9 ms·

As much as Apple has neglected pros moving to ARM would definitely put video editors using Final Cut in the lurch. Its only really their macbook/macbook air li

As much as Apple has neglected pros moving to ARM would definitely put video editors using Final Cut in the lurch. Its only really their macbook/macbook air lineups where a move to ARM/RISC looks truly appealing. I don't really think Apple wants to completely abandon creative Pros. Yet if you aren't going to move the ENTIRE lineup off of Intel CPUs you're really limiting any potential benefit.

zenexer 8y ago
How are you arriving at this conclusion? It’s possible to transition from x86 to ARM without any loss of performance in the scenario you’ve described.
gumby 8y ago
First of all the ARM chips Apple ships are designed for iOS workloads (check out the dominance of graphics in area and internal bandwidth on the iPad Pro chips -- it calls to mind the Alto's explicit tradeoff of bus bandwidth and screen refresh). Apple's ARM parts have been smokers for the past few generations but they aren't designed for the sustained performance and instruction mix targeted by Intel. But that doesn't have to be true. MacOS CPUs need not be the same as the ones in iOS devices (after all they aren't today either). Apple could very well design ARM instruction set chips with special pipelined or SIMD instructions that support, say, the kind of transforms Final Cut needs. They could make sure Photoshop was a barn burner on their machines if they thought that mattered. Etc. The real question is the rest of the silicon (peripheral controllers like TB etc which they also get from their Intel deal). As for the tired trope of "As much as Apple has neglected pros" -- I'm tired of hearing that. Yes of course their mix isn't perfect for everyone but holy cow, my MacBook Pro has 4 40GPS TB ports! Hell, I get a lot of work done on their smaller devices, and the ever increasing portability is a huge plus. I'm a pro, not prosumer, and it works for me. I know my needs aren't the same as everybody else's, but I would bet that Apple knows the usage profile of everyone willing to share analytics, including how often people plug what sort of device into their machines. And I'm sure that informs their designs. Every design will fail to be a perfect match for some set of pros, and of course they should complain. Maybe it's just that they are designing machines for me? (in which case: fix the damned touch bar).
- consp 8y ago
  You are basically talking about the A76 core[1] and modified derivatives and its successors (e.g. neoverse[2]). [1] https://en.wikipedia.org/wiki/ARM_Cortex-A76 https://en.wikipedia.org/wiki/ARM_Cortex-A76 [2] https://en.wikichip.org/wiki/arm_holdings/microarchitectures/neoverse_n1 https://en.wikichip.org/wiki/arm_holdings/microarchitectures...
  - close04 8y ago
    The N1 is actually expected to be on par or faster than a Zen core in single threaded applications. Overall it would be a match for both Zen and Skylake-SP. Add to this Apple's specific brand of control over their software stack and you've got some exceptional potential to deliver high performance in a low power envelope. And all this while giving Apple full control over every aspect of their product (whether it leads to good or bad outcomes). Which could mean actual yearly refreshes instead of forgetting about a product for years because Intel had nothing in the pipeline. I mean if you're going to pay a hefty price premium you probably want something only Apple can give you. Not just a more expensive version of the same thing shipped by every other OEM out there.
    - speedplane 8y ago
      Moore's law is dead. Switching architectures to something that is more customizable to your current process is one of the only ways to squeeze more performance out of a given chip. It's why companies are increasingly building their own ASICs. If Moore's law was alive, they wouldn't need to, they'd just wait for the next general purpose chip to come out.
      - close04 8y ago
        This isn't just a matter of Moore's law. The lack of competition over the past decade also played a major role. Intel didn't feel pressed to deliver anything and instead decided to milk the market. It doesn't even matter if tomorrow we find a way to bring back Moore's law (performance increase, not necessarily transistor density). Apple probably doesn't want to be back here 10 years from now if competition is dwindling. And again, if they want to charge a lot more than the competition they have to justify it. This is the justification. Give me a super optimized CPU for which major software companies optimize their software and you've got my attention. I imagine Adobe would make sure their SW gets tha last drop of performance on Apple's machines. May not be the case with the competition's run of the mill ARM core.
zaphirplane 8y ago
Real question. Wouldn’t a video editor lean on GPU ?
- gumby 8y ago
  There's no general answer to that. For example the MPU could have a set of vector operations that use a special range of memory not mapped by the cache and that has its own path to the MPU. Those instructions could conceivably be faster (or have a different power profile) than GPU for certain tasks. I say "MPU" rather than "CPU" because I think of the CPU in a more traditional sense (ALU+register sets) while I think of system design as encompassing a whole range of functional elements which can include computational, memory, etc. Even within the CPU these days you'll find very complex behavior including speculative execution, various shadow registers and register windows etc.
writepub 8y ago
The comment seems to suggest some special secret sauce so damn specific to the x86 Arch that final cut pro simply stalls without it. This is inaccurate. Video editing - a DSP heavy compute scenario, probably benefits from a few SSE/SIMD instructions in x86, and possibly a bigger L1/L2/L3 cache in desktop grade Intel processors. None of the stated features are unique to Intel, and an ARM based SoC can easily incorporate similar techniques. Also, current ARM SoCs from Apple target phones and iPads, which automatically necessitates tradeoffs on inclusion of higher capacity L1/L2... caches & novelty accelerator instructions, not just for battery imposed power restrictions, but also based on use cases. The iPhone isn't designed for final-cut-pro, meaning it's CPU isn't designed for it either! But that doesn't mean anything is lacking in the instruction set (ARM)
- speedplane 8y ago
  Even X86 chips aren't really X86, they internally compile down to a different instruction set that is more optimally executed. Technical competition between architectures is hardly relevant now. It's more about building a developer network and access to partners with experience with the architecture.
- gumby 8y ago
  There's a lot more to the design of an MPU than its instruction set (as you point out yourself by discussing cache), different mixes of functional units and different paths to memory for example.
- CoolGuySteve 8y ago
  iMovie already ships on iOS and does roughly all the things Final Cut needs to do using GPU acceleration (which Apple also designs).
CoolGuySteve 8y ago
I wouldn't be surprised if a heavy duty Mac Pro-class ARM chip from Apple looks like AMD's Epyc, with an array of smaller 5-7nm ARM cores surrounding a larger 12-14nm bus core. Apple and AMD both use TSMC and this fab technology is apparently ready for prime time. It would also make sense from a manufacturing standpoint to use a slightly modified phone core in parallel for their poweruser desktop refreshes rather than design a significantly different CPU (otoh Apple seems to have no problem leaving the Mac Pro to languish for years at a time). They might even get better yields by fabbing a single core with both shared-MMU and the phone SoC features and then disabling one or the other if there are defects.

zenexer 8y ago

How are you arriving at this conclusion? It’s possible to transition from x86 to ARM without any loss of performance in the scenario you’ve described.

gumby 8y ago

First of all the ARM chips Apple ships are designed for iOS workloads (check out the dominance of graphics in area and internal bandwidth on the iPad Pro chips -- it calls to mind the Alto's explicit tradeoff of bus bandwidth and screen refresh). Apple's ARM parts have been smokers for the past few generations but they aren't designed for the sustained performance and instruction mix targeted by Intel. But that doesn't have to be true. MacOS CPUs need not be the same as the ones in iOS devices (after all they aren't today either). Apple could very well design ARM instruction set chips with special pipelined or SIMD instructions that support, say, the kind of transforms Final Cut needs. They could make sure Photoshop was a barn burner on their machines if they thought that mattered. Etc. The real question is the rest of the silicon (peripheral controllers like TB etc which they also get from their Intel deal). As for the tired trope of "As much as Apple has neglected pros" -- I'm tired of hearing that. Yes of course their mix isn't perfect for everyone but holy cow, my MacBook Pro has 4 40GPS TB ports! Hell, I get a lot of work done on their smaller devices, and the ever increasing portability is a huge plus. I'm a pro, not prosumer, and it works for me. I know my needs aren't the same as everybody else's, but I would bet that Apple knows the usage profile of everyone willing to share analytics, including how often people plug what sort of device into their machines. And I'm sure that informs their designs. Every design will fail to be a perfect match for some set of pros, and of course they should complain. Maybe it's just that they are designing machines for me? (in which case: fix the damned touch bar).

consp 8y ago
You are basically talking about the A76 core[1] and modified derivatives and its successors (e.g. neoverse[2]). [1] https://en.wikipedia.org/wiki/ARM_Cortex-A76 https://en.wikipedia.org/wiki/ARM_Cortex-A76 [2] https://en.wikichip.org/wiki/arm_holdings/microarchitectures/neoverse_n1 https://en.wikichip.org/wiki/arm_holdings/microarchitectures...
- close04 8y ago
  The N1 is actually expected to be on par or faster than a Zen core in single threaded applications. Overall it would be a match for both Zen and Skylake-SP. Add to this Apple's specific brand of control over their software stack and you've got some exceptional potential to deliver high performance in a low power envelope. And all this while giving Apple full control over every aspect of their product (whether it leads to good or bad outcomes). Which could mean actual yearly refreshes instead of forgetting about a product for years because Intel had nothing in the pipeline. I mean if you're going to pay a hefty price premium you probably want something only Apple can give you. Not just a more expensive version of the same thing shipped by every other OEM out there.
  - speedplane 8y ago
    Moore's law is dead. Switching architectures to something that is more customizable to your current process is one of the only ways to squeeze more performance out of a given chip. It's why companies are increasingly building their own ASICs. If Moore's law was alive, they wouldn't need to, they'd just wait for the next general purpose chip to come out.
    - close04 8y ago
      This isn't just a matter of Moore's law. The lack of competition over the past decade also played a major role. Intel didn't feel pressed to deliver anything and instead decided to milk the market. It doesn't even matter if tomorrow we find a way to bring back Moore's law (performance increase, not necessarily transistor density). Apple probably doesn't want to be back here 10 years from now if competition is dwindling. And again, if they want to charge a lot more than the competition they have to justify it. This is the justification. Give me a super optimized CPU for which major software companies optimize their software and you've got my attention. I imagine Adobe would make sure their SW gets tha last drop of performance on Apple's machines. May not be the case with the competition's run of the mill ARM core.

zaphirplane 8y ago

Real question. Wouldn’t a video editor lean on GPU ?

gumby 8y ago
There's no general answer to that. For example the MPU could have a set of vector operations that use a special range of memory not mapped by the cache and that has its own path to the MPU. Those instructions could conceivably be faster (or have a different power profile) than GPU for certain tasks. I say "MPU" rather than "CPU" because I think of the CPU in a more traditional sense (ALU+register sets) while I think of system design as encompassing a whole range of functional elements which can include computational, memory, etc. Even within the CPU these days you'll find very complex behavior including speculative execution, various shadow registers and register windows etc.

writepub 8y ago

The comment seems to suggest some special secret sauce so damn specific to the x86 Arch that final cut pro simply stalls without it. This is inaccurate. Video editing - a DSP heavy compute scenario, probably benefits from a few SSE/SIMD instructions in x86, and possibly a bigger L1/L2/L3 cache in desktop grade Intel processors. None of the stated features are unique to Intel, and an ARM based SoC can easily incorporate similar techniques. Also, current ARM SoCs from Apple target phones and iPads, which automatically necessitates tradeoffs on inclusion of higher capacity L1/L2... caches & novelty accelerator instructions, not just for battery imposed power restrictions, but also based on use cases. The iPhone isn't designed for final-cut-pro, meaning it's CPU isn't designed for it either! But that doesn't mean anything is lacking in the instruction set (ARM)

speedplane 8y ago
Even X86 chips aren't really X86, they internally compile down to a different instruction set that is more optimally executed. Technical competition between architectures is hardly relevant now. It's more about building a developer network and access to partners with experience with the architecture.
gumby 8y ago
There's a lot more to the design of an MPU than its instruction set (as you point out yourself by discussing cache), different mixes of functional units and different paths to memory for example.
CoolGuySteve 8y ago
iMovie already ships on iOS and does roughly all the things Final Cut needs to do using GPU acceleration (which Apple also designs).

CoolGuySteve 8y ago

I wouldn't be surprised if a heavy duty Mac Pro-class ARM chip from Apple looks like AMD's Epyc, with an array of smaller 5-7nm ARM cores surrounding a larger 12-14nm bus core. Apple and AMD both use TSMC and this fab technology is apparently ready for prime time. It would also make sense from a manufacturing standpoint to use a slightly modified phone core in parallel for their poweruser desktop refreshes rather than design a significantly different CPU (otoh Apple seems to have no problem leaving the Mac Pro to languish for years at a time). They might even get better yields by fabbing a single core with both shared-MMU and the phone SoC features and then disabling one or the other if there are defects.