I should mention that @fastdotai really makes a point of making this very clear, and they do a stellar job of teaching you how to be effective in the low budget realm.
Show this thread
You can actually do awesome state of the art work with humble resources! DeOldify models are trained on a single GeForce 1080TI. I make it a point to try harder on optimization if I’m tempted to get more hardware. It has worked out wonderfully.https://twitter.com/rctatman/status/1148648091345543168 …
Estimates of the training cost (in US dollars) of a few SOTA NLP models. 
https://link.medium.com/hlu71MTTRX
https://link.medium.com/hlu71MTTRX -
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
The 1080Ti has been amazing, it's fast enough for most things, I just wish they'd release another consumer card with >11 gigs of memory. When I just full-stop can't run something is the only time I feel limited by it.
-
32GB in consumer cards would be so good - although that's likely 5 years away
- Show replies
New conversation -
-
-
Not sure I would call a 1080Ti humble tbh. It costs like 1300$ up front!
-
That’s not the normal price. I bought mine for way less than that.... That being said, yes, “humble” is definitely relative. Relative to what the article is talking about- definitely
- Show replies
New conversation -
-
-
How long does training DeOldify take?
-
These days- about 5 days with my latest models. I’m extremely skeptical it’s actually necessary so I’ll look into cutting that down soon.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.