• using nvidia for cuda only with optimus


    I have an optimus laptop with intel hd4000 and nvidia gt640m graphics. Like many, I’ve had loads of trouble with these drivers before.

    I want to use the hd4000 for all actual graphics and I want to use the gt640m for cuda computation only (namely tensorflow). I also want to be able to turn off the gt640m when I’m not using it (most of the time) and not boot up to a blank screen.

    My bios gives options of discreet graphics or switchable. I can’t disable the hd4000. This gives instructions which sound like what I want but they are for ubuntu so I was wondering if anyone here could advise me of anything I might have to do differently.

    My other option is to follow the bumblebee tutorial on the antergos wiki. Does anyone know if bumblebee works well with tensorflow and cuda?

    Many thanks.

  • The article you mention expalins not only how to install Bumblebee and proprietary Nvidia driver, but also how to completely uninstall them.

    If Bumblebee is not what you’re looking for, simply remove it. Or restore the system from a backup, made right before Bmb installation.

  • haha I underestimated the wiki. It all worked perfectly. Now I need to install cuda, cudnn and the frameworks. thanks @just

  • Brilliant! cuda, cudnn, tensorflow and keras all work out of the box from the arch repos, although I had to install werkzeug for keras

  • @Deanodirector Very well. I never had a CUDA card, and wasn’t sure that Bmb works well enough on them. Your post is important. It is (at least) the third confirmation that Bmb works well on CUDA. Thanks.

    I’ll add your GT 640M to the list of cards. on which Bumblebee was tested and works. Changing topic’s status to Solved.

nvidia99 cuda2 optimus23 Posts 5Views 171
Log in to reply