Notices by Corey Quinn (quinnypig@awscommunity.social), page 3

@dalias each p5 server at AWS (featuring the H100) costs Amazon a third of a million dollars.

There’s an enormous waiting list for the cards.

@paul_ipv6 Pretty sure it’s all AI driven these days.

Uhhhh… this one took me by surprise. Amazon is a $1.790T company. That’s ridiculously huge; I can’t conceive of that scale.

As of today, Nvidia is larger.

Amazon RDS Aurora can helpfully store your credentials in the console via AWS Secrets Manager. Surprise! It costs 40¢ per month to store the creds.

If their underpants store division pulled the same thing:

“You don’t understand, AWS started charging per IPv4 address assigned to resources in your account at the start of the month. Check your billing console, it’s gonna be material by month-end!”

ChatGPT, create an image of a normal human villain wearing a glove.

Okay, I've been dancing around it for years--it's time to call it:

There's no economic story around using the cloud for steady-state HPC workloads. It's not even close to cost competitive.

"Well what about Spot?"

HPC workloads generally hate nodes going away mid-computation, plus there's no way to model the cost in advance given Spot's variable nature.

For some of these runs it's hard enough to get on-demand capacity, let alone finding enough *excess*.

In exactly one month Microsoft 365 is going to take a outage because… well, it’s not Microsoft 366 now is it?

“The new RHEL pricing to cloud partners will scale by vCPU count, which is consistent with the most common model for cloud virtual machines (VMs) and software" twenty years ago.

https://www.redhat.com/en/blog/red-hat-enterprise-linux-scalable-pricing-cloud-partners-announcement

“For some reason our AWS bill spikes 4% in February once every four years. I don’t understand why.”

Adopting this stray dog I found on the beach. He sure is bitey!

@evan all so recent!

The most surprising thing about building a Kubernete on a bunch of disparate hardware is how few of the projects I use publish Arm Docker images.

I have to run my own registry of (trivially) rebuilt images.

Everyone: “Kubernetes divorces node failures from the applications running on top of them.”

Me: *builds a kubernete, rips the power cable out of a worker node*

Kubernetes: “Ach! Mein Leben!”

Pods: *endlessly stuck in Terminating state*

k8s has played us for absolute fools.

Wake up asshole; you're the pod scheduler now.

23andme is apparently telling customers that suing them for their data breach is “futile.”

Pro tip: don’t take legal advice from the opposing party.

I cannot believe I have to start 2024 by explaining this to large companies.

It's once again the most wonderful time of the year: the newly-renamed Gartner Magic Quadrant for Strategic Cloud Platform Services!

This year there are no visionaries or challengers, just "cloud" vs. "you pretend to be a cloud." Let's explore together!

We're going to ignore the "niche players" because for three of them I don't speak Mandarin, and for IBM I don't speak ancient Greek.

That leaves AWS, Azure, Google Cloud, and Oracle Cloud.

First up is AWS due to its undisputed alphabetical supremacy.

Strengths include its "everything but the kitchen sink" approach, its innovation in hardware design, and its large feeding ground--I mean, partner ecosystem.