Apple walks Ars through the iPad Pro’s A12X system on a chip
arstechnica.com
Apple walks Ars through the iPad Pro’s A12X system on a chip
1–10 of 173 posts
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#2Would have been interesting for Ars to have asked them how their GPU differs from the IMGTec/PVR IP they acquired besides adding more cores, what did they change?
They give the impression by saying it’s a custom GPU that it’s a from scatch in-house design but it’s unlikely to be the case.
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#3I would have figured anand would give this interview to anandtech.
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#4Would have been interesting for Ars to have asked them how their GPU differs from the IMGTec/PVR IP they acquired besides adding more cores, what did they change? They give the impression by saying it’s a custom GPU that it’s a from scatch in-house design but it’s unlikely to be the case.
The GPU they were licensing from Imagination was already very tailored to their needs. Apple already had a strong saying in the architecture, in the driver, in the compiler and in the implementation details of the GPU they were buying from them.
Moving to a "custom GPU" was basically Apple saying "Ok, thanks, we are taking over from here".
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#5I would have figured anand would give this interview to anandtech.
I guess his involvement may have made that look like a conflict of interest? AnandTech already has a reputation Apple favoritism (deserved or not, I can't say).
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#6I wonder how long it will be before Apple releases a laptop with an A-X chip.
They’re already shipping their custom T2 chips in their laptops. The compiler toolchain can build great binaries for their A chips. They’ve swapped CPU architecture before and the modern Mach binary format can hold versions of the executable built for different architectures.
They will probably need a Rosetta equivalent to emulate x86 for all the applications that are slow to switch. That might be tricky because of the huge surface area of the x86 instruction set. But I think it will only be a matter of time. It might also explain why they have kept the MacBook and MacBook Air product lines - they might want one of them to stay with intel’s cpus and the other to switch to their A* chips going forward. Or maybe they’ll just wait another generation or two and switch CPUs across their whole line in one go.
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#7Would have been interesting for Ars to have asked them how their GPU differs from the IMGTec/PVR IP they acquired besides adding more cores, what did they change? They give the impression by saying it’s a custom GPU that it’s a from scatch in-house design but it’s unlikely to be the case.
Idunno, the PVR-based products were pretty awful, drivers were full of system crashing bugs which could be triggered by reasonable shaders (On two separate occasions, I accidentally wrote shaders which would crash iOS and cause it to restart). I haven't touched their recent products, but I suspect they have been working on this stuff for a long time, because I think it'd be pretty hard to put up with that crap for any longer.
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#8I wonder how long it will be before Apple releases a laptop with an A-X chip. They’re already shipping their custom T2 chips in their laptops. The compiler toolchain can build great binaries for their A chips. They’ve swapped CPU architecture before and the modern Mach binary format can hold versions of the executable built for different architectures. They will probably need a Rosetta equivalent to emulate x86 for a…
This is what I'm waiting for. I can't find a reason to upgrade to the current line of MacBook (Pro, Air, etc.) Better performance, better battery life than Intel and I will no longer be funding Intel's lazy way.
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#9I wonder how long it will be before Apple releases a laptop with an A-X chip. They’re already shipping their custom T2 chips in their laptops. The compiler toolchain can build great binaries for their A chips. They’ve swapped CPU architecture before and the modern Mach binary format can hold versions of the executable built for different architectures. They will probably need a Rosetta equivalent to emulate x86 for a…
I don't see the point of an A-X chip in a laptop. You aren't taking photos on your laptop, why does it need the image signal processor? There's no FaceID (yet, maybe it's coming) on laptops, so we don't need the ML chip.
When you strip away all the stuff a laptop doesn't need, you're left with... an x86 chip!
Re: Apple walks Ars through the iPad Pro’s A12X system on a chip
#10I wonder how long it will be before Apple releases a laptop with an A-X chip. They’re already shipping their custom T2 chips in their laptops. The compiler toolchain can build great binaries for their A chips. They’ve swapped CPU architecture before and the modern Mach binary format can hold versions of the executable built for different architectures. They will probably need a Rosetta equivalent to emulate x86 for a…
>They will probably need a Rosetta equivalent to emulate x86 for all the applications that are slow to switch.
Well, and all the applications that won't switch. Also even on the Mac virtualization/containerization is not nothing. The Mac is a different market and use profile then iOS, and while Apple based on past history won't support an old arch indefinitely neither are they likely to completely blow off backwards compatibility. Compared to previous transitions dropping x86 would have extra complexities as well, so previous experience may not be entirely applicable. In particular Apple would be moving away from the full fat computer standard rather then no change or towards one, which may change the payoff for users despite Apple being much bigger. The absolute performance differences (immediate and future) also aren't likely to be as big.
I don't want to underestimate them, and huge disruption is inevitably coming down the pipe anyway and Arm may well emerge a winner there regardless, but it's also just a really big challenge.
>That might be tricky because of the huge surface area of the x86 instruction set.
Transmeta was able to do a decent job, and I think Novafora is still around and licensing their IP? Granted a lot of instructions have been added since then, but Apple certainly has a lot of expertise there as well and a great deal of capital to aim at the issue.