Javascript is not enabled. This site can still works but it'll be more interactive when javascript is enabled.
loading...
Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
jkingsman
1 day ago
|
on: Pete Warden on local AI, open models, voice interfaces, the future of hardware
I'm across a Pro 6000, 2x3090, and 128GB sys RAM (built a while back; I could not afford that these days heh). I get ~1-3ktok/s prefill, ~120tok/s decode.
When I had only twin 3090s, I made heavy use of Qwen 3.5 27b.
reply