Apple A14 - 5 nm, 11.8 billion transistors

[Eug]

A14 - First announced for iPad Air.

 

[Hitman928]

You are about to get a glimpse of the future tomorrow. Stay tuned.

What happens tomorrow? Is there a press release or something that I'm not aware of?

Edit: After a quick search, I'm assuming you are referring to Apple's event where they are expected to announce their ARM based Mac products. . .

 

[name99]

Nice narative, but not true.
Just the last 4 years Intel spent more than $12 Bn p.a. in R&D.
They are not cheap, they are inefficient.

I wasn't in the mood for writing up again a story I have written up a few times already.
What I meant with my synecdoche is that Intel prioritized finance thinking over engineering thinking. This means
- not taking risks, just doing more of the same slightly updated
- selling only what has already been proved to work rather than trusting that something new will be adopted by the market and someone will figure out a way to make it valuable (for example how slow they have been to add an NPU to their lineup)
- more concern for extracting money via market segmentation than with growing the ecosystem (witness the mess that has been AVX512 because of this, along with various other security features and misfires that have been restricted to narrow segments of the market)

All of this is not the behavior of tech companies in their growth phase, when they are confident about the future.

It's not exactly that they are cheap; inefficient yes, but also stupid/short-sighted in how they choose to spend money -- priorities chosen by finance fold not engineering folk, like ten thousand different SKUs.

 

[Eug]

You are about to get a glimpse of the future tomorrow. Stay tuned.

Apple's announcements tomorrow have nothing to do directly with AMD. Sure, both AMD and Apple use TSMC, but there's nothing stopping Intel from using TSMC as well, when push comes to shove.

This was in fact my point, that the issues here may be more related to process technology than CPU design per se. IOW, it isn't necessarily that AMD has suddenly bested Intel in everything related to the latter. They just happen to be working with a foundry that has been more successful than Intel has been of late for process shrinks.

 

[dmens]

IOW, it isn't necessarily that AMD has suddenly bested Intel in everything related to the latter.

Actually, that is exactly what I am saying. In my opinion, AMD has superior design, methodology, verification, and architecture, in addition to superior management at every level.

Tomorrow you will see that AMD is not the only company that left Intel in the dust.

 

[Eug]

Regarding A14 vs A14X in a Mac:

I've been watching the various iPad Air 4 reviews, and just about NOBODY is complaining about CPU performance of A14 in the iPad Air 4.

Yeah, I realize it's a different OS and all, but I still think that if A14 could address 16 GB RAM, it'd be perfectly fine for lower end Macs.

The only performance related issue I've come across in the reviews has to do with the lack of 120 Hz display support. However, you likely wouldn't be putting 120 Hz display support into a low end Mac anyway. Furthermore, we don't even know if the lack of 120 Hz is due to A14, or if it's just a limitation for market segmentation purposes and cost.

 

[Roland00Address]

So I missed the event, and have not had time to backtrack and watch the original thing or a summary.

Thus I ask, did Apple release any useful numbers?

 

[Eug]

So I missed the event, and have not had time to backtrack and watch the original thing or a summary.

Thus I ask, did Apple release any useful numbers?

No, but my gut feeling is M1 is roughly A14X/Z with tweaks.

4 high perf + 4 low perf CPU cores, plus 8 core GPU, although there is a 7 core GPU bin as well

Mac mini: M1 with 8-core GPU, fan
MacBook Pro: M1 with 8-core GPU, fan
MacBook Air: M1 with 7 or 8-core GPU, fanless

So basically, the MacBook Air is roughly the same machine as the Pro, but without the fan.

 

[IvanKaramazov]

I think the difference between the MBP and MBA being essentially a fan make sense if we assume this is the A14X. The A12Z in the iPad Pro apparently runs right around 9-10w in sustained load, which is precisely the sustained thermal budget for the MBA. But as we all know that's because the iPad throttles, and all the cores engaged at once must be closer to 16-18w. We'll have to wait for benchmarks but it's probable that the MBP can maintain those peak speeds.

EDIT: If this is accurate, incidentally, it may have ramifications for battery life on each device. The trade-off on the MBA may be that it's much, much faster under load than the Intel variant (which throttles horribly) but actually may not have improved battery life while under load (light usage should be significantly increased, due to the efficiency cores). On the MBP, on the other hand, it's very possible that peak power draw may still fall under the 28w TDP of the Ice Lake chips, meaning battery life under load should improve as well?

 

[Doug S]

So I missed the event, and have not had time to backtrack and watch the original thing or a summary.

Thus I ask, did Apple release any useful numbers?

Unfortunately no real numbers, as with any event Apple targets at consumers. The product pages for the new Macs don't even show clock rates.

We'll have to wait until they ship next week and Anandtech et al can run benchmarks on them. Even if Apple had released numbers people would be complaining they didn't release the right numbers.

I'd be very interested to see them run SPEC, then have Rosetta 2 translate x86 compiled SPEC and run that and compare the results. The differences between the various subtests would tell us a lot about how well Rosetta 2 works in the real world.

 

[name99]

So I missed the event, and have not had time to backtrack and watch the original thing or a summary.

Thus I ask, did Apple release any useful numbers?

Apple didn't. AnandTech did. Go read their article as a floor to what to expect.

 

[IntelUser2000]

The A12Z in the iPad Pro apparently runs right around 9-10w in sustained load, which is precisely the sustained thermal budget for the MBA. But as we all know that's because the iPad throttles, and all the cores engaged at once must be closer to 16-18w.

How did you get those numbers? If its the system power consumption the SoC is going to be lower.

25W Intel/AMD systems get load power consumption of 45W in the ultrabook form factor. The non-CPU part takes a significant portion in both light and heavy load.

 

[insertcarehere]

How did you get those numbers? If its the system power consumption the SoC is going to be lower.

25W Intel/AMD systems get load power consumption of 45W in the ultrabook form factor. The non-CPU part takes a significant portion in both light and heavy load.

Notebookcheck pegs the (entire) iPad pro averaging ~11-12w under load fwiw.

 

[IvanKaramazov]

How did you get those numbers? If its the system power consumption the SoC is going to be lower.

25W Intel/AMD systems get load power consumption of 45W in the ultrabook form factor. The non-CPU part takes a significant portion in both light and heavy load.

@insertcarehere is correct, and I goofed. That 9-10w was, as you suggested, the total power draw of the iPad Pro while playing an intensive game per the Anandtech review.

In terms of the M1 MBA, then, I suppose how aggressively and how soon it has to throttle the chip to keep at 10w sustained probably depends quite a lot on the impact of the frequency increase over the A14.