New cores (A15)

[cmaier]

YOU DON'T SAY. I've only been telling you this for months.

And you may be right, or you may be wrong. There’s just as much evidence the other way. The fact that the new cores appear to be similar to the old, but optimized for N5P, suggests that Apple would use those instead (no way these new chips, whatever they are, are N5, and Apple is not going to optimize twice).

 

[wyrdness]

Here's my wild speculation. The current top end MBP 16" has an 8 core i9. So the forthcoming Apple Silicon MPB 16" will have at least 8 performance cores, maybe even 10 or 12, together with (most likely) 4 energy efficient cores.
Could Apple be focusing on reducing the power consumption of this years cores, since there could be quite a few of them in the next MPB's. More cores = quicker battery drain and they might be focusing on battery life.

I've no idea if this is anywhere near the mark, but it would seem to fit the observed facts.

 

[Mr. Bear]

I'm just curious...I assume that most of Apple's CPU design budget is oriented towards iPhone SoCs. So I'm wondering how much of the budget is spent on the Mac SoCs. I expect that it's (relatively) nothing, because all of the heavy lifting handled by the main design team. Or, to put it more bluntly, how much did it cost Apple to develop the M1, ignoring the costs associated with the A14, etc? My WAG is that Apple is spending a couple of billion on chip design, but less than 5% of that goes to Mac-specific SoC design. Am I anywhere in the ballpark?

 

[Sydde]

I wonder if Apple might have a SoC, for notebooks, that uses a clock splitter for the P cores, so that a few of them could run at a lower speed in order to act as semi-E cores. That might make a 6+2 configuration more sensible if pairs of the P cores could just run down-clocked instead of the whole CPU.

 

[Populus]

Hey people, I have a question about the cache.

The new A15 has doubled the system cache from 16 to 32MB.
What’s the cache of the M1? Is it 16MB? 32MB? Is there a system cache on the M1? Do you expect Apple to double it on the next M2 as well?

Those were several questions but the topic is the same: The system’s cache. I’ve only found data about L1 and L2 caches on the M1, not about a system’s cache.

Thank you

 

[leman]

Hey people, I have a question about the cache.

The new A15 has doubled the system cache from 16 to 32MB.
What’s the cache of the M1? Is it 16MB? 32MB? Is there a system cache on the M1?

As far as I know, it’s 16MB, same as A14

Those were several questions but the topic is the same: The system’s cache. I’ve only found data about L1 and L2 caches on the M1, not about a system’s cache.

L1 and L2 caches are CPU(cluster)-specific, LLC is shared by all the clusters on the SoC (it’s basically the shared cache before the RAM). This is the cache that allows quick data transfers between CPU, GPU, ML accelerators etc.

 

[Lemon Olive]

And you may be right, or you may be wrong. There’s just as much evidence the other way. The fact that the new cores appear to be similar to the old, but optimized for N5P, suggests that Apple would use those instead (no way these new chips, whatever they are, are N5, and Apple is not going to optimize twice).

Funny thing is, I now feel the opposite way, because the increased GPU performance of the A15 makes me think this chip is what is being fabbed for the MBP.

 

[Kpjoslee]

Looking at Metal Score leak that was posted on MacRumors news, it seems like there is a clockspeed increase. From ~3.0 Ghz to 3.23Ghz.

 

[cmaier]

View attachment 1832359

Looking at Metal Score leak that was posted on MacRumors news, it seems like there is a clockspeed increase. From ~3.0 Ghz to 3.23Ghz.

I don’t trust clock speed leaks until the benchmark people have a chance to debug with the new chip, but an interesting data point at Least.

 

[Populus]

As far as I know, it’s 16MB, same as A14


L1 and L2 caches are CPU(cluster)-specific, LLC is shared by all the clusters on the SoC (it’s basically the shared cache before the RAM). This is the cache that allows quick data transfers between CPU, GPU, ML accelerators etc.

Thank you for your explanation.
Hopefully the M1X or the M2 will jump to 32MB of system cache as well.

 

[Kpjoslee]

I don’t trust clock speed leaks until the benchmark people have a chance to debug with the new chip, but an interesting data point at Least.

I agree with you but with iPhone 13 being really close to launch at this point, I think it is safe to assume that may end up being true.

 

[cmaier]

I agree with you but with iPhone 13 being really close to launch at this point, I think it is safe to assume that may end up being true.

Could be. Looks like the GPU scores are very nice, fwiw

 

[ADGrant]

I'm just curious...I assume that most of Apple's CPU design budget is oriented towards iPhone SoCs. So I'm wondering how much of the budget is spent on the Mac SoCs. I expect that it's (relatively) nothing, because all of the heavy lifting handled by the main design team. Or, to put it more bluntly, how much did it cost Apple to develop the M1, ignoring the costs associated with the A14, etc? My WAG is that Apple is spending a couple of billion on chip design, but less than 5% of that goes to Mac-specific SoC design. Am I anywhere in the ballpark?

Well the M1 replaced what would have been the A14X in the iPad Pros so the incremental cost over what would have been the A14X was probably not that great. However, the M1 has replaced the Intel CPU in the lower end consumer Macs. To replace the Intel CPUs in the higher end Macs (in particular the desktop Macs), Apple is going to have to invest a lot more time and money. They need to support a lot more RAM, more high performance CPU cores and a lot more GPU cores.

I suspect the reason the A15 may not be a huge improvement over the A14 and the reason the CPU in the Apple Watch 7 is the same as last years is Apple's SoC team is focused on the new Mac SoCs.

 

[Kung gu]

Geekbench 5
ST: 1734
MT: 4818
3.23ghz clock

 

[crazy dave]

Geekbench 5
ST: 1734
MT: 4818
3.23ghz clock
View attachment 1832516

If true that’s a 10% performance increase in ST and a 24% increase in MT! That’s actually not too shabby … we’ll see

EDIT: where did you find this?

 

[quarkysg]

Geekbench 5
ST: 1734
MT: 4818
3.23ghz clock

Looks like the numbers from the A14 running at similar clock speed.

 

[Kung gu]

Looks like the numbers from the A14 running at similar clock speed.

No its higher than A14. clock speed increased from 3ghz to 3.2ghz.
This is iPhone 12 Pro Max

 

[Gnattu]

No its higher than A14. clock speed increased from 3ghz to 3.2ghz.
This is iPhone 12 Pro Max

View attachment 1832526

Apple M1 runs at about the same clock and has about the same score.
Seems like Avalanche has no obvious iso-frequency advantage over Firestorm.

 

[Kung gu]

Apple M1 runs at about the same clock and has about the same score.

Because the CPU cores are the same. Just a higher clock, thats why M1 scores the same 1700ST because it is also clocked at 3.2ghz

 

[crazy dave]

No its higher than A14. clock speed increased from 3ghz to 3.2ghz.
This is iPhone 12 Pro Max

View attachment 1832526

Weird I got smaller numbers for the iPhone 12 Pro Max

iPhone 12 Pro Max Benchmarks - Geekbench

browser.geekbench.com

 

[crazy dave]

Apple M1 runs at about the same clock and has about the same score.
Seems like Avalanche has no obvious iso-frequency advantage over Firestorm.

In performance no, hopefully yes in power efficiency at that point.

 

[cmaier]

Geekbench 5
ST: 1734
MT: 4818
3.23ghz clock
View attachment 1832516

Pretty close to the 14% single core performance I *um* guessed. Of course, the 14% is not based on any single benchmark, so it may actually be around 14%.

 

[cmaier]

In performance no, hopefully yes in power efficiency at that point.

if they run the same frequency at the same clock, then since the voltage is the same (presumably), it’s going to be the same power efficiency.

 

[Kung gu]

if they run the same frequency at the same clock, then since the voltage is the same (presumably), it’s going to be the same power efficiency.

The clock speed on A15 is 3.2ghz per geekbench.

 

[cmaier]

The clock speed on A15 is 3.2ghz per geekbench.

Sure, I just meant “if you normalize per clock tick”