risc-v seems to be gaining traction

Amazon gets behind free rival to Arm’s microchips https://www.telegraph.co.uk/technology/2021/03/20/amazon-gets-behind-free-ri... I didn't get beyond the paywall, but it sounds like the article is based on ads that suggest Amazon is looking to hire engineers with risc-v experience in response to NVIDIA buying Arm. I thought that was an interesting development.

Hi, On Mon, Mar 22, 2021 at 08:39:15PM -0400, Warren McPherson via talk wrote:
Amazon gets behind free rival to Arm’s microchips https://www.telegraph.co.uk/technology/2021/03/20/amazon-gets-behind-free-ri...
I didn't get beyond the paywall, but it sounds like the article is based on
Thanks for the article (no paywall when I tried it). -- Znoteer znoteer@mailbox.org

| From: Warren McPherson via talk <talk@gtalug.org> | Amazon gets behind free rival to Arm’s microchips Most Amazon comsumer products contain ARM. But they are just part of SoCs that ARM buys (it doesn't design or produce them). Of course that could change, but why? The main Amazon exposure to ARM is the Graviton processor. I *think* they buy the design of the ARM core used in that chip from ARM corporation. Nobody yet has a RISCV core for sale that is as fast. Amazon could develop one in-house but that would probably take several years. My guess: a clean sheet RISCV design could beat an ARM, modestly. That's because, after all these years, the ARM architecture has quite a few barnacles. Apple designs the cores of its ARM implementations. But they have a pretty good license perhaps from the founding of ARM (with Acorn). It's hard for Apple to switch because of the binary software distribution model. They are just starting to migrate users from X86-64 to ARM. Quite a history: Motorola 68000 => Power => x86-32 => x86-64 => ARM64

On Mon, Mar 22, 2021 at 09:44:15PM -0400, D. Hugh Redelmeier via talk wrote:
Most Amazon comsumer products contain ARM. But they are just part of SoCs that ARM buys (it doesn't design or produce them). Of course that could change, but why?
Certainly most Android devices run ARM, with a few x86 exception.
The main Amazon exposure to ARM is the Graviton processor. I *think* they buy the design of the ARM core used in that chip from ARM corporation. Nobody yet has a RISCV core for sale that is as fast. Amazon could develop one in-house but that would probably take several years.
Graviton 2 is ARM Neoverse N1 core, so a standard core. Graviton 1 is ARM Cortex A72 cores so also standard cores.
My guess: a clean sheet RISCV design could beat an ARM, modestly. That's because, after all these years, the ARM architecture has quite a few barnacles.
ARMv8 actually cleaned up a lot of legacy from the design, and a number of ARMv8 designs don't include any backwards compatibility (so they can't run ARMv7 or older code at all).
Apple designs the cores of its ARM implementations. But they have a pretty good license perhaps from the founding of ARM (with Acorn).
Anyone can buy a license to design their own ARM implementations. They cost more than a license to just use standard ARM designed cores. Not unusual though. Qualcomm, Marvell, Apple, Applied Micro, Nvidia, Cavium, AMD, Samsung and probably others have done custom ARM compatible cores.
It's hard for Apple to switch because of the binary software distribution model. They are just starting to migrate users from X86-64 to ARM. Quite a history: Motorola 68000 => Power => x86-32 => x86-64 => ARM64
Apple has done it multiple times, and have managed to use software emulation to handle the transision. They will be fine. Of course given the iphone and ipad (we will pretend the newton never happened although it was partially responsible for ARM being around today) have been ARM forever, so it makes sense for Apple to expand the use of their own custom ARM designs to the laptop and desktop market too. No need to depend on what intel decides to make going forward. They had enough trouble with what IBM and freescale(motorola) was making in the past that didn't suit their future, and intel's failure to bring up new wafer processes probably doesn't inspire confidence at Apple. -- Len Sorensen
participants (4)
-
D. Hugh Redelmeier
-
lsorense@csclub.uwaterloo.ca
-
Warren McPherson
-
Znoteer