Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
codedokode
28 days ago
|
parent
|
context
|
favorite
| on:
Qwen3-Coder-Next
30-A3B model gives 13 t/s without GPU (I noticed that token/sec * # of params matches memory bandwidth).
yencabulator
26 days ago
[–]
Something like 21 t/s on pure CPU on a mini PC that's <2 years old.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: