By: Maynard Handley (name99.delete@this.name99.org), June 22, 2020 12:28 pm
Room: Moderated Discussions
Linus Torvalds (torvalds.delete@this.linux-foundation.org) on June 22, 2020 12:46 pm wrote:
> I have to admit that it looks good.
>
> First reaction: it sounds like "ARM" is a dirty word at Apple. It's "Apple Silicon". Lol.
Well, to be fair, most of the heavy lifting IS Apple's work. Sure ARM defines the architecture (and does a great job) and some of the inter-block comms protocols, but Apple defines the micro-architectures, the GPU, the NPU, the ISP, the memory controller, media, ...
> And I find it gratifying how they demoed Linux as a foreign environment, but the cynic in me wonders
> if it was because it was much easier to do an ARM Linux distribution with a paravirtualized kernel
> and virtual GPU interface. Which is obviously the technically right thing to do anyway, but might
> be hiding some of the darker corners. Windows might be a lot harder to do efficiently.
>
> And showing a legacy Maya app doing rendering similarly might be a "oh, it performs really well because all
> the heavy lifting is done by the Metal libraries and there's hardly any critical x86 code running".
>
> Which was how most of the original Rosetta performance goals were done
> too, of course (back then not Metal, but all the GUI libraries etc).
>
> Will be interesting to see the reactions from people leaking out the development
> hardware (despite possibly NDA's in place for early hardware access).
>
> Linus
> I have to admit that it looks good.
>
> First reaction: it sounds like "ARM" is a dirty word at Apple. It's "Apple Silicon". Lol.
Well, to be fair, most of the heavy lifting IS Apple's work. Sure ARM defines the architecture (and does a great job) and some of the inter-block comms protocols, but Apple defines the micro-architectures, the GPU, the NPU, the ISP, the memory controller, media, ...
> And I find it gratifying how they demoed Linux as a foreign environment, but the cynic in me wonders
> if it was because it was much easier to do an ARM Linux distribution with a paravirtualized kernel
> and virtual GPU interface. Which is obviously the technically right thing to do anyway, but might
> be hiding some of the darker corners. Windows might be a lot harder to do efficiently.
>
> And showing a legacy Maya app doing rendering similarly might be a "oh, it performs really well because all
> the heavy lifting is done by the Metal libraries and there's hardly any critical x86 code running".
>
> Which was how most of the original Rosetta performance goals were done
> too, of course (back then not Metal, but all the GUI libraries etc).
>
> Will be interesting to see the reactions from people leaking out the development
> hardware (despite possibly NDA's in place for early hardware access).
>
> Linus