You seen what local models can do? Theres no point in paying for the surveillance apparatus anymore when a 2080 TI Ventus can run models just as good as an entire data center.
In this case it’s an AMD RYZEN AI MAX+ 395 w/ Radeon™ 8060S and 128GB of unified memory, 96GB of which I’ve dedicated to being VRAM, contained inside a ASUS ROG Flow Z13, but yeah.
And it runs during max load with less power than most people’s gaming PCs use on idle.
Hell yeah bro, strix halo is a fantastic example of how we can leverage the power of contemporary open source LLMs in an environmentally friendly way.
If you roll it with Linux you can squeeze out even more VRAM - running mine with 126G available for models.
120W peak at max load. It’s nothing. The data centers are a complete sham, ruining our environment for the sole purpose of capital owners trying to corner an emerging market and accumulate more wealth. Fuck em!
Alright, I’m mad jealous. I’m gonna spin one up on my threadripper server when I can afford the power supply repairs and really test it out. But Idk how much the CPU matters.
yeah, it’s pretty nice. for an average image with SDXL it takes about 5 seconds to generate. of course, my setup is a little more advanced than that since I enjoy convincing lonely conservative men on Twitter to give me their money by pretending to be a white christian woman who likes Trump and posting racy pictures, so that’s what paid for it.
I wish I was rich enough for stuff like that tbh
pretend to be a young, white, christian Trump-loving woman who posts racy images to Twitter and you too can have some extra pocket money.
well, I don’t pretend that, I have an abliterated version of Qwen 3.6 35B loaded on my Z13 who does the pretending. I just cash in the bitcoin.
you’re right, I didn’t tag it. good catch, I sometimes forget to do that and need a nudge.
