> Apple wouldn't go down this path if they weren't confident that their designs would scale and keep them in the performance lead for a long time.
There are two reasons to think this might not be the case. The first is that they could justify continuing to do this to their shareholders based solely on the cost savings from not paying margins to Intel, even if the performance is only the same and not better. Their customers might not appreciate having the transition dumped on them in that case, but Apple has a specific relationship with their customers.
And the second is that these things have a long lead time. They made the call to do this at a time when Intel was at once stagnant and the holder of the performance crown. Intel is still stagnant but now they have to contend with AMD. And with whatever Intel's response to AMD is going to be now that they've finally got an existential fire lit under them again.
So it was reasonable for them to expect to beat Intel's 14nm++++++ with TSMC's 5nm, but what happens now that AMD is no longer dead and is using TSMC too?
The first is that they could justify continuing to do this to their shareholders based solely on the cost savings from not paying margins to Intel, even if the performance is only the same and not better.
You know Apple’s market capitalization is a little over $2 trillion dollars, right? And Apple's gross margins have been in the 30-35% range for many years. This isn't a shareholder issue. They are by far the most profitable computer/gadget manufacturer around.
So it was reasonable for them to expect to beat Intel's 14nm++++++ with TSMC's 5nm, but what happens now that AMD is no longer dead and is using TSMC too?
No matter what AMD does in the short term, they're not going to beat the performance per watt of the M1, let alone the graphics, the Neural Engine and the rest of components of Apple's SoC. It's not just 14nm vs. 5nm; it's also ARM’s architecture vs. x86-64.
Apple has scaled A series production for more than a decade and nobody has caught them yet in iPhone/iPad performance. There were 64-bit iPhones for at least a year before Qualcomm and other ARM licensees could catch up.
There's no evidence or reason to believe it'll be any different with the M series in laptops and desktops.
> You know Apple’s market capitalization is a little over $2 trillion dollars, right?
That's the issue. Shareholders always want to see growth, but when you're that big, how do you do that? There isn't much uncaptured customer base left while they're charging premium prices, but offering lower-priced macOS/iOS devices would cannibalize the margins on existing sales. Solution: Increase the margins on existing sales without changing prices so that profitability increases at the same sales volume.
> No matter what AMD does in the short term, they're not going to beat the performance per watt of the M1
Zen 3 mobile APUs aren't out yet, but multiply the performance of the Zen 2-based Ryzen 7 4800U by the 20% gain from Zen 3 and the multi-threaded performance (i.e. the thing power efficiency is relevant to) is already there, and that's with Zen 3 on 7nm while Apple is using 5nm.
> it's also ARM’s architecture vs. x86-64.
The architecture is basically irrelevant. ARM architecture devices were traditionally designed to prioritize low power consumption over performance whereas x86-64 devices the opposite, but that isn't a characteristic of the ISA, it's just the design considerations of the target market.
And that distinction is disappearing now that everything is moving toward high core counts where the name of the game is performance per watt, because that's how you get more performance into the same power envelope. Epyc 7702 has a 200W TDP but that's what allows it to have 64 cores; it's only ~3W/core.
> Apple has scaled A series production for more than a decade and nobody has caught them yet in iPhone/iPad performance.
Ryzen 7 4800U has eight large cores vs 4 large plus 4 small in the M1 and even with your (hypothetical) 20% uplift multicore is just about matching M1. Single core is nowhere near as good.
'Architecture is basically irrelevant' not the biggest factor but not irrelevant - x64 still has to support all those legacy modes and has more complex front end.
You're working very hard to try to deny that Apple has passed AMD and Intel in this bit of the market. We'll have to see what happens at higher TDPs but they clearly have the architecture and process access to do very well.
> Ryzen 7 4800U has eight large cores vs 4 large plus 4 small in the M1 and even with your (hypothetical) 20% uplift multicore is just about matching M1.
We're talking about performance per watt. The little cores aren't a disadvantage there -- that's what they're designed for. They use less power than the big cores, allowing the big cores to consume more than half of the power budget and run at higher clocks, but the little cores still exist and do work at high power efficiency. It would actually be a credit to AMD to reach similar efficiency with entirely big cores and on an older process.
> Single core is nowhere near as good.
Geekbench shows Zen 3 as >25% faster than Zen 2 for single thread. Basically everything else shows it as ~20% faster. Geekbench is ridiculous.
> 'Architecture is basically irrelevant' not the biggest factor but not irrelevant - x64 still has to support all those legacy modes and has more complex front end.
This is the same argument people were making twenty years ago about why RISC architectures would overtake x86. They didn't. The transistors dedicated to those aspects of instruction decoding are a smaller percentage of the die today than they were in those days.
> No idea what Qualcomm has to do with this.
The claim was made that Apple has kept ahead of Qualcomm. But Intel and AMD have kept ahead of Qualcomm too, so that isn't saying much.
> You're working very hard to try to deny that Apple has passed AMD and Intel in this bit of the market.
People are working very hard to try to assert that Apple has passed AMD and Intel in this bit of the market. We still don't have any decent benchmarks to know one way or the other.
Half the reason I'm expecting this to be over-hyped is that we keep getting synthetic Geekbench results and not real results from real benchmarks of applications people actually use, which you would think Apple would be touting left and right if they were favorable.
We'll find out soon enough how things stand but just to point out that your first comment on small vs large cores really doesn't work - the benchmarks being quoted are absolute performance not performance per watt benchmarks. Small cores are more power efficient but they do less in a given period of time and hence benchmark lower.
AMD should easily beat Apple in graphics, all they have to do is switch to the latest Navi/RDNA2 microarchitecture. They are collaborating with Samsung on bringing Radeon into mobile devices, surely that will translate into efficiency improvements for laptops too.
> ARM’s architecture vs. x86-64
x86 will always need more power spent on instruction decode, sure, but it's not a huge amount.
Perhaps you haven't read the Anandtech article? [1]
Intel has stagnated itself out of the market, and has lost
a major customer today. AMD has shown lots of progress
lately, however it’ll be incredibly hard to catch up to
Apple’s power efficiency. If Apple’s performance trajectory
continues at this pace, the x86 performance crown might
never be regained.
There are two reasons to think this might not be the case. The first is that they could justify continuing to do this to their shareholders based solely on the cost savings from not paying margins to Intel, even if the performance is only the same and not better. Their customers might not appreciate having the transition dumped on them in that case, but Apple has a specific relationship with their customers.
And the second is that these things have a long lead time. They made the call to do this at a time when Intel was at once stagnant and the holder of the performance crown. Intel is still stagnant but now they have to contend with AMD. And with whatever Intel's response to AMD is going to be now that they've finally got an existential fire lit under them again.
So it was reasonable for them to expect to beat Intel's 14nm++++++ with TSMC's 5nm, but what happens now that AMD is no longer dead and is using TSMC too?