The ARM-ification of the datacenter has begun, kick-started by the Apple M1
The Apple M1 chip will spur the ARM-ification of the datacenter, even though Apple is likely not going into the datacenter business.
Traditionally x86 CPUs were faster at running single thread tasks. x86 CISC (Complex Instruction Set Computing) offers more instructions vs ARM’s RISC. ARM was meant for low power and correspondingly smaller instruction set workloads, lowering single thread performance. Apple M1’s single-threaded performance benchmarks show that it blows Intel’s x86 CPUs out of the water, and that too in a Fan-less MacBook Air where the M1 is likely throttled to control heating. x86 CPUs come close to the M1 in multi-threaded performance, however this isn’t a good comparison due to the presence of 4 low-power cores on the M1. Let’s wait till an Apple M1 like Desktop-class ARM CPU with all high powered cores comes out.
So Apple has shown it’s ARM M1 SoC offers a higher single-threaded performance per watt than any x86 CPU out there. So why wouldn’t I, as a developer, demand the same performance and price from my data centre as well. In fact, once developers get used to developing on ARM MacBooks, they are going to demand ARM servers to deploy their workloads. This would make deployments seamless.
The other systemic change driving the ARM transition is that workloads and applications are changing. With micro-services and containerisation, developers create horizontally-scalable, non-blocking/async, stateless backends which work efficiently on single CPU threads, e.g. NodeJS. Also, all the apps we use on smartphones are ARM apps. In addition, Multi-threaded workloads, like machine learning, have moved to specialised GPUs rather than CPUs. Signal processing, voice recognition to FPGAs. So there aren’t many multi-threaded workloads available to run on CPUs.
Enterprises who don’t have any incentive to transition apps to ARM will be slow to transition. However, developers would start to vote with their feet and move to companies embracing this change in order to get a modern development environment to work with.
Also, NVIDIA bought ARM, and I expect big things to come from this acquisition.https://stackoverflow.blog/2020/09/21/nvidia-has-acquired-arm-what-does-this-mean-for-the-future-of-ai-edge-computing-and-the-people-who-write-software-for-these-chips/
A version of this article has been published on on twitter as well: https://twitter.com/mayankkapoor/status/1327540846502178818?s=20
