from Hacker News

Ask HN: Is it still meaningful learning deep learning for GPU tight lab?

by spark_chicken on 7/7/24, 5:00 PM with 2 comments

Hi,

These days when I do literature reviews for the trending research, I find that most of the useful models need insane amount of GPU to train. As a student in a small lab, we are lack of GPU power. Is it still meaningful if I stay in this lab to learn deep learning??? I felt like all my work is like a toy...feeling lost, any advice?

  • by talldayo on 7/7/24, 5:10 PM

    You need GPUs to train models. Right now that's just the way it is; you can enjoy inference on weaker hardware, but training is plainly unforgiving.

    Thankfully, you can get pretty good results finetuning on a much smaller and cheaper GPU like the 3060ti. Going from GPU-poor to GPU-rich is easier than you might think.

  • by psyklic on 7/7/24, 7:40 PM

    Apparently this is true of nearly all academic labs, e.g. https://twitter.com/Yuchenj_UW/status/1789319531430912042.