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

 

[richardskrad]

https://twitter.com/x/status/1635502297714397187

The first Apple A17 Geekbench 6 scores have potentially leaked:


ST 3019
MT 7860

 

[Cardyak]

The first Apple A17 Geekbench 6 scores have potentially leaked:


ST 3019
MT 7860

If this is true, then compared to the A16 the Single core performance is a ~20% increase, and the Multicore performance is a ~24% increase.

 

[Ajay]

If this is true, then compared to the A16 the Single core performance is a ~20% increase, and the Multicore performance is a ~24% increase.

Would make sense if true. N3 node plus some architectural improvements.

 

[moinmoin]

If this is true, then compared to the A16 the Single core performance is a ~20% increase, and the Multicore performance is a ~24% increase.

Geekbench 6's "multicore" anyway.

 

[poke01]

Geekbench 6's "multicore" anyway.

Yeah, we need Geekbench 5 for better multicore. So here:

https://twitter.com/x/status/1635506320479784961

Also single core is fast that's nearly on par with Intel and AMDs latest i9 and Ryzen 9 CPUs. That's in a phone CPU.

M3 if based on A17 will likely have the best Single core as they clock them higher. If it comes out at the end of this year. MTL will suck so it's likely.

 

[richardskrad]

When the M1 came out in 2020, it had the best single-core performance of any CPU on the planet while simultaneously having groundbreaking performance per watt. Now, AMD, Intel and Qualcomm have shortened the gap and the M2 isn't the fastest anymore but it's still the performance/watt champion.

There's absolutely a chance that Apple reclaims the single-core crown again with the A17/M3. They did it once before and the Johny Srouji's team know that they have to make a statement and answer all the bad press about Apple losing key chip designers the last few years.

 

[Heartbreaker]

Johny Srouji's team know that they have to make a statement and answer all the bad press about Apple losing key chip designers the last few years.

 

[Eug]

I haven't heard of Revegnus before. Are we to take those Geekbench 6 numbers seriously? Also, ShrimpApplePro didn't tell us from where the Geekbench 5 numbers came either, and even s/he says to take them with a grain of MSG.

 

[Doug S]

We've seen leaked numbers that were later shown to be correct, and numbers that were later shown to be completely made up (sometimes too high and sometimes too low, I guess depending on the goal of the person making them up)

We won't know which category these are until September, so it is pointless to discuss it. Someone could have just decided to make something up and figured "well N3 will give them a boost, plus some normal year to year increase so let's add 20% to the numbers then fudge them a bit so it isn't obvious I simply added 20%"

 

[Ajay]

I haven't heard of Revegnus before. Are we to take those Geekbench 6 numbers seriously? Also, ShrimpApplePro didn't tell us from where the Geekbench 5 numbers came either, and even s/he says to take them with a grain of MSG.

Yeah, I made a look be you leap comment. Random dude on internet says.....

Still, it would be nice if his numbers were true. I'm thinking moving to N3 provides the best opportunity for Apple Silicon to up it's game since the M1 came out.

 

[scannall]

We've seen leaked numbers that were later shown to be correct, and numbers that were later shown to be completely made up (sometimes too high and sometimes too low, I guess depending on the goal of the person making them up)

We won't know which category these are until September, so it is pointless to discuss it. Someone could have just decided to make something up and figured "well N3 will give them a boost, plus some normal year to year increase so let's add 20% to the numbers then fudge them a bit so it isn't obvious I simply added 20%"

Well yeah, leaks and rumors are for entertainment purposes only.Nothing to take seriously. Wait until parts are actually shipping and reviewed before making any judgements.

 

[Mopetar]

If we did that what in the hell would we have to talk about around here the other 95% of the time?

 

[Eug]

Correct me if I’m wrong but I don’t think an Apple Silicon benchmark score “leak” this early has ever been legit. We’re talking literally 6 months early here.

 

[Doug S]

Correct me if I’m wrong but I don’t think an Apple Silicon benchmark score “leak” this early has ever been legit. We’re talking literally 6 months early here.

I don't know the comparison between accuracy and lead time but it would seem to make it less likely. Heck, Apple may not have even selected the final operating frequency for A17 and they have time for at least one more stepping before they need to set it in stone and enter mass production.

Plus if I'm correct in assuming that Apple will not be using TSMC's N3 or N3E, but some Apple specific version of N3 that's got some of N3E's characteristics TSMC may not be quite done tweaking process parameters either.

So even if it is a legit leak from one of the likely very few Apple employees who have access to an iPhone 15 prototype at this time and would be risking their job, the numbers may not reflect what will ship in September.

 

[smalM]

I think N3 is the Apple specific version of N3. It was developed with Apple for Apple.
N3E will just be a relaxed, easier to design for, and cheaper to produce version of N3 for everyone else.
And btw I assume Apple will only make a moderate step forward with the CPU but a big step forward with the GPU, the NPU, and the AMX units.

 

[Doug S]

I think N3 is the Apple specific version of N3. It was developed with Apple for Apple.
N3E will just be a relaxed, easier to design for, and cheaper to produce version of N3 for everyone else.
And btw I assume Apple will only make a moderate step forward with the CPU but a big step forward with the GPU, the NPU, and the AMX units.

The reason I suggest Apple won't be using standard N3 is because in the past Apple has used new TSMC processes right out of the gate. TSMC enters mass production in April/May/June and Apple ships iPhones in September. Apple is involved in the design of the process, but it isn't truly specific to Apple's needs since TSMC is rolling it out as a standard foundry process.

N3 is different. It enters mass production a half year earlier than Apple needs it for iPhone 15, and N3E is following quickly on its heels but all evidence is it comes too late for Apple. Given that lead time and the fact N3E has been in risk production for months there would be plenty of time for TSMC and Apple to bring in some of N3E's improvements to an Apple specific flavor of N3. Remember the claims that N3 would only see a thousand wafers per month? That fits with expectations it would be used for lower volume stuff like Bitcoin ASICs (which were still a thing when that statement was made, there is probably a lot less of a market for those now) and not any Apple products.

I see no reason why Apple wouldn't want something "easier to design for and cheaper to produce". Sure iPhones are not cheap but per wafer costs are a lot higher with N5 than they were with N7, and with N3 or N3E get even higher. A more relaxed process means slightly larger chips but if it is made up for by cheaper wafers they may still come out ahead on cost which gives them more room to either once again hold the pricing steady in the face of inflation or at least use that BOM for other stuff, and the ability to select different transistor types in a more fine grained manner is something Apple could clearly benefit from.

Now maybe if M3 stuff starts shipping in June that lead time argument doesn't hold as much water, we'll have to see what happens there. Because past history with M1/A14 and M2/A15 points to using the same process and same core for M3/A17.

 

[soresu]

Now, AMD, Intel and Qualcomm have shortened the gap

ARM Ltd, not Qualcomm.

Until QC start cranking out those Oryon cores it's still ARM Ltd's µArch design work since they ditched the original v8-A Kryo core from the Snapdragon 820.

 

[Eug]

There is an article in the Financial Times about ARM wanting to charge more for licensing. It's paywalled, but here is Ars' report on it:

RISC-Y Business: Arm wants to charge dramatically more for chip licenses

Arm is tired of seeing device makers bring in billions while it makes millions.

arstechnica.com

According to the report, Arm wants to stop charging chip vendors to make Arm chips, and instead wants to charge device makers—especially smartphone manufacturers—a fee based on the overall price of the final product.

Apparently, this may not affect Apple (or Samsung)... yet, but who knows in the future.

 

[Heartbreaker]

There is an article in the Financial Times about ARM wanting to charge more for licensing. It's paywalled, but here is Ars' report on it:

RISC-Y Business: Arm wants to charge dramatically more for chip licenses

Arm is tired of seeing device makers bring in billions while it makes millions.

arstechnica.com

According to the report, Arm wants to stop charging chip vendors to make Arm chips, and instead wants to charge device makers—especially smartphone manufacturers—a fee based on the overall price of the final product.

Apparently, this may not affect Apple (or Samsung)... yet, but who knows in the future.

Why wouldn't it affect Apple or Samsung? The story just says the aren't using them as examples because they are more complicated, not that they would be exempted.

From the story:

The new plan is to raise prices across the board and charge "several times more" than it currently does for chip licenses. According to the report, Arm wants to stop charging chip vendors to make Arm chips, and instead wants to charge device makers—especially smartphone manufacturers—a fee based on the overall price of the final product.
...
Our example isn't picking on obvious companies like Apple or Samsung because they have much more complicated deals with Arm.

Awesome, so if there is a ARM CPU in the Ferrari, ARM gets a percentage cut of Million Dollar car's selling price. Brilliant!

Also from story:

Arm has been "frustrated by customers’ reluctance to accept the new arrangement."

No kidding. You want to increase pricing by hundreds of percent, without any improved service, and your customers aren't happy. Did Softbank put Martin Shkreli in charge of ARM?

This is just Rumor at this point. If ARM pushes ahead with this, they risk killing the Goose laying the Golden eggs. Apple is likely their single largest source of revenue, and Apple is the company most capable of designing their own RiscV processors, and most capable of switching their ecosystem over to RiscV, and as the story states Google plans to fully support RiscV in Android.

 

[Doug S]

Why wouldn't it affect Apple or Samsung? The story just says the aren't using them as examples because they are more complicated, not that they would be exempted.

Apple has a permanent "founders" architectural license. Everyone else's architectural licenses are subject to occasional renewal, so changes to the terms could be made at that time.

It wouldn't make sense to make changes to licensing intended to increase revenue that only affect companies licensing ARM designed cores and not companies designing their own cores under an architectural license. If there was a lot of money to be saved that way, everyone would stop using ARM designed cores and either design their own or buy from someone who does, and ARM would end up making LESS money and be paying a bunch of designers to design cores no one wants.

I thought Softbank was crazy to pay what they did for ARM, and thought Nvidia's rumored bid was even crazier. ARM has never had the revenue to justify that, and there was no feasible growth strategy to get there other than "huge price increase". If there were no alternatives other than commercial CPUs like x86 then this would work, but RISC-V is "good enough" for many and only needs a few more years to get everything in place to where pretty much all ARM customers except Apple could leave ARM behind.

Certain people have been theorizing for years and years that Apple would eventually abandon ARM for a proprietary architecture, without any reason or evidence - I guess just because Apple equals proprietary in their mind? Apple may end up with a proprietary architecture if ARM follows through on this plan, not because Apple abandons ARM, but because everyone else does.

 

[Heartbreaker]

Apple has a permanent "founders" architectural license. Everyone else's architectural licenses are subject to occasional renewal, so changes to the terms could be made at that time.

They do not. That is just made up nonsense from the internet. It's reported that Apple took out it's architecture license in 2008, after they purchased PA Semi back in 2008. This is years after Apple sold all their ARM stock.

In the real world when you sell your stake in a company, you aren't entitled to free access to all future tech that company ever develops.

 

[Roland00Address]

Awesome, so if there is a ARM CPU in the Ferrari, ARM gets a percentage cut of Million Dollar car's selling price. Brilliant!

Also from story:

Yes this is what Qualcomm does with their "patents" (a different form of IP), any phone that is $400 or less is charged a sliding fee, and it maxes out on $400 so an iPhone vs an iPhone Max pays Qualcomm the same.

=====

Good Luck to ARM *raises a glass* , for this will be fought tooth and nail. Including unofficial collusion where certain large seller of chips may buy more shares in ARM to make this harder to happen, and if it does happen it is seen as an arbitrage for they are paying ARM anyway so might as well get some of that fee back on the back end. A defensive form of acquisition.

 

[lopri]

That single-core performance is quite similar to Raptor Lake and Raphael's. Process + power envelope dictate performance?

 

[Heartbreaker]

Good Luck to ARM *raises a glass* , for this will be fought tooth and nail. Including unofficial collusion where certain large seller of chips may buy more shares in ARM to make this harder to happen, and if it does happen it is seen as an arbitrage for they are paying ARM anyway so might as well get some of that fee back on the back end. A defensive form of acquisition.

ARM is not a publicly traded company so they can't buy shares.

 

[moinmoin]

ARM is not a publicly traded company so they can't buy shares.

Not yet anyway. Which makes these rumours especially smelly.