just_another_person @ just_another_person @lemmy.world

Posts

528

Comments

7,269

Joined

2 yr. ago

It depends on the use-case. If this VPS doesn't have layer7 routing out in front, you have a service that needs to respond from an A/CNAME hostname in combination with an SRV lookup, you have service that uses bidirectional TLS 1.2...etc.

There's all kinds of reasons you may need multiple static IPs for an instance. If youre just using vhosts for HTTP capable services, don't worry about it though.

"Old PC" meaning what, exactly? Need some specs.

In general, a Qnap or Synology box is going to be much lower wattage than a full PC.

Lol, okay, bud. Not only are you absolutely wrong and seem to have no professional experience with this whatsoever: search engines, engineering blogs, Wikipedia, history, and every other known source of truth on this disagree with you, yet here you are arguing anyway. Amazing. 😎

Yes, Windows sucks. What's your question?

Your machine isn't shutting down, it's trying to sleep.

You also have active KVM instances which are fighting to keep it alive.

Run a LiveUSB of anything else. If it works, install that instead.

Stop listening to AI.

Okay, well it doesn't NEED to be Docker Desktop for sure. There is just a difference in the configurations of how it works, but whatever Desktop can do, docker-ce can do.

For starters, let's find out why you can't install drivers. Do you have an error?

It will with Tailscale though, as OP mentioned using. Tailscale clients reach out to DERP endpoints to create the routes they need to communicate with other clients. dNAT wouldn't matter.

Source: don't this with Tailscale vanilla and Headscale

Any distro will work.Your issues with driver install should be obvious, but also let us know what card you're working.

There is a big difference between Docker Desktop and docker-ce when you're talking about CUDA integration and performance. You probably just want docker-ce with the nvidia-runtime configured.

The easier distros with Nvidia drivers are going to be the Debian, Fedora or Ubuntu flavors, and you need to pay attention to the CUDA requirement which is not always in-line with the stock gaming GPU version of drivers. Nvidia has a compatibility matrix you can lookup with detailed instructions on how to navigate what is what.

🍿

DynamicDNS will get you there without having to pay for VPS host traffic prices for ingress though.

How/why are you double NAT'd? Is it the ISP's fault, or did you setup your own router behind their NAT'd handoff? If the latter, you can work around this.

Some other options to deal with the former:

• Talk to them about running their handoff device in passthrough mode, or turn it on yourself (lookup the model and see instructions)
• Get a static IP for your router on their network (not applicable to CGNAT)

Regarding Tailscale: you shouldn't need a VPS for this. Even though you're stuck with dNAT, the Tailscale client should be able to figure out a way to communicate with you wherever you are with DERP servers. Install on your phone, then a machine at home, and try it out.

That is not what that term generally means. Somebody COULD be running their own cloud platform, but if you're speaking to a large group of people and you say "Cloud deployed", they understand that to be deployed to a Cloud Provider on a secured platform and location (AWS, Google, Azure...etc).

We don't say "cloud" in engineering anywhere without meaning this. We may refer to a non-colocated deployment of something as "edge" or "off-site", but never "cloud". There isn't a single engineer on this planet who would ever confuse "deployed to cloud" to mean somebody's basement.

"Cloud" infers centralized consolidation of resources in a datacenter. A PaaS, for example.

"Decentralized" infers any number of running instances of something that are not tied to any specific vendor, infrastructure, or location.

Cloud can be distributed, but not decentralized since the underlying controls of the infrastructure are themselves centralized.

According to this paste, you're not even using inference at all, or rather, it's using CPU.

Change the release tag and "cpu" to openvino and see if that performs any better.

Wtf

You just "snapshot" the existing disk and run it in a VM like VirtualBox or something. Then that new disk image works as the "new" disk for the VM.

You're overcomplicating things.

If you have an existing Win10 install, just format the new SSD for whatever distro, install, and set the boot target for the new SSD drive.

Then run Linux for a bit, make sure everything works, and make an image of the already existing Win10 partition to run in a VM.

Much simpler.

Also: almost all tax software runs in a browser now. If not then it will run under Wine or Proton. I think you have less to worry about than you think.

They said they want to "learn Linux". Immutable distros are not how Linux runs in its native form, but a utilitarian way of running it for a specific purpose. You must understand the thing before you speak on the thing.