this post was submitted on 05 Jun 2026
14 points (85.0% liked)
Fuck AI
7069 readers
2748 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Not at all.
It's a Gigabyte AI Top PC. It's one of several such mini-PCs based on the Nvidia Blackwell GB-10 processor, in the emerging category of "personal AI supercomputer". It's not cheap but it's not out of reach either.
I got my company to buy me this thing on the argument that the company shouldn't use cloud services if possible on our sensitive code, and American cloud services even less, for obvious political and moral reasons. I argued that we should investigate the capabilities of this machine - or one like it - either to use as a shared AI server, or as an individual machine for all the developers here to use as locally-hosted AI coding assistants. My pitch made sense to my boss, so the company bought me this machine for me to investigate.
I've been using it since Monday for development and other things, such as translations, computer vision, general chatty thing, but also for traditional applications such as CAD, Blender, video editing and gaming too ๐ I've been trying different models for different things, see how they perform, how snappy they are... I'm impressed enough that I plan on buying one for me and my wife at home. That one or maybe a slightly cheaper equivalent based on the same processor / memory, but without the connectors I don't need. It costs a little over 5 grand.
Cool, thanks for the info. When I saw the memory bandwidth for the integrated systems like these, I thought perhaps it wouldn't be worth it over a custom build with a couple of 5090s (2.5x the price before the memory price hike). What models have you thrown at it? What's your tokens/second? The local-sized models keep getting more capable.
My potato 2060 plus 64GB of system RAM is enough to crawl along with some of the MoE models, but I've got projects that need way more compute (would like to content filter my torrent crawler database, finetune of one of the small Qwens to produce my novel programming language, ML experiments that currently take days/weeks to iterate, etc). Maybe an integrated system like yours is worth it after all.