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

 

[igor_kavinski]

But once turbo boost is off, I never see it reach 80C and the 1650 ti is locked at max boost.

Excellent! Learned something new

 

[Heartbreaker]

I know. When under load the M2 can use up to 30watts. All I am saying is the Air could use better cooling say a vapour chamber for a fanless design.

The problem is ultimately the fanless design. A metal case with no fins is not great for passive heat dissipation, especially when it's potentially sitting on your lap.

 

[moinmoin]

If you want better sustained performance, buy something designed for that.

I think we have different expectations of "sustained performance" there. What I expect out of the box is performance that can be reliably sustained, not some sporadic hell heat in a box that's immediately followed by a full on brake throttling. That that abomination of a mis-behavior is followed by all kinds of manufacturers is sad enough already. That Apple apparently didn't notice that much of the positive public feedback around M1 was exactly about avoiding the above is simply unfathomable to me.

 

[gdansk]

I think we have different expectations of "sustained performance" there. What I expect out of the box is performance that can be reliably sustained, not some sporadic hell heat in a box that's immediately followed by a full on brake throttling. That that abomination of a mis-behavior is followed by all kinds of manufacturers is sad enough already. That Apple apparently didn't notice that much of the positive public feedback around M1 was exactly about avoiding the above is simply unfathomable to me.

MBA M1 throttled about 20% too.

It's not new and is the consequence of having a fanless design. MBA M2 has to throttle more, relatively, because it uses more power at peak.
Throttling would be a deal breaker in a MBP but it was fine for the last generation MBA. I don't see why that would change. You've bought the wrong device if you expect sustained peak performance from a fanless laptop.

And conveniently ignored is that even at this throttled level it's faster than most ADL chips with a 28W PL1.

 

[igor_kavinski]

2022 MacBook Air review: Apple’s clean slate | Ars Technica

The good

• Sleek, slim design
• Excellent build quality
• Out-of-this-world performance for its size and weight
• Great keyboard and trackpad
• Strong battery life

The bad

• Throttles and warms under heavy sustained load compared to Apple’s various bigger, more expensive alternatives
• Port situation could still be better
• Expensive, once you factor in attractive configuration options

The ugly

• Like M1-based laptops, it can only drive one external display

Throttles and expensive. Confirmed.

 

[Eug]

MBKHD states that for any sustained load under 5 minutes, there isn't really any noticeable throttling for the fanless M2 MacBook Air. However, it can get quite warm on the lap if you're pegging the CPU for several minutes. Throttling becomes more significant after about that 5 minute mark.

This is borne out by Dave2D's throttle testing, comparing sustained high loads on the M2 MacBook Pro vs M2 MacBook Air.





So, it appears to be roughly a 15% to 25% performance loss after 20 loops of these stress tests. (I don't know how long the loops take though.)

This is the M1 vs M2 vs M1 Max vs M1 Ultra for an export of a very complex video, done by Brian Tong.



The difference in time here between the MBA and MBP is 13%.

So basically, if you want to do hardcore video stuff regularly, you should probably get a 14" (or 16") MacBook Pro. However, if you are only dabbling in video, getting the M2 MacBook Pro doesn't really offer that much performance improvement over M2 MacBook Air, and misses out on all the design improvements of the M2 MacBook Air.

Again, I still don't really see the point of the M2 MacBook Pro for most users, unless you absolutely love the Touch Bar.

 

[Doug S]

If Apple operated like PC OEMs they'd sell the Macbook Air with an M2 that was clocked lower to reflect it being a lower end product sold at a lower price. If they did people would be whining "it is the same M2 that goes in the Macbook Pro, why don't they let it run at full speed for as long as it can instead of artificially limiting it? Apple is deliberately making people spend more money on the higher priced product to get the full performance of the M2!"

Apple haters gonna hate.

 

[moinmoin]

And conveniently ignored is that even at this throttled level it's faster than most ADL chips with a 28W PL1.

Not sure why you bring in ADL there. Intel hasn't done any true mobile chip in generations anymore, and ADL is the worst yet.

 

[poke01]

Not sure why you bring in ADL there. Intel hasn't done any true mobile chip in generations anymore, and ADL is the worst yet.

What do mean a 9 watts ADL chips is not mobile enough for a fanless design for you?

Oh wait it's not going to be good experience cause single core would suck cause Intel relies on boot clock so much.

 

[poke01]

2022 MacBook Air review: Apple’s clean slate | Ars Technica



Throttles and expensive. Confirmed.

Even the M1 Air throttled. Its fanless for fudges sake. FANLESS. Your phone throttles, your tablet throttles cause they are fanless but do you complain about those...

Why you expecting a fanless laptop to NOT throttle?

 

[poke01]

I think we have different expectations of "sustained performance" there. What I expect out of the box is performance that can be reliably sustained, not some sporadic hell heat in a box that's immediately followed by a full on brake throttling. That that abomination of a mis-behavior is followed by all kinds of manufacturers is sad enough already. That Apple apparently didn't notice that much of the positive public feedback around M1 was exactly about avoiding the above is simply unfathomable to me.

The M1 Air also throttled and what do you expect from a fanless design??

Perfect sustained worked on a thin 11mm laptop that is fanless is impossible. I would like to see a 10-20Watt x86 laptop that's fanless do the same without throttling. x86 would not even survive in that scenario.

The 108C reading is junction temperature and this is confirmed by an ex-nuvia employee. You want sustained work loads get the 14" or 16".

Mind you all laptops throttle, even ARM and low end x86 and high end windows laptops with fans.

Fanless designs are good for certain laptops like the Air. You don't need clean the insides for dust. Less maintenance.

 

[gdansk]

Not sure why you bring in ADL there. Intel hasn't done any true mobile chip in generations anymore, and ADL is the worst yet.

Sorry, I was comparing the MBA M2 to its predecessors (with M1) and to hypothetical competitors (with ADL). Its predecessor throttled. Its competitor throttles even in actively cooled design.

 

[moinmoin]

The M1 Air also throttled and what do you expect from a fanless design??

Perfect sustained worked on a thin 11mm laptop that is fanless is impossible.

I prefer a fanless design, and I prefer actually sustained performance. (I got an M1 MBA for iOS app work before M2 was revealed and would pick it again now as well.) Sustained performance is possible if the CPU/GPU are not allowed to boost past the point that can be sustained without throttling. But great, everybody wants useless short term peak performance even in mobile form factors now, all companies fall in line and start caring more about peak performance than efficiency, and everybody starts spreading myths that any other approach would be impossible. I honestly hoped Apple would be more immune to this.

 

[poke01]

I prefer a fanless design, and I prefer actually sustained performance. (I got an M1 MBA for iOS app work before M2 was revealed and would pick it again now as well.) Sustained performance is possible if the CPU/GPU are not allowed to boost past the point that can be sustained without throttling. But great, everybody wants useless short term peak performance even in mobile form factors now, all companies fall in line and start caring more about peak performance than efficiency, and everybody starts spreading myths that any other approach would be impossible. I honestly hoped Apple would be more immune to this.

Why would pick the M1 Air if does not have sustained perf just like M2 Air?

No one has done that because it's impossible. No chip can ever not thottle. That's why companies offer low power modes in OS. To limit the chip yourself.

 

[moinmoin]

No one has done that because it's impossible. No chip can ever not thottle.

How young are you?

 

[gdansk]

Why would pick the M1 Air if does not have sustained perf just like M2 Air?

No one has done that because it's impossible. No chip can ever not thottle. That's why companies offer low power modes in OS. To limit the chip yourself.

There are plenty of older chips that don't throttle. But they have worse responsiveness. Average workloads are better on CPUs which have a higher clock rate for a while. Most users workloads fall into this window. But unfortunately M1/M2 in fanless designs have to reduce power to match dissipation at some point.

Oddly, Apple's behavior hasn't significantly changed from M1 to M2. This remains much ado about nearly nothing.

 

[Eug]

There was an old video of a VIA x86 chip running even without a heatsink, but the performance of VIA chips in general was terrible of course.

Mind you, I'd be perfectly happy performance-wise with an Apple A16 MacBook, as long as it had 16 GB RAM and sufficient I/O.

 

[Markfw]

How young are you?

Exactly ! I will give an example. Take a 5950x. Put a heatsink rated at at least 150 watt. Stock BIOS. when the chip gets to its power threshhold , it downgrades the speed to maintain the power limits, but will NOT even throttle a little, EVER. It is designed so that if the proper heatsink is used, the heat will be fine under any load the chip is rated for as far a power.

I am pretty sure this also is what Intel chips do, and chip in the last few years. I am not as well versed on them though.

 

[repoman27]

Exactly ! I will give an example. Take a 5950x. Put a heatsink rated at at least 150 watt. Stock BIOS. when the chip gets to its power threshhold , it downgrades the speed to maintain the power limits, but will NOT even throttle a little, EVER. It is designed so that if the proper heatsink is used, the heat will be fine under any load the chip is rated for as far a power.

I am pretty sure this also is what Intel chips do, and chip in the last few years. I am not as well versed on them though.

"It downgrades the speed, but will NOT even throttle a little, EVER."

So yeah, it throttles but then it doesn't throttle ever after that? Did you even think about what you just wrote there?

Modern SoCs are dynamic. Very few people utilize their personal computing devices continuously at their thermal dissipation limits. If you don't allow higher clocks when you're below a reasonable thermal threshold, then you'll always be leaving performance on the table. 80% of PCs sold in recent years and 100% of phones and tablets are portable, battery powered devices. Providing them with a cooling solution that would allow the CPU and GPU to sustain maximum clock speeds indefinitely is a non-starter. Throttling is by design, it's not at bug. It's by far the best engineering tradeoff to maximize both portability and performance at the expense of some performance consistency.

The M2 offers the highest performance per watt of any PC microarchitecture on the market today. You can't take a less efficient SoC and put it in the same chassis and expect it to perform better or throttle less.

 

[eek2121]

"It downgrades the speed, but will NOT even throttle a little, EVER."

So yeah, it throttles but then it doesn't throttle ever after that? Did you even think about what you just wrote there?

Modern SoCs are dynamic. Very few people utilize their personal computing devices continuously at their thermal dissipation limits. If you don't allow higher clocks when you're below a reasonable thermal threshold, then you'll always be leaving performance on the table. 80% of PCs sold in recent years and 100% of phones and tablets are portable, battery powered devices. Providing them with a cooling solution that would allow the CPU and GPU to sustain maximum clock speeds indefinitely is a non-starter. Throttling is by design, it's not at bug. It's by far the best engineering tradeoff to maximize both portability and performance at the expense of some performance consistency.

The M2 offers the highest performance per watt of any PC microarchitecture on the market today. You can't take a less efficient SoC and put it in the same chassis and expect it to perform better or throttle less.

AMD and Intel chips have what is called a base clock. That is the clock you should be looking at. Boost clocks just mean 'hey, I am capable of boosting much higher if there is extra headroom'.

The base clock of a 5950x is 3.4 ghz. The chip will always operate at that speed unless you do a poor job at cooling it. Note that a $20 air cooler can cool the chip.

So, I guess it depends on how you want to define throttling. No, it will not go below base clocks if provided the minimum rated CPU cooler that can dissipate 115W of heat. Yes, if you cheap out on the cooler, you'll see it drop below the base clocks.

The reason the MBA throttles is because Apple had two goals:

1. It must be thin and silent
2. It must be cheaper than the 13" MBP

Working with those two goals, they chose to differentiate the air from the pro by allowing the air to throttle (among other differences).

After all, if the air were every bit of awesome as the pro is, maybe minus some ports, why would you even buy the pro?

 

[Doug S]

I prefer a fanless design, and I prefer actually sustained performance.

Those two preferences are in conflict. I prefer to fly in first class comfort, and also prefer my plane tickets to cost less than $500.

 

[Markfw]

"It downgrades the speed, but will NOT even throttle a little, EVER."

So yeah, it throttles but then it doesn't throttle ever after that? Did you even think about what you just wrote there?

Modern SoCs are dynamic. Very few people utilize their personal computing devices continuously at their thermal dissipation limits. If you don't allow higher clocks when you're below a reasonable thermal threshold, then you'll always be leaving performance on the table. 80% of PCs sold in recent years and 100% of phones and tablets are portable, battery powered devices. Providing them with a cooling solution that would allow the CPU and GPU to sustain maximum clock speeds indefinitely is a non-starter. Throttling is by design, it's not at bug. It's by far the best engineering tradeoff to maximize both portability and performance at the expense of some performance consistency.

The M2 offers the highest performance per watt of any PC microarchitecture on the market today. You can't take a less efficient SoC and put it in the same chassis and expect it to perform better or throttle less.

Do you understand the difference between throttling and internally downgrading the clock ? Throttling is what the Old chips did when they got too hot. Internally downgrading the clock is a calculated action by the CPU to stay in spec on power limits, NOT HEAT. Now if you are too stupid to put the correct heatsink on there, then it may throttle, but I qualified my answer, saying with a heatsink rated at 150 watt or more. His statement was that a CPU would never NOT throttle, and thats totally wrong.

 

[poke01]

The M1 and M2 Air will thottle LESS when when low power mode is on. Its clocks the P cores to 2Ggz

How young are you?

In laptops with fanless cooling it's impossible. You can push a SoC to the max it will get hot with fanless, no matter how efficient or how low power it is. There is no laptop fanless laptop in existence that can get hot and not slow down and throttle

I knows desktops with proper cooling can be throttling can be completely avoided.

 

[poke01]

Exactly ! I will give an example. Take a 5950x. Put a heatsink rated at at least 150 watt. Stock BIOS. when the chip gets to its power threshhold , it downgrades the speed to maintain the power limits, but will NOT even throttle a little, EVER. It is designed so that if the proper heatsink is used, the heat will be fine under any load the chip is rated for as far a power.

I am pretty sure this also is what Intel chips do, and chip in the last few years. I am not as well versed on them though.

This is not the case for laptops. Because laptops such as ultrabooks are thin and have very little space for cooling.

Mac Desktops like the Mac Pro and Mac Studio and your 3950x do NOT throttle as they have large coolers and maintain the chips speed.

 

[poke01]

Throttling is what the Old chips did when they got too hot.

wrong people wrong.
Modern mobile chips do that still from Intel to AMD to Apple to Qualcomm...

https://twitter.com/x/status/1548771664430858241

The ex-nuvia in twitter thread also agrees with dylan