability) on Macs. 75 percent is just the OS default limit. There is no hardware limit.

Great. Apple invented how to exceed the physical memory, and download more RAM from the Internet!

 

[mvprod123]

 

[mvprod123]

 

[Eug]

View attachment 111207

M4 Pro uses slightly more power than M3 Max here. They're both 14-core variants.

 

[poke01]

M4 Pro uses slightly more power than M3 Max here. They're both 14-core variants.

it delivers significantly more performance for that extra 1 watt tho. Like it makes the M3 Max looks it’s 2 generations old

 

[mvprod123]

M4 Pro uses slightly more power than M3 Max here. They're both 14-core variants.

Unbinned M3 Max uses 51 Watts. M4 Pro delivers the same performance as unbinned M3 Max using 14% less power.

 

[Eug]

Unbinned M3 Max uses 51 Watts. M4 Pro delivers the same performance as unbinned M3 Max using 14% less power.

View attachment 111210

I wonder how hot M4 Max 14-core is, and how hot unbinned M4 Max 16-core is too. Will it be not too far off that 51 Watt peak for M3 Max?

 

[mvprod123]

I wonder how hot M4 Max 14-core is, and how hot unbinned M4 Max 16-core is too. Will it be not too far off that 51 Watt peak for M3 Max?

View attachment 111207

The unbinned 16-core M4 Max consumes 57 watts. +11% power delivers 20-22% better perf.

 

[Eug]

I wonder at what point will the MacBook Pro fans will become a nuisance again. However, it will take a lot more power consumption to get to the crazy noise of the Intel MacBook Pros.

 

[Glo.]

We are still discussing chips that consume LESS than even Core i3's of yesteryear.

Insane efficiency, and as a person who has M4 I can vouch for all of its performance.

M4 iPad Pro is better for trading than my 12100F desktop machine, while using less than 10% of the power that desktop uses.

 

[poke01]

Mac mini M4 Pro, análisis: un titán de rendimiento en un tamaño casi imposible

Análisis a fondo del Mac mini M4 Pro: batería, pantalla, procesador, diseño y acabado, software y relación calidad precio. Todos los resultados de las pruebas...

www.applesfera.com

 

[FlameTail]

Strix Halo will have worse memory bandwidth

M4 Max vs Strix Halo
546 GB/s vs 273 GB/s

STH has basically half the bandwidth of M4 Max.

 

[FlameTail]

Over on RWT we've recently been discussing the possibility that Apple Silicon is binned on power. It would make sense, for example, to select M4/M4P dies able to operate on less power for Macbook Pro and use ones that require more in Mac Mini since it isn't running on battery and has (or at least should have, I guess we haven't seen the inside of the new one yet) better cooling.

I have no idea if they do this, but it would be interesting if someone had access to a Mac Mini and Macbook Pro with the same model of M4 and compare power usage when running the same tasks

That's a very interesting theory. They could be doing it to iPhones too. Put the more efficient ones in the smaller iPhones with smaller batteries.