4 ms·
On the serve side, AWS Graviton3 and Ampere Altera are pushing server workloads to optimize for ARM. On the consumer side, Apple Silicon is pushing the industr
by senttoschool 4y ago
On the serve side, AWS Graviton3 and Ampere Altera are pushing server workloads to optimize for ARM.
On the consumer side, Apple Silicon is pushing the industry to optimize for ARM.
I think we will see a ton of companies switch to ARM workloads in the server as soon as this year. I can see AWS Graviton3 taking over 80% of all new EC2 instances. Graviton2 already took over 50% of all new EC2 instances in the year 2020. [0]
[0] https://www.anandtech.com/show/16640/arm-announces-neoverse-v1-n2-platforms-cpus-cmn700-mesh https://www.anandtech.com/show/16640/arm-announces-neoverse-...
- quectophoton 4y ago> I can see AWS Graviton3 taking over 80% of all new EC2 instances. Graviton2 already took over 50% of all new EC2 instances in the year 2020. Speaking of this. Friendly reminder that AWS currently has a free trial[0] of 750 hours/month free for t4g.small EC2 instances, until the end of the year. You'd still need to pay for storage, but it's a good deal if you want to use it as DNS or WireGuard server for the remaining of 2023, since 1GB storage is enough for Alpine Linux. [0]: https://aws.amazon.com/ec2/instance-types/t4/ https://aws.amazon.com/ec2/instance-types/t4/
- KronisLV 4y ago> On the serve side, AWS Graviton3 and Ampere Altera are pushing server workloads to optimize for ARM. My hope is that eventually we'll get something as affordable as Scaleway's old ARM offerings in budget cloud providers (e.g. Hetzner, Scaleway, maybe Vultr and DigitalOcean etc.). Ideally, I'd get an ARM VPS for a few euros a month with 4 GB of RAM or something to that order and would be able to host whatever I need on those for cheaper than the x86 equivalents. Going from something like 5 Euros/month per node to 3 Euros/month doesn't seem like a big change, but it adds up, especially when your workloads would be well served by multiple separate low performance nodes. Of course, that'd be nice if I could easily build and test ARM container images or something to that tune from my x86 development machine, given that I already have the hardware and I doubt I'll personally be able to switch to ARM on the desktop/homelab CI nodes anytime soon. But hey, that's an inconvenience at worst, because you can still build things "in the cloud" from the source as well, provided that there aren't many ARM specific bugs, debugging which would be hard.