Apple's leading CPU team looks to be struggling as they have now delivered no CPU gains for the first generation ever. Apple likely had to delay their next core architecture due due to personnel overturn from the Nuvia exodus. Furthermore, with a new exodus surrounding Rivos, the future doesn't look so bright. Apple currently has the best client CPU cores, but can they hold this lead against stronger competition from Intel, AMD, and Qualcomm?
You are contradicting yourself, because you said in the article :
"SemiAnalysis believes that the next generation core was delayed out of 2021 into 2022 due to CPU engineer resource problems."
So was it delayed or not? Or maybe you think they delayed it to 2023?
There is no indication that Apple isn't following the roadmap they always had, which takes into account production costs, which you seem to completely ignore.
Yes. The reality is core architecture has made tiny improvements for the big core since firestorm.e what is contridictory here. Did you not see the drama with Mike f, after this was published. Even more turmoil than expected or known at the time. Fact is performance per clock stagnated super hard and that wasn't the plan and Intel, AMD, Arm haven't stagnated.
Well it's specifically about the CPU and then it's specifically about unfounded, guess-journalism, speculation about the issues with the team, when in fact it looks like the team that's still in place and was involved in the development decided to concentrate on GPU, power drain and a whole host of other clever not CPU speed-headlining issues which actually make more noticeable real-world differences to the performance of the phone. We know they design and decide on the camera specs three years out, so a lot of the decisions on what the chip needs to do are baked in from that (the phones are already as fast as they could possibly need to be on CPU performance, so it's kind of meaningless at this point).
The leap between this year and last year is about 4x iPhone 5S (The A7 chip had 1B transistors). I don’t get why this advancement is described as “no gains”.
There’s a whole bunch of new functionality as you mentioned, which requires “more transistors” to maintain the same performance metrics, if we’re dumbing it down.
Apple simply delivered a faster and less power hungry chip.
There are definitely improvement to the SoC. Less clear is whether there are improvements (beyond, perhaps, a few tweaks and bug fixes) to the CPUs (large and small).
After that, there is the issue of why this happened (minimal CPU improvement) and what this means for the future.
I'm not nearly as pessimistic as Dylan is, or as willing to pin this all on an exodus of engineers or (the second theory people are floating) that the Apple design has hit a wall.
My guess is that there is *essentially* a plan here:
What I suspect is that they have hit the limits of an annual CPU redesign cycle. They have been under astonishing pressure (worse every year!) not just to substantially improve the core (and the rest of the SoC) but to do it by a fixed, unmovable date. It's amazing they kept this up for ten years, but even Stakhanovites eventually crack.
An alternative would be to switch to a biannual cycle, something like
- odd years (starting with 2021 [or early 2022]) Mac gets new "extreme" core
- even years (like 2022) that core gets optimized and whatever didn't work as well as hoped for (mac is a much more forgiving power environment!) is tweaked down to the new iPhone core.
Most Apple products are already on a two year (or slower) when you look at the tech details, especially the core; look at eg aWatch or iPads or aTV, or Macs. Even the iPhone 5C/SE/SE2 has operated on a slower cadence.
It can still look like annual updates for iPhone (and perhaps even for Macs) via improvements to the SoC (just not the core) -- as we saw this year [and this may be the essence of what the M1X is about vs the M2].
Essentially this makes Apple's life *much* easier.
It decouples the riskiest part of the design, ie new core (and associated complex IP) from the forced schedule of iPhone, since new Macs ship when they ship -- We all want to see what the M2 looks like and its associated iMac and Mac Pro; but no-one has any expectation that it's in Nov this year or May next year.
And so Apple can be more aggressive in future core designs if they have two years to work on the changes, AND can slip a month or three from the target date with no serious consequences.
We always assumed new cores would debut on iPhone, and till this year that made sense; but going forward it actually makes a lot more sense to have the newest most performant core debut in the devices for which they are most desired, ie iMac Pro and Mac Pro.
In a way this is a classic Apple curveball, where we have a certain expectation of how to solve a problem because of the way things have always been done, but Apple offers a different (and probably better) solution; cf the way they handled the "we need to sell a cheaper iPhone" problem by selling last year's iPhone rather than designing an alternative line.
This may be too optimistic, of course; but I'd keep it in mind as a theory at least until if the M2 disproves it (which it will if the M2 is not a substantially new core).
Wait the A14 on the ipad air (4th gen) also has ~ 6100 in SFFT, does this means apple either disabled half of the FP32 units on the iphone 12s A14’s or they just ran it on a much lower frequency ?
Even Apple silicon must deal with Moore's law, so gains may start to diminish after a couple refreshes. I think the ARM model works well for Apple, specifically iPhones, iPads, and Macbook's. I am not so sure on heavy lifting desktops like the Mac Pro or an iMac Pro.
The A15 would have been taped out and in risk production well before staff left for Nuvia, and in fact before the Covid-19 shutdowns hit California hard. TSMC has also missed with 3nm by four months, it'll be ready for the next ipad/macbook generation in 04/2023.
Whatever happened to Apple's CPU team won't be visible until then.
Yup took them multiple years after the team left to pick up the pieces, and still very meager gains versus what AMD and Intel have done last few years.
Embarrassing take based on a single iteration and tons of baseless speculation.
And returning back years later. The ST perf per clock has gone nowhere. It wasn't based on a single iteration lol.
You are contradicting yourself, because you said in the article :
"SemiAnalysis believes that the next generation core was delayed out of 2021 into 2022 due to CPU engineer resource problems."
So was it delayed or not? Or maybe you think they delayed it to 2023?
There is no indication that Apple isn't following the roadmap they always had, which takes into account production costs, which you seem to completely ignore.
Yes. The reality is core architecture has made tiny improvements for the big core since firestorm.e what is contridictory here. Did you not see the drama with Mike f, after this was published. Even more turmoil than expected or known at the time. Fact is performance per clock stagnated super hard and that wasn't the plan and Intel, AMD, Arm haven't stagnated.
How did this bullshit age?
And returning back years later. The ST perf per clock has gone nowhere. It wasn't based on a single iteration lol.
No Mea Culpa, Dylan? You schooled yourself. Get your facts straight, please, before shilling for Intel and Qualcomm.
Thinking critically is important Joseph. Apparently I can't call out a fact without being attacked. You informed the first sentence too.
"the best CPU cores for consumer workloads for years. They have by far the highest performance per clock and efficiency"
Why didn’t you think critically while writing this claim chowder?
Well it's specifically about the CPU and then it's specifically about unfounded, guess-journalism, speculation about the issues with the team, when in fact it looks like the team that's still in place and was involved in the development decided to concentrate on GPU, power drain and a whole host of other clever not CPU speed-headlining issues which actually make more noticeable real-world differences to the performance of the phone. We know they design and decide on the camera specs three years out, so a lot of the decisions on what the chip needs to do are baked in from that (the phones are already as fast as they could possibly need to be on CPU performance, so it's kind of meaningless at this point).
CPU teams do not work on GPU and other items. They are independent.
The leap between this year and last year is about 4x iPhone 5S (The A7 chip had 1B transistors). I don’t get why this advancement is described as “no gains”.
There’s a whole bunch of new functionality as you mentioned, which requires “more transistors” to maintain the same performance metrics, if we’re dumbing it down.
Apple simply delivered a faster and less power hungry chip.
This is specifically about the CPU.
The issue is CPU vs SoC.
There are definitely improvement to the SoC. Less clear is whether there are improvements (beyond, perhaps, a few tweaks and bug fixes) to the CPUs (large and small).
After that, there is the issue of why this happened (minimal CPU improvement) and what this means for the future.
I'm not nearly as pessimistic as Dylan is, or as willing to pin this all on an exodus of engineers or (the second theory people are floating) that the Apple design has hit a wall.
My guess is that there is *essentially* a plan here:
What I suspect is that they have hit the limits of an annual CPU redesign cycle. They have been under astonishing pressure (worse every year!) not just to substantially improve the core (and the rest of the SoC) but to do it by a fixed, unmovable date. It's amazing they kept this up for ten years, but even Stakhanovites eventually crack.
An alternative would be to switch to a biannual cycle, something like
- odd years (starting with 2021 [or early 2022]) Mac gets new "extreme" core
- even years (like 2022) that core gets optimized and whatever didn't work as well as hoped for (mac is a much more forgiving power environment!) is tweaked down to the new iPhone core.
Most Apple products are already on a two year (or slower) when you look at the tech details, especially the core; look at eg aWatch or iPads or aTV, or Macs. Even the iPhone 5C/SE/SE2 has operated on a slower cadence.
It can still look like annual updates for iPhone (and perhaps even for Macs) via improvements to the SoC (just not the core) -- as we saw this year [and this may be the essence of what the M1X is about vs the M2].
Essentially this makes Apple's life *much* easier.
It decouples the riskiest part of the design, ie new core (and associated complex IP) from the forced schedule of iPhone, since new Macs ship when they ship -- We all want to see what the M2 looks like and its associated iMac and Mac Pro; but no-one has any expectation that it's in Nov this year or May next year.
And so Apple can be more aggressive in future core designs if they have two years to work on the changes, AND can slip a month or three from the target date with no serious consequences.
We always assumed new cores would debut on iPhone, and till this year that made sense; but going forward it actually makes a lot more sense to have the newest most performant core debut in the devices for which they are most desired, ie iMac Pro and Mac Pro.
In a way this is a classic Apple curveball, where we have a certain expectation of how to solve a problem because of the way things have always been done, but Apple offers a different (and probably better) solution; cf the way they handled the "we need to sell a cheaper iPhone" problem by selling last year's iPhone rather than designing an alternative line.
This may be too optimistic, of course; but I'd keep it in mind as a theory at least until if the M2 disproves it (which it will if the M2 is not a substantially new core).
Wait the A14 on the ipad air (4th gen) also has ~ 6100 in SFFT, does this means apple either disabled half of the FP32 units on the iphone 12s A14’s or they just ran it on a much lower frequency ?
Even Apple silicon must deal with Moore's law, so gains may start to diminish after a couple refreshes. I think the ARM model works well for Apple, specifically iPhones, iPads, and Macbook's. I am not so sure on heavy lifting desktops like the Mac Pro or an iMac Pro.
The A15 would have been taped out and in risk production well before staff left for Nuvia, and in fact before the Covid-19 shutdowns hit California hard. TSMC has also missed with 3nm by four months, it'll be ready for the next ipad/macbook generation in 04/2023.
Whatever happened to Apple's CPU team won't be visible until then.
Nuvia team left in early 2019. A15 taped early 2020 AFAIK.
Pretty funny stuff. The charts look good & all, but overall the comic tone becomes repetitive and dull.
Apple tick-tocks between performance and screen upgrades.
Do you guys know that if Apple uses Arm V9 architecture for its A15 bionic chips or sill stick to Arm's V8 architecture?
It appears V8
Mostly through clock speed gains via node and TSMC performance improvement and the new metal stack.
Yup took them multiple years after the team left to pick up the pieces, and still very meager gains versus what AMD and Intel have done last few years.
They had to push out v9, correct.
A16 power efficiency gains weren't huge on CPU side outside node. Most the gains came from fabric improvements.
It was a delay. Talk to some their engineers lol