Loading…
26
Rented a beefy cloud GPU for a weekend vs training on my own 3060, and the cloud lost
I wanted to fine tune a small image model last month and figured renting an A100 for two days would smoke my home setup, so I dropped about $85 on a cloud instance. Turns out the upload, the setup, and the network lag ate half my weekend, and my little 3060 in my desk tower beat it on actual time per step because there was no data moving over my crappy apartment wifi. I tried again the next weekend with the same dataset on my own card and finished in 9 hours, flat. The lesson for me was that local wins when your dataset is huge and your pipe is small. Anyone else run into this, or am I just doing cloud setup wrong?
0 comments
Log in to join the discussion
Log In0 Comments
No comments yet
Be the first to share your thoughts on this discussion.