6 ms·

Every analysis I've read, including this one, assumes that Apple would start at the bottom rather than the top, as they have with every other CPU transition in

by ken 7y ago

Every analysis I've read, including this one, assumes that Apple would start at the bottom rather than the top, as they have with every other CPU transition in the history of the company. Is there something especially difficult about making a high-core-count CPU? The A-series is already crushing single-threaded performance and they're competitive with Intel desktop CPUs for some applications. Once you have a high-performance CPU with multi-core capabilities, is there any reason you can't just copy/paste cores?

_ph_ 7y ago
It isn't especially difficult, but is is expensive. The big challenge with CPU design is to make up for the costs and for that you need numbers. In the 90ies, there were plenty of high-performance CPU designs, Alpha, Sparc, Mips. They were alle killed by two mechanisms, that Intel had a lead in fab technology and that Intel had much higher numbers, consequentely could gain back much higher investments as the competitors could afford. The big question wouldn't be whether Apple could make a high-performance ARM, but if the financials play out for it. The Mac Pro numbers certainly wouldn't pay for such a CPU development. This could change of course, if they would bring back the Xserve in a modern incarnation or find some other uses for a high-performance ARM cpu.
- yoz-y 7y ago
  But Apple does not need to be a good or efficient fab because TSMC and Global Foundries already exist. One of the cited reasons for ARM transition is specifically that Intel fab sucks and their CPUs have been lagging in performance per watt for years.
  - snazz 7y ago
    Performance per watt isn’t too important on the Mac Pro and the low volumes mean that developing a purpose-built chip would be extremely cost-prohibitive.
- ken 7y ago
  > The Mac Pro numbers certainly wouldn't pay for such a CPU development. Again, though, what exactly does "development" mean here? Apple is already fabless. TSMC makes 32-core chips for AMD, so that scale is not new to them.
  - _ph_ 7y ago
    Well, the process of designing the chip to fabricate. This means paying a lot of electrical engineers to design the chip, probably paying for several iterations of lithography masks. Yes, TSMC certainly can make any chip of the desired size - they are already making enough of those - but the development is very expensive. The Mac Pro by itself is certainly not selling in high enough numbers to pay for such a development and considering how long Apple neglected the pro, I am somewhat dubious they would be willing to spend a huge amount of effort on it.
  - ksec 7y ago
    You will need to buy different set of tools to work and design for TSMC 's node. And while it is still from TSMC, the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ). The cost of designing and testing a 5nm chips is roughly a billion. For iPhone that is spread out over 200M unit over its life time, or roughly $5 per chip excluding the actual cost of wafer. For Mac you are looking at 25M unit per year, with TDP going from 15W to 250W. Apple could follow the same path as AMD and make one die to fit all. But in reality it is waste of engineering recourse and focus for little benefits.
    - ken 7y ago
      > the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ) The iPhone CPUs are already competitive with Intel chips. They've got incredible single-core performance, great memory bandwidth, and solid 64-bit and SIMD support. What specific feature of a "high performance" node would they require that they aren't using for iPhones now? They've been pushing the limits on allegedly "mobile"-oriented architectures for years. Every time this comes up, I hear from the naysayers "Yes, Apple is great at mobile, but desktop is a different game", but then also "Apple's mobile chips are faster than many Intel desktop chips". The skeptics sound awfully similar to those who said "Apple is great at iPods but smartphones are a completely different game". > The cost of designing and testing a 5nm chips is roughly a billion. AFAICT, nobody in the world has a production 5nm chip yet. How can you put a price on something that nobody has successfully produced yet? If I were Apple and I were confident that just throwing $1B at it would solve all the engineering problems, I would absolutely do it. $40 per Mac for the best CPU (and independence from Intel) sounds like a steal.

by ken 7y ago

_ph_ 7y ago

It isn't especially difficult, but is is expensive. The big challenge with CPU design is to make up for the costs and for that you need numbers. In the 90ies, there were plenty of high-performance CPU designs, Alpha, Sparc, Mips. They were alle killed by two mechanisms, that Intel had a lead in fab technology and that Intel had much higher numbers, consequentely could gain back much higher investments as the competitors could afford. The big question wouldn't be whether Apple could make a high-performance ARM, but if the financials play out for it. The Mac Pro numbers certainly wouldn't pay for such a CPU development. This could change of course, if they would bring back the Xserve in a modern incarnation or find some other uses for a high-performance ARM cpu.

yoz-y 7y ago
But Apple does not need to be a good or efficient fab because TSMC and Global Foundries already exist. One of the cited reasons for ARM transition is specifically that Intel fab sucks and their CPUs have been lagging in performance per watt for years.
- snazz 7y ago
  Performance per watt isn’t too important on the Mac Pro and the low volumes mean that developing a purpose-built chip would be extremely cost-prohibitive.
ken 7y ago
> The Mac Pro numbers certainly wouldn't pay for such a CPU development. Again, though, what exactly does "development" mean here? Apple is already fabless. TSMC makes 32-core chips for AMD, so that scale is not new to them.
- _ph_ 7y ago
  Well, the process of designing the chip to fabricate. This means paying a lot of electrical engineers to design the chip, probably paying for several iterations of lithography masks. Yes, TSMC certainly can make any chip of the desired size - they are already making enough of those - but the development is very expensive. The Mac Pro by itself is certainly not selling in high enough numbers to pay for such a development and considering how long Apple neglected the pro, I am somewhat dubious they would be willing to spend a huge amount of effort on it.
- ksec 7y ago
  You will need to buy different set of tools to work and design for TSMC 's node. And while it is still from TSMC, the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ). The cost of designing and testing a 5nm chips is roughly a billion. For iPhone that is spread out over 200M unit over its life time, or roughly $5 per chip excluding the actual cost of wafer. For Mac you are looking at 25M unit per year, with TDP going from 15W to 250W. Apple could follow the same path as AMD and make one die to fit all. But in reality it is waste of engineering recourse and focus for little benefits.
  - ken 7y ago
    > the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ) The iPhone CPUs are already competitive with Intel chips. They've got incredible single-core performance, great memory bandwidth, and solid 64-bit and SIMD support. What specific feature of a "high performance" node would they require that they aren't using for iPhones now? They've been pushing the limits on allegedly "mobile"-oriented architectures for years. Every time this comes up, I hear from the naysayers "Yes, Apple is great at mobile, but desktop is a different game", but then also "Apple's mobile chips are faster than many Intel desktop chips". The skeptics sound awfully similar to those who said "Apple is great at iPods but smartphones are a completely different game". > The cost of designing and testing a 5nm chips is roughly a billion. AFAICT, nobody in the world has a production 5nm chip yet. How can you put a price on something that nobody has successfully produced yet? If I were Apple and I were confident that just throwing $1B at it would solve all the engineering problems, I would absolutely do it. $40 per Mac for the best CPU (and independence from Intel) sounds like a steal.