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

 

[okoroezenwa]

https://twitter.com/x/status/1883276535643455790

View attachment 115664


when M4 ultra?

I want to be optimistic and say March but it could just be WWDC.

 

[mvprod123]

 

[SpudLobby]

How much will Apple going 2.5D impact power/latency? Namely, the former. It should be much lower than with something like Meteor Lake, right? Guess it depends on the materials and whatever substrate TSMC does. There will certainly be some tradeoff but how obvious/large it is is the question.

 

[Curious_Inquirer]

How much will Apple going 2.5D impact power/latency? Namely, the former. It should be much lower than with something like Meteor Lake, right? Guess it depends on the materials and whatever substrate TSMC does. There will certainly be some tradeoff but how obvious/large it is is the question.

TSMC says that chiplet and later on 3D chip stacking is coming. Given that they make nearly all the mobile chips and is in constant communication and collaboration with their customers, I believe they have future sight on what is coming.

https://twitter.com/x/status/1865811156583231995

Edit: X is borked. Whenever I link a tweet in a thread it always puts the previous tweet of the linked tweet thread on top. Just focus on the mobile tweet and disregard the WiFi tweet.

 

[The Hardcard]

How much will Apple going 2.5D impact power/latency? Namely, the former. It should be much lower than with something like Meteor Lake, right? Guess it depends on the materials and whatever substrate TSMC does. There will certainly be some tradeoff but how obvious/large it is is the question.

I don’t have an official source, but in another forum, one poster says he is a former CPU architect from AMD. He made a post that included the claim that latency increased 6 picoseconds per millimeter.

I have been wanting to see documentation on that, but if that is true, memory on package is not a latency play. You would have to put the memory more than 3 feet away to see a noticeable difference. There would surely be a huge difference in power needed.

 

[DavidC1]

I don’t have an official source, but in another forum, one poster says he is a former CPU architect from AMD. He made a post that included the claim that latency increased 6 picoseconds per millimeter.

You can use on package stuff to decrease latency and power, but it's only doable if you overhaul the system to take advantage of it. On package stuff for the vast majority is cost/area play.

6 ps/mm assumes the speed of transfer at 0.7-0.8x the speed of light which is the speed in a wire.

Just slapping memory on package does nothing. It doesn't even save power in this case. Saving power needs optimization too. Intel had on package memory stuff prior to Lunarlake(as far back as Haswell), and mobile vendors had Package-on-Package which means the stuff is on top of each other, so physically even closer. None of them claims much lower latency. Intel Atoms since the 32nm generation had memory controller on-die, but had crappy latency. It wasn't until Silvermont when they actually optimized for low latency.

Putting in on package gives you a potential to optimize for lower latency and/or lower power. They key word is potential.

 

[poke01]

Mac mini time. It’s a great little machine been doing some tests and loving it so far.

Got the base model and I’m using a 500GB external drive as the boot drive

 

[poke01]

Everybody's favourite benchmark
heres CPU-Z x86-64 running on M4 via crossover:


7700X:





Well, the M4 has faster single core than a 2022 CPU while running via Crossover. Very impressed and its not that far from a 285K.

 

[poke01]

I'll be posting ARM Linux benchmarks that use TSO and Rosseta2 enabled and oh nested virt as well soon. M4 was the right time to get an ARM mac, lots of fun stuff

 

[poke01]

Apple Inc. Apple Virtualization Generic Platform - Geekbench

Benchmark results for an Apple Inc. Apple Virtualization Generic Platform with a VirtualApple processor.

browser.geekbench.com

this score is via Linux ARM VM , running an x86 Geekbench test. It on par with a Pixel 7

 

[igor_kavinski]

heres CPU-Z x86-64 running on M4 via crossover:
View attachment 116127

Very nice! Appreciate if you make a separate thread about all the interesting things you try on your M4, especially related to x86 emulation.

Better also for people in future otherwise these experiments will be lost in this mega thread.

 

[poke01]

Very nice! Appreciate if you make a separate thread about all the interesting things you try on your M4, especially related to x86 emulation.

Better also for people in future otherwise these experiments will be lost in this mega thread.

I'll do it at some point.

-------

Also whatever Apple did with these new cores for 3D rendering is amazing. Thanks to Igor, I benched the M4 in V-Ray 6 and its a big improvement over M3. ~30% faster than the M3 scores listed on Vray site. But the biggest suprise to me is the CPU power consumption, it was under 19 watts for the CPU cluster and the P cores and E cores were being maxed out.

This is amazing perf/w. ( I also measured with a wall meter and the total power consumption of the Mac listed in the terminal instance is correct.)

 

[Eug]

I'll do it at some point.

-------

Also whatever Apple did with these new cores for 3D rendering is amazing. Thanks to Igor, I benched the M4 in V-Ray 6 and its a big improvement over M3. ~30% faster than the M3 scores listed on Vray site. But the biggest suprise to me is the CPU power consumption, it was under 19 watts for the CPU cluster and the P cores and E cores were being maxed out.

This is amazing perf/w. ( I also measured with a wall meter and the total power consumption of the Mac listed in the terminal instance is correct.)
View attachment 116265

One of the reasons I bought the M4 Mac mini was because of the GPU upgrades, not 3D performance-wise, but for 2D screen resolution support. It is now HDMI 2.1 with 8K 60 Hz and 4K 240 Hz support, a big upgrade from previous non-Pro models.

However, Apple seems to be artificially limiting its non-integer scaled resolution support options for monitors. For example, here are the 5K Studio Display resolution options with the M4 non-Pro:



OTOH, those with any of the Pro/Max models also have 3200x1800 as an option. Here is the M2 Max Mac Studio:



Even my ancient 2017 27" Core i5 iMac has the 3200x1800 option:



IOW, Apple is limiting the M4's resolution options to the same ones as M1/M2/M3 despite the very different GPUs. Is there another reason for this limitation besides marketing?

 

[igor_kavinski]

IOW, Apple is limiting the M4's resolution options to the same ones as M1/M2/M3 despite the very different GPUs. Is there another reason for this limitation besides marketing?

If a 2017 iMac can do it, there is no excuse other than Apple wanting moar monehhhhh.

 

[digitaldreamer]

One of the reasons I bought the M4 Mac mini was because of the GPU upgrades, not 3D performance-wise, but for 2D screen resolution support. It is now HDMI 2.1 with 8K 60 Hz and 4K 240 Hz support, a big upgrade from previous non-Pro models.

However, Apple seems to be artificially limiting its non-integer scaled resolution support options for monitors. For example, here are the 5K Studio Display resolution options with the M4 non-Pro:

View attachment 116269

OTOH, those with any of the Pro/Max models also have 3200x1800 as an option. Here is the M2 Max Mac Studio:

View attachment 116270

Even my ancient 2017 27" Core i5 iMac has the 3200x1800 option:

View attachment 116271

IOW, Apple is limiting the M4's resolution options to the same ones as M1/M2/M3 despite the very different GPUs. Is there another reason for this limitation besides marketing?

What's with the two sets of equal resolutions with one marked as "low resolution?" Is a refresh frequency difference?

 

[Eug]

What's with the two sets of equal resolutions with one marked as "low resolution?" Is a refresh frequency difference?

Low resolution is non-Retina, so you don't want to use that. IOW, if you use the low resolution 2560x1440, it's functioning as if the screen is natively 2560x1440. Essentially every 2x2 pixel cluster is treated as a single pixel with no further processing.

OTOH, if you use the regular 2560x1440, the resolution is actually 5120x2880 with 2X scaled UI assets, to mimic a 2560x1440 screen but with much greater detail - Retina.

Why would you want to use low resolution mode? I guess maybe for faster fps in games, or if you absolutely need to mimic the pixel structure of a specific resolution for whatever reason. As for higher refresh rates, that would make sense but I cannot test that as this monitor is 60 Hz.

HiDPI:



Low resolution:



I'm using a 28.2" 3840x2560 (3:2) monitor, non-integer scaled to 2304x1536.

 

[Thibsie]

Low resolution for everything 3D games included, indeed.

 

[name99]

One of the reasons I bought the M4 Mac mini was because of the GPU upgrades, not 3D performance-wise, but for 2D screen resolution support. It is now HDMI 2.1 with 8K 60 Hz and 4K 240 Hz support, a big upgrade from previous non-Pro models.

However, Apple seems to be artificially limiting its non-integer scaled resolution support options for monitors. For example, here are the 5K Studio Display resolution options with the M4 non-Pro:

View attachment 116269

OTOH, those with any of the Pro/Max models also have 3200x1800 as an option. Here is the M2 Max Mac Studio:

View attachment 116270

Even my ancient 2017 27" Core i5 iMac has the 3200x1800 option:

View attachment 116271

IOW, Apple is limiting the M4's resolution options to the same ones as M1/M2/M3 despite the very different GPUs. Is there another reason for this limitation besides marketing?

Will Better Display give you what you want?

GitHub - waydabber/BetterDisplay: Unlock your displays on your Mac! Flexible HiDPI scaling, XDR/HDR extra brightness, virtual screens, DDC control, extra dimming, PIP/streaming, EDID override and lots more!

Unlock your displays on your Mac! Flexible HiDPI scaling, XDR/HDR extra brightness, virtual screens, DDC control, extra dimming, PIP/streaming, EDID override and lots more! - waydabber/BetterDisplay

github.com

M4 Mac Mini @ Sequoia - HiDPI Resolution missing · waydabber BetterDisplay · Discussion #3591

Hi @waydabber, I have the LG 38GN950 (3840x1600) screen. I usually use this with the resolution 3200x1333 @ HiDPI mode with my macbook pro (M2 Pro @ Sonoma). Since today I also have a mac mini (m4 ...

github.com

You apparently can't get EXACTLY what you seem to want, but you can get close; with the constraint (and the difference relative to Max/Pro) apparently being in the amount of storage available to the pool of attached monitors. There may be(or may not) be an issue with how this is allocated that may (or may not) be fixable/fixed in new, or even current latest, firmware.

You can complain about Apple's *UI* choices forever.
Or you can just solve the problem as well as possible.

 

[poke01]

IOW, Apple is limiting the M4's resolution options to the same ones as M1/M2/M3 despite the very different GPUs. Is there another reason for this limitation besides marketing?

the cores are much better than M1/M2 but Apple locked the extra resolutions thru software but it’s an easy fix, just get BetterDisplay like name99 suggested, it’s what I use too.

 

[poke01]

Also using USB-C instead of HDMI offers higher limits

 

[Eug]

Will Better Display give you what you want?

You apparently can't get EXACTLY what you seem to want, but you can get close; with the constraint (and the difference relative to Max/Pro) apparently being in the amount of storage available to the pool of attached monitors. There may be(or may not) be an issue with how this is allocated that may (or may not) be fixable/fixed in new, or even current latest, firmware.

Actually 3200x1800 is not something I’d use on a 27” 5K display. It was a friend of mine who sold his Mac Studio and got the M4 Mac mini. I currently use a 28.2” 4K+ 3840x2560 (3:2) display which has same 164 ppi pixel density as 27” 4K. I run 2304x1296 on that one because I prefer larger default text sizing.

I am more interested in the coming 31.5” 5K and 6K displays this year. Regardless if it’s 5K or 6K, on a 31.5” screen, I’d probably run something like 2880x1620. (I might prefer something like 2720x1536 but I haven’t seen that option offered anywhere.) It seems macOS includes this option for 5K displays so that’s fine. However, AFAIK, neither macOS nor BetterDisplay allows anything in between 2560x1440 and 3008x1692 on 6K displays for some reason. We shall see once the new monitors are out though.

Acer will be releasing a 31.5” 5K 144 Hz Predator gaming monitor.
LG will be releasing a 32” 6K 60 Hz monitor with IPS Black and Thunderbolt 5 for creative professionals.
Asus will also be releasing a 32” 6K ProArt 60 Hz monitor also for creatives. It won't have IPS black or Thunderbolt but apparently it will retail for only US$1199.

Also using USB-C instead of HDMI offers higher limits

On the M4 Mac mini, it actually supports HDMI 2.1 up to 8K 60. It supports HDMI 2.1 and DisplayPort 1.4. The M4 Pro Mac mini supports HDMI 2.1 and DisplayPort 2.1.