New cores (A15)

[cmaier]

A15 has the same core configuration (2+4) as A14, which suggests to me that they are of similar relative complexity as Firestorm vs. IceStorm, and the new high performance cores aren’t hardware-multithreaded. Which suggests to me that rumors of 8+2 or 10+2 for M1X/m2 are likely not right. Rumors only made sense if the cores were radically different in relative performance - either the low power cores can handle more pipelines or the high power cores can multi thread to compensate for blocking low priority threads. Based on all that, I doubt the 8+2/10+2 rumors. Also means that the new MBPs coming next month may indeed have M1X instead of M2(x).

 

[Gnattu]

In fact, from the figures Apple published, Firestorm is almost as fast as the A15's big core...

 

[Kung gu]

Why the low performance CPU uplift?

 

[cmaier]

Transistor count went way up, too, right? From around 12M to 15M? Wonder where they added.

 

[Argon_]

In fact, from the figures Apple published, the A15's big core is almost as fast as Firestorm...

Do you mean little core? Because the big core should be substantially faster than Firestorm, the big core of the A14 generation.

 

[Joelist]

If the A15 performance gains on the performance and efficiency cores are correct, the A15 efficiency cores are in the same speed class as Firestorm. So a 2+4 arrangement is going to be crazy fast - it's like having 6 Firestorms.

 

[Kung gu]

If the A15 performance gains on the performance and efficiency cores are correct, the A15 efficiency cores are in the same speed class as Firestorm. So a 2+4 arrangement is going to be crazy fast - it's like having 6 Firestorms.

Can you please elaborate everyone is saying there has been no CPU improvement.

 

[Kung gu]

Do you mean little core? Because the big core should be substantially faster than Firestorm, the big core of the A14 generation.

How can the little core be as fast Firestorm when no CPU uplift was told by Apple.

 

[Kung gu]

Apple says the A15 CPU is a 40% improvement compared to the A12.

 

[firewood]

Transistor count went way up, too, right? From around 12M to 15M? Wonder where they added.

Could there be significantly more 3rd level cache?

 

[Kung gu]

Transistor count went way up, too, right? From around 12M to 15M? Wonder where they added.

The extra GPU core. 5 core GPU now.

 

[crazy dave]

The extra GPU core. 5 core GPU now.

As big as GPU cores are I’m sure that’s not all of it, but yeah probably a decent fraction.

 

[cmaier]

The extra GPU core. 5 core GPU now.

There can’t be 3billion transistors in that core. After all, the entire chip only has 15 billion transistors - that would leave nothing for the rest of the chip.

 

[Gerdi]

Sounds to me as no performance uplift for the cores at all. Transistor budget was going into GPU, AI compute, additional video encoder/decoders and perhaps system level cache.

 

[crazy dave]

Apple says the A15 CPU is a 40% improvement compared to the A12.

Which is the same uplift as the A14 compared to the A12. Interesting.

 

[crazy dave]

There can’t be 3billion transistors in that core. After all, the entire chip only has 15 billion transistors - that would leave nothing for the rest of the chip.

Extra GPU core + more/larger accelerators? Dunno

 

[Sydde]

Size difference relative to A14: M1 +35%, A15 +27%. This is with the addition of one GPU core and a video en-decoder to the A15. My WAG would be that M2 will go to 2+6 CPU, no more than 4 more (improved) GPU cores and the video processor (probably no change in neural). Apple is inching toward offloading heavy work to dedicated logic, which makes a lot of sense. CPU and GPU cores will not proliferate very quickly because they will have much to do as dedicated logic starts to take on the heavy workloads.

 

[Kung gu]

@cmaier what is point in doubling SLC cache??

 

[Sydde]

There can’t be 3million transistors in that core.

billion

 

[cmaier]

billion

I think I fixed that before you actually posted this

 

[cmaier]

@cmaier what is point in doubling SLC cache??

Improved effective memory bandwidth. How much improved depends on the situation.

 

[Gnattu]

Do you mean little core? Because the big core should be substantially faster than Firestorm, the big core of the A14 generation.

It should, but it is not.

 

[Argon_]

It should, but it is not.

It wouldn't make sense for an A15 performance core to be almost as powerful, in other words less than an A14 core. In that case why not use a binned A14.

 

[Gnattu]

It wouldn't make sense for an A15 performance core to be almost as powerful, in other words less than an A14 core. In that case why not use a binned A14.

My bad, what I meant is that A15's big core does not have observable performance advantage over Firestorm.

Edited my original post

 

[Argon_]

I'm curious to see how the A15 benchmarks once numerous devices are in private hands.