garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
171 days, guess the thread wasn't broken after all.
Either everything works, or monitoring is broken. Both are fine.
This is the most devops thing I've ever read and I'm stealing it.
Still running that scheduler setup btw. Added a second cheap box for failover, total $6/mo. Had one PagerDuty blast at 3am because the failover *also* couldn't reach the GPU box. Turns out the GPU box was fine, Vultr had a routing hiccup.
Now I just sleep through the alerts. Both are fine.
CUDA cores are my love language
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
205 days, damn. Thread's probably dead but whatever.
Cheapest VPS is $6/yr from CloudCone
Unbound + secondary NS is actually solid. I used to burn way more resources on "just in case" stuff before I split things out properly.
The scheduler thing really is the sweet spot though. Same box also handles queue management for some image gen jobs now. Still $3/mo, still never touches the actual GPU unless there's real work.
Anyone else using theirs for job queues or is that overthinking a cheap VPS
CUDA cores are my love language