I get about 45 t/s after some tweaking, getting the thermals under control helped a lot; mainly by reducing the power consumption to 261 W (and downclocking a little bit).

What vision model do you use?

I use Qwen 3.8 27b Q4 with 100k context (llama.cpp and the Mistral AI harness) on a 24 GB AMD Radeon 7900 XTX and it's been great. It can handle pretty much anything I throw at it and I can let it work indepedently overnight / over days on long tasks.

Try Cfait ;) (CalDAV / local task manager with VJOURNAL support)

Dealing with multiple computers is a pain (slow boot time and/or more power consumption and multiplied effort) and more expensive in most cases (you still need the minimum requirements of each part, but now there's the rest of the each PC to get too).

I did it for power consumption; one small low power computer that's on 24/7 for NAS + providing some services like home assistant, and a powerful computer for gaming, photo development, etc. I ended up being too lazy to boot up the powerful computer most of the time so I was doing less of what I like to do with it and my time spent on the computer was slower. I switched back to one do-it-all computer and I'm much happier.

The only reason I would want a multi-systems setup is if I need to run some things in the background that would make the computer unresponsive or unusable for some interactive tasks (e.g. local AI inference or training, video transcoding, heavy data transfers, ...).

Nazi scum

Nazi scum

Nazi scum

No one wants copilot because it's highly unpleasant hot garbage. There is definitely a market for AI for the competent providers.

Old high end phones are much better than new low end even after a few years. They are definitely not getting high end components.

Yes :-) thank you!

Not exactly that but Cfait (which I develop) is a CalDAV task manager (which has an option to create calendar events). I added a time tracking feature so you can add / manage sessions (which show up in the calendar if you chose to activate it) and you do get a total time.

3.8 MB, wtf did you do to this comic? This should have been less than 100 420 kB.