There’s a lot to cover here but I’ll try to touch on each point:

The key requirement is fast memory that can be addressed by your GPU, and ideally a lot of it - hence the insane cost of this hardware right now.

Remember that you need space for the model’s weights (think of this as its ‘knowledge base’) and the context window, which is basically the data needed for the LLM to keep track of your current conversation with it (effectively its short term memory).

With smaller pools of VRAM (8-16gb) you will have to compromise and either have a more capable model that will lose context quickly and start hallucinating, or a less capable model that can maintain a session for a bit longer but overall less ‘smart’.

For software - there are a couple of options for running the LLM itself, Llama.cpp is one of the more popular tools and is the one that I use. It has a web UI with the usual chat interface, and also exposes an API that you can plug other tools (e.g. opencode) into, depending on your use case.

In terms of hardware recommendations, at 20GB+ of VRAM you do have a bit more headroom compared to more consumer grade GPUs, but to be honest the most cost effective way to get a shitload of VRAM is likely not with a dedicated GPU but actually using a system based around a recent APU.

I got a Minisforum MS-S1 last year for exactly this purpose. It is based on AMD’s Strix Halo platform which it has in common with the Framework Desktop and a couple of other similar devices.

It has 128gb of unified RAM which can be divided between the GPU and CPU however you like, so plenty of capacity for even fairly chunky models. It also uses a tiny amount of power compared to a more traditional system with a dedicated GPU, while also giving really reasonable performance for most AI workloads, more than enough for use in a homelab.

For cloud rental - doable, but pricing is a factor, and of course this will not actually be running locally.

Usability - manage your expectations, but overall for a lot of use cases and of course depending on the model that you are running and the resources you throw at it, it can be comparable with especially older iterations of ChatGPT, Gemini etc.

But remember, you are not a Google or an Anthropic and do not have an infinite pool of compute to throw at your model, nor do you have access to the specific models they are using.

I’m not sure if this is triggering either thalassophobia or submechanophobia but either way thanks I hate it

They have. The Doorbell Lite is $99.

I’ve switched both my laptop and desktop over to Linux (Bazzite and Fedora respectively) in the last 6 months.

The last time I tried to daily Linux (over a decade ago) I ended up switching back eventually, but this time I really don’t think I’ll need to. All of the games I play most often work perfectly, the dev tooling is even better than it is on Windows, and the hardware compatibility side has been completely flawless.

Gone are the days of having to hunt down obscure Linux drivers for your touchpad or webcam. Everything just works out of the box.

Well, I’m currently writing a service and frontend, both in C# (Blazor for the UI), and using docker-compose to build and deploy them to a Raspberry Pi running Linux. So not only cross-platform, but cross-architecture as well.

This is not a new thing either. Since .NET Core was released almost 10 years ago, it has supported cross platform development.

We refer to it as kew-bee-cuttle

I have a IIIXe (very similar to the one in OP) somewhere. Really limited in what it could do, but very cool for the time. I also have a later model Zire somewhere that had enough horsepower (with a mild overclock) to play Quake.

Yeah I can imagine it taking time getting used to as a new player. I played the original as a kid on DOS so the UI is deeply ingrained lol.

Baldur’s Gate 3 (~600 hours), BeamNG.drive (~550), Cities Skylines (~300), Space Engineers (~300), 7 Days to Die (~250) and Satisfactory (~230).

These are all stats from Steam and probably not fully representative. Satisfactory for example I used to play on Epic when I got it as a free game over there, probably logged at least another 500 hours or so on that platform.

My most played game of all time is most likely TES: Oblivion, which I started playing at release back when I was a teenager and had almost infinite free time. I’m not sure if I still have my oldest save to confirm, but I suspect it would be at least 1,500 hours, probably more across several characters.

OpenTTD.

I decided to set up Fedora on my new laptop as it was either take a chance on that or spend like 3 hours debloating a Win11 install.

It’s been over 10 years since I last tried dailying Linux, we have come a long way in that time. Everything just worked out of the box. No fucking around needed.

Even relatively niche stuff like my thunderbolt dock and the laptop’s fingerprint sensor was picked up. And, thanks to the investment Valve has been putting into Wine and Proton, pretty much every game I’ve tried has worked with no issue.

Next time my desktop is due for a clean install I’ll definitely be doing the same there.

Not at all.

Lemmy is overwhelmingly militantly anti-Tesla, which is understandable considering who owns it, but it does mean that users tend to interpret any neutral or factual statements (basically anything that is not outright criticism) as having a pro-Tesla bias.

In this case, all I am stating is the fact that this specific change currently only affects corporate users. That could of course change in the future.

There is a rich history of cloud based data providers pulling the rug from under users with no warning. Look at what happened to Nest users when Google took over.

There is most likely an overlap on what you can get from the OBD port, but generally speaking the API will provide more high level info e.g driving status, mileage, live location - and the OBD port will provide more low level data e.g. detailed battery stats from the BMS, energy usage, etc.

Highlight where in the above post I am defending anything.

Something to note: Tesla has two vehicle APIs, the Fleet API for commercial accounts and the Owner API for individuals. This change currently only impacts the Fleet API.

If you are an individual owner who accesses your vehicle data from the Owner API (usually via a self hosted tool like TeslaMate), this does not affect you. Yet.

How often do you need to travel the entire range your car allows?

If you do need to drop everything and drive across country in an EV, you should be stopping at service stations to do short fast charge sessions anyway, as with modern fast chargers and battery tech you will typically go from something like 30% SoC to say 70% in only a few minutes. This saves a lot of time on longer trips.

If you are driving an EV by depleting the battery completely and then charging it back to 100% every time, you are doing it wrong.

That also means we can still use the expansion cards for the Framework in any other device that also has a USB-C port. Need an SD card reader or a 2.5Gb LAN adapter? Not a problem, I’ll just grab one from my laptop.

Solution: don’t read that shitrag. It was always a waste of paper, now it is a waste of bandwidth as well.

Not exactly crazy but just mysterious…this was at a software company I worked at many years ago. It was one of the developers in the team adjacent to ours who I worked with occasionally - nice enough person, really friendly and helpful, everyone seemed to get on with them really well and generally seemed like a pretty competent developer. Nothing to suggest any kind of gross misconduct was happening.

Anyway, we all went off to get lunch one day and came back to an email that this person no longer worked at the company, effective immediately. Never saw them again.

No idea what went down - but the culture at that place actually became pretty toxic after a while, which led to a few people (including me) quitting - so maybe they dodged a bullet.