16

6h attempting to correctly install
nvidia driver
cuda
cudnn
pytorch from source
anaconda environment

and this

Comments
  • 3
    940M is not even that old, from March 2015
    guess I'll rent a machine
  • 3
    That is fucked.
  • 3
    Try installing graphic drivers for a laptop that has two graphics cards.

    Impossible. Bumblebee fucks off, nvidia drivers dont work, linux drivers dont work, F community packages, none work. I've spent too many hours on this to count.
  • 0
    Most laptop graphics cards have kinda shitty support for cuda, or at least outdated. I guess NVIDIA never really thought that people are gonna need it that much.
  • 0
    @that-guy but this time is not nvidia's fault, pytorch is the one lacking support for my gpu :/
  • 0
    I have a Laptop with a 3rd Gen i7 and a Quadro M1200, formerly a K2100M, and both OpenCL and the latest Cuda work like a charm. Thanks to Bumblebee I can use both at the same time.
    Although those are BOINC projects running in parallel (6 on CPUs, 2 on GPUs) it still means it works.

    So what problem does PyTorch have?
  • 1
    #InstallWindows

    Seriously though, OOTB is a great thing.
  • 0
    @RememberMe yeah, until the OS stabs you in the back with weird behavior. I've had it with Windows. It has a decent OOTB functionality overall, but the blue screens, freezes, excessive memory usage, unexplicable errors in apps, bloatware, disk fragmentation, and so on pile up.

    In contrast, I often leave my linux computer on for weeks without any issue at all: current record is 35 days and I just had to turn it off for traveling. Don't get me wrong, if Windows works for you, great. For me it's not worth it.
Add Comment