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

 

[oak8292]

You guys know what really rubs salt into the wound?

RAM prices were down this year by as much as 30% !

Apple is raking in astronomical profits.

Apple is raking in ‘astronomical’ profits on the high volume but there gross margin actually dipped a bit in the last quarter from last year, from 43% to 42%. You know who else has good margins in computing, Microsoft, Intel and AMD. Microsoft gross margins on Windows has been close to 90%. Intel had gross margins of about 60%. Even AMD currently has gross margins higher than Apple.

RAM prices may be down 30% for the year but is that contract for high volumes and does either the manufacturer or DRAM vendor have any margin. In other words is it a sustainable business model. Moore’s Law is dead and manufacturing costs in DRAM are not dropping by 30% in a year. Micron margins were negative for the year and they are looking to raise prices. If you know what you will need in DRAM you should buy it now.

 

[Doug S]

Apple is raking in ‘astronomical’ profits on the high volume but there gross margin actually dipped a bit in the last quarter from last year, from 43% to 42%. You know who else has good margins in computing, Microsoft, Intel and AMD. Microsoft gross margins on Windows has been close to 90%. Intel had gross margins of about 60%. Even AMD currently has gross margins higher than Apple.

RAM prices may be down 30% for the year but is that contract for high volumes and does either the manufacturer or DRAM vendor have any margin. In other words is it a sustainable business model. Moore’s Law is dead and manufacturing costs in DRAM are not dropping by 30% in a year. Micron margins were negative for the year and they are looking to raise prices. If you know what you will need in DRAM you should buy it now.

Apple has multi year contracts for stuff like DRAM and NAND with guaranteed availability, etc. These are great in times of shortages or rising prices, but they aren't getting the benefit of DRAM spot prices being down 30%. But they don't care, what COOs want is not the absolute lowest price but an absolutely predictable price. And more importantly, guaranteed availability (including flexibility if product sales are higher or lower than anticipated) so they know that such components will always be there when they need it. Even if there's a fire or earthquake that takes out 20% of the production as has happened before.

That's probably also why they tend to be a bit behind in DRAM speeds, a DRAM company might not want to guarantee pricing and availability on LPDDR5X if the contract was written before they were making any and didn't want to commit themselves to something that would cost them a lot of money if they couldn't deliver (and also because Apple requires huge quantities so going LPDDR5X in an iPhone is very different than going LPDDR5X in some high end Chinese phone that sells fewer than a million units a year)

Now theoretically Apple could buy the minimum on their DRAM contract and buy up some cheaper stuff at current prices but I'll bet their contract prevents them from doing so. If a DRAM OEM is going to commit to being flexible in supplying Apple and could conceivably lose money on the deal (if something happened which greatly increased the spot price so they were selling to Apple for far less than they could get on the open market) then they want to be sure they get the benefit of making more money when the spot market price is falling as it has the past year or two.

 

[FlameTail]

Question: Does dividing CPU cores into clusters reduce the multithreaded performance? Is the effect significant?

For example (Cinebench 2024):

Ryzen 7950X3D: 122 / 2100
M2 Max (8P+4E): 122 / 1025

Both have the same ST performance. However it seems the per core contribution to the MT score is worse on the M2 Max.

Ryzen 7950X3D: 2100 ÷ 16 = 131.25 per core

M2 Max: 900 ÷ 8P = 112.5

As you can see, the per core contribution to the MT score is higher on Ryzen.

What is the reason for this?

*Note: I subtracted 125 points from the 1025 points of the M2 Max, to remove the E-cores oit of the equation. I think 125 is a reaonable assumption for the E-cores' MT contribution.

 

[F-no]

Question: Does dividing CPU cores into clusters reduce the multithreaded performance? Is the effect significant?

For example (Cinebench 2024):

Ryzen 7950X3D: 122 / 2100
M2 Max (8P+4E): 122 / 1025

Both have the same ST performance. However it seems the per core contribution to the MT score is worse on the M2 Max.

Ryzen 7950X3D: 2100 ÷ 16 = 131.25 per core

M2 Max: 900 ÷ 8P = 112.5

As you can see, the per core contribution to the MT score is higher on Ryzen.

What is the reason for this?

*Note: I subtracted 125 points from the 1025 points of the M2 Max, to remove the E-cores oit of the equation. I think 125 is a reaonable assumption for the E-cores' MT contribution.

Yes there is a detrimental effect, assuming data needs to be passed in between clusters. Accessing data between clusters is slower than accessing data within clusters.

 

[TESKATLIPOKA]

Question: Does dividing CPU cores into clusters reduce the multithreaded performance? Is the effect significant?

For example (Cinebench 2024):

Ryzen 7950X3D: 122 / 2100
M2 Max (8P+4E): 122 / 1025

Both have the same ST performance. However it seems the per core contribution to the MT score is worse on the M2 Max.

Ryzen 7950X3D: 2100 ÷ 16 = 131.25 per core

M2 Max: 900 ÷ 8P = 112.5

As you can see, the per core contribution to the MT score is higher on Ryzen.

What is the reason for this?

*Note: I subtracted 125 points from the 1025 points of the M2 Max, to remove the E-cores oit of the equation. I think 125 is a reaonable assumption for the E-cores' MT contribution.

7950X3D has SMT. SMT adds another ~20-30% to the score.

 

[FlameTail]

On the CPU side, there is a performance cluster with 4 cores that reach a maximum of 4.056 GHz (or ~3.6 GHz when all cores are loaded) and an efficiency cluster with 4 cores that reach a maximum of 2.748 GHz

Apple M3 SoC analyzed: Increased performance and improved efficiency

A Notebookcheck analysis of the performance and efficiency of the new Apple M3 SoC, Apple's first 3nm chip, in comparison with AMD, Intel and Qualcomm.

www.notebookcheck.net

This is something that has been fuzzy for a long time. So the CPUs don't reach the max clock speed in multi-core scenarios?

 

[poke01]

Apple M3 SoC analyzed: Increased performance and improved efficiency

A Notebookcheck analysis of the performance and efficiency of the new Apple M3 SoC, Apple's first 3nm chip, in comparison with AMD, Intel and Qualcomm.

www.notebookcheck.net

This is something that has been fuzzy for a long time. So the CPUs don't reach the max clock speed in multi-core scenarios?

Nope. It's been like that since M1

 

[FlameTail]

Then the Dual-core boost in the Oryon CPU is no different to what Apple has been doing!

 

[FlameTail]

Nope. It's been like that since M1

Do you know of a source where all this information about varying clock speeds of Apple CPUs has been documented?

 

[poke01]

core boost in the Oryon CPU is no different to what Apple has been doing!

The whole industry does this but Apple's doesn't advertise clocks.

In multi-core workloads laptop CPUs from AMD and Intel lower their clocks in multi-core as well

 

[Mopetar]

The whole industry does this but Apple's doesn't advertise clocks.

In multi-core workloads laptop CPUs from AMD and Intel lower their clocks in multi-core as well

It's not that they lower the clock speeds. They list a specific all core clock speed that you can expect under a normal load, but if there's additional headroom a few of those cores may exceed that clock speed up to some designated boost number. Typically games benefit from that behavior as even those that use all of the cores typically have one or two threads that will scale the overall performance based on clock speed as they're the performance bottleneck on the CPU side of things.

 

[FlameTail]

Yes, but are you telling us that we do not know what Apple's clock speeds are in multi-threaded workloads?


Look at this. From Andrei's review on A14 Bionic.

Where can we find this information regarding the M series SoCs ?

 

[poke01]

I ran Cinebench 2024 on M3 8GB.

SC: 140
MT: 608

On par with 14900K for SC I believe the 16GB M3 would have gone higher as Geekerwan got 148SC with 24GB RAM.

Very impressive for a thin laptop. I don't Intel will get reach M3 levels of ST with MTL. I think AMD will with Zen5.

 

[capricornsunvoyager2]

Yes, but are you telling us that we do not know what Apple's clock speeds are in multi-threaded workloads?


Look at this. From Andrei's review on A14 Bionic.

Where can we find this information regarding the M series SoCs ?

Have in mind these are the maximum frequencies. Given these processors, especially for phones and laptops, have very tight thermal and energy limits, I wonder if they do sustain those frequencies for the multithreaded loads.

For instance, in the most recent Geekerwan video on the 9300 on the vivo X100, in one moment, they specifically created a timeline graph with the frequencies of the cores running a game. I guess we need that kind of curve while running a heavy multithreaded workload (maybe a benchmark) to observe if those maximum frequencies are sustained and explain the discrepancy between the highly high single-core scores and the more moderate multi-threaded performance.

 

[Eug]

This may not mean much, but it was amusing nonetheless.

Even Apple doesn't have a use for the M2 Ultra Mac Pro

While taking screenshots of Apple’s “Scary Fast” event video for an article this week, I noticed something in Johnny Srouji’s lab: The four $7,000 Apple Pro Display XDRs that were prominently positioned around the room weren’t hooked up to Mac Pros—they were powered by Mac Studios.

 

[Glo.]

This may not mean much, but it was amusing nonetheless.

Even Apple doesn't have a use for the M2 Ultra Mac Pro

While taking screenshots of Apple’s “Scary Fast” event video for an article this week, I noticed something in Johnny Srouji’s lab: The four $7,000 Apple Pro Display XDRs that were prominently positioned around the room weren’t hooked up to Mac Pros—they were powered by Mac Studios.

Mac Studio is hands down the best desktop Apple computer they have ever designed. Its good that they are promoting it, over MP.

 

[StinkyPinky]

I honestly don't get the point of the Mac Pro since it cannot use discrete graphics, why even put it in such a huge tower.

 

[Glo.]

I honestly don't get the point of the Mac Pro since it cannot use discrete graphics, why even put it in such a huge tower.

So that you can show on Social media that you are "Pro", not just... "Studio".

 

[FlameTail]

Where is the Extreme chip (4×Max)?

That's about the only thing that could justify the Mac Pro

 

[Doug S]

This may not mean much, but it was amusing nonetheless.

Even Apple doesn't have a use for the M2 Ultra Mac Pro

While taking screenshots of Apple’s “Scary Fast” event video for an article this week, I noticed something in Johnny Srouji’s lab: The four $7,000 Apple Pro Display XDRs that were prominently positioned around the room weren’t hooked up to Mac Pros—they were powered by Mac Studios.

If they aren't using any PCIe cards in their work then there is no reason to use a Pro over a Studio.

 

[eek2121]

Me < wished I had invested in Apple in 2022 as I intended to, then i could be retiring now and grateful for Apple's greed

I once owned a few hundred Apple shares back at the lowest point. I ended up selling them due to a combination of my own hardships and fears if them going banrupt.

Bye bye Apple Store monopoly? https://www.tweaktown.com/news/9430...ever-as-hidden-ios-17-2-code-shows/index.html

Only in Europe.

Yes, but are you telling us that we do not know what Apple's clock speeds are in multi-threaded workloads?

View attachment 88853
Look at this. From Andrei's review on A14 Bionic.

Where can we find this information regarding the M series SoCs ?

Because it varies based on workload, length of workload, and based on device.

I honestly don't get the point of the Mac Pro since it cannot use discrete graphics, why even put it in such a huge tower.

Same. The only reason I would ever have considered a Mac Pro is the ability to cram GPUs inside (oh and upgradable RAM/CPU). The Mac Pro is useless to me.

I see a lot of hate towards pricing, but IMO Macbook Pros are still the best laptops money can buy.

 

[Ajay]

I once owned a few hundred Apple shares back at the lowest point. I ended up selling them due to a combination of my own hardships and fears if them going banrupt.

Well that sucks . I made a typo - I was going to invest in Apple in 2002 (not 2022). I was going to do it on my lunch break and instead got laid off with 1,500 other people that day. Not sure why I didn't revisit that investment after got back on my feet. Oh well, spilt milk and all that.

 

[Doug S]

Well that sucks . I made a typo - I was going to invest in Apple in 2002 (not 2022). I was going to do it on my lunch break and instead got laid off with 1,500 other people that day. Not sure why I didn't revisit that investment after got back on my feet. Oh well, spilt milk and all that.

I made what I viewed at the time as a fairly modest investment in Apple in 2007, added to it in 2010, and ever since kept expecting "it has to be getting pretty close to played out". That holding alone is worth a few million now, but if I had been more willing to keep adding to it over time instead of looking at how much it had gained and thinking it couldn't continue growing at that pace I might be flying around the country on a private jet lol

Obviously the market as a whole has climbed so it isn't like what I did with that money instead of buying more AAPL was a loser, all the other investments just look puny compared to it on the graph showing the investment gains over time of each position!

But you can think that about any stock that goes on a tear. I added some NVDA far too late, if I had paid attention to the hype surrounding AI sooner that would have been a pretty big win over a fairly short time frame. Looking at INTC the last week I can't help but wonder what the heck is going on. I have been watching it for ages wondering about jumping in. I looked more closely and almost pulled the trigger a few weeks ago. When I saw what it had done today and looked back over the past couple weeks I couldn't help but mentally kick myself.

What I'm wondering about now is how much stuff like possibly forcing Apple to open up to competition for its App Store, prevent Google from paying to be the Safari search default, and so forth would affect the stock. Apple's entire strategy the past five years has been one of increasing installed base - because that increases post sale revenue. If those sources of post sale revenue are greatly restricted, then beyond the obvious affect on the stock price from reduced expectation for future profit you have to think Apple's management might rethink the way they approach the iOS market and then we're into real unknown territory.

 

[GC2:CS]

While M3 Max can pull much more power on CPU the performance between the 14” / 16” is almost identical (in high power mode) while fans are louder and battery much lower on the 14”. Tha wattage is over 10 W less than 16” but this does not show in scores. Possibly due to extreme power scaling at tail end or that the 14” gets better binned chips ?

Still this is insane from such a small notebook.

M3 Pro is pretty much identical in perf to M2 Pro save for a few tasks like RT rendering.
However the power cosumption is much reduced.

This reminds me of the M2 vs M1 comparisons. While all direct power measurements showed thet the power went up gen on gen, the battery life and thermals improved in pretty much all situations.

We need more M3 battery tests now performance is quite a known at this point.

 

[mikegg]

What I'm wondering about now is how much stuff like possibly forcing Apple to open up to competition for its App Store, prevent Google from paying to be the Safari search default, and so forth would affect the stock. Apple's entire strategy the past five years has been one of increasing installed base - because that increases post sale revenue. If those sources of post sale revenue are greatly restricted, then beyond the obvious affect on the stock price from reduced expectation for future profit you have to think Apple's management might rethink the way they approach the iOS market and then we're into real unknown territory.

I don't think the Safari search default revenue is a risk. Yes, it's insanely profitable for Apple to just get 37% of pure profit from Google. But if anything, Apple will just open up default to Bing if the court forces them. They'll still make a ton of money whether it's from Google or Bing.

Lastly, the nuclear option is that Apple has to make their own search engine and they'll start their own search business which will be profitable down the line. I don't see this option as likely. The court can't really do much about default search engine revenue because if they ban it, it'll kill browsers like Firefox and Opera, giving all the market share to Chrome, Safari, and Edge.

As for the app store, I don't think it'll change much. I think 99% of people will continue to use the default app store no matter what. iOS users trust that any app they download on the default App Store isn't a scam or virus.

AAPL stock hasn't really been effected by app store and search engine news. IE. bad news for these 2 topics have not decreased the price much on the same day. I think traders are reacting to the news accurately - that the revenue is not really at risk. Even if it is, it'll be many many years before the effects take hold.