Friday, September 16th 2022
Report: Apple to Move a Part of its Embedded Cores to RISC-V, Stepping Away from Arm ISA
According to Dylan Patel of SemiAnalysis sources, Apple is moving its embedded cores from Arm to RISC-V. In Apple's Silicon designs, there are far more cores than the main ones that power the operating system and end-user applications. For example, embedded cores are present, and there are 30+ in M1 SoCs responsible for all kinds of workloads not related to the operating system. These tasks are usually associated with other functions such as WiFi/BlueTooth, ThunderBolt retiming, touchpad control, NAND chips having their own core, etc. They run their own firmware and power everything around the central cores that run the OS, so the whole SoC functions appropriately.
It appears that a lot of these cores are based on Arm M-series or lower-end A-series IP that Apple is currently looking to replace with RISC-V. Given that a large portion of software runs on the main big.LITTLE configuration, other secondary SoC tasks can migrate to a different ISA like RISC-V, with a small firmware adjustment. Given that these cores can be placed with custom IPs, Apple would save licensing fees if custom RISC-V cores were used. Additionally, developing firmware for these cores at an Apple engineering team size shouldn't be a problem. Of course, we have no information about when these custom cores will appear inside Apple Silicon. Even when they are used, no formal announcement is expected given that the main cores remain to be powered by Arm ISA, with everything else invisible to the end-user.
Source:
SemiAnalysis
It appears that a lot of these cores are based on Arm M-series or lower-end A-series IP that Apple is currently looking to replace with RISC-V. Given that a large portion of software runs on the main big.LITTLE configuration, other secondary SoC tasks can migrate to a different ISA like RISC-V, with a small firmware adjustment. Given that these cores can be placed with custom IPs, Apple would save licensing fees if custom RISC-V cores were used. Additionally, developing firmware for these cores at an Apple engineering team size shouldn't be a problem. Of course, we have no information about when these custom cores will appear inside Apple Silicon. Even when they are used, no formal announcement is expected given that the main cores remain to be powered by Arm ISA, with everything else invisible to the end-user.
23 Comments on Report: Apple to Move a Part of its Embedded Cores to RISC-V, Stepping Away from Arm ISA
I will concede that lack of competition in the mobile space may also be a factor in this, but I will remind you that Apple is also using these designed in their laptops, so they are competing with AMD and Intel on that front.
So, I think it's time for the next generation of low power CPUs, and I don't think it will be ARM based.
If you don't know what AIM was or what happened, grab a barf bag & hold on to your seat when you google it... man I was using ALOT of pepto & Aspirin back then....hehehe :)
Plus any savings turn into big fat checks for the execs ;) So yeah it's worth the hustle.For IPC improvements you can optimize something so much without adding more instructions=transistors and the more instructions they add the closer they get to x86 and it will defeat the purpose.
RISC/ARM is the future at least should be.
Apple proved that there is nothing that ARM can't do even with apps that are not native. So if the world switched to ARM right this second and everybody just made ARM version of their software in about a year x86 would be totally useless especially in the mobile world.
Take Steam Deck that tiny APU probable uses same power as M1 Pro/Max and is like 20 times slower.
With ARM you can have Ryzen 5800 / RTX 3070 in something small like the Steam Deck today where with the x86 in about 10 years.
One of your predictions is already correct though: nobody cares about x86 in the mobile world unless you count laptops.
In my opinion, it is not that ARM have reached some sort of “peak” but rather the fab improvements have slowed down drastically. Name of the node is not the same as the actual product, and while something like a 7nm to 4nm sounds like a big improvement, it is not. With fab improvement falling behind, the improvement in SOC is also falling behind. You can tell that modern SOC be it from the likes of Qualcomm, Mediatek or Apple, they throttle fairly significantly under sustained load. So you can make a ridiculously power SOC, but the fab just can’t accommodate it.
That's what you get for short sighted lack of diversification for the sake of profit$$$.
I also don't see any danger coming from custom extensions - as long as they don't break compatibility with basic ISA and standard extensions defined by Arm. Here I suppose that Arm demands full compatibility for any product to be branded or advertised as Arm.
this completely ignores the hpc cluster that is optional vector unit, and clunky variable-length vectors