Hacker News new | ask | show | jobs
by gonyea 2602 days ago
It's amazing that Apple seems to be holding on to their grudge against Nvidia... to their own detriment.

You haven't been able to buy a Mac/Macbook Pro with an Nvidia chip for many years now, and using one in an external enclosure is iffy. So data scientists are increasingly using Linux to do their work.

4 comments

Are the embedded NVidia chips really that great for DL? I have a GeForce 150 in my laptop that I keep permanently disabled in Linux to save a couple watts of power. Aside from the GPU generally being weak, my battery would only last for a couple hours if I cranked my CPU/GPU near 100% usage vs 10 hours just editing files via wifi.

Almost all my work requiring a GPU is done by either ssh'ing into my desktop with a GTX 1080 or on AWS. And in that scenario, Linux has less friction (I can migrate setups between home and AWS) and is significantly cheaper than a MacPro.

I have a System76 Linux laptop with a 1070 and it works fine for deep learning. I used to SSH into cloud services but decided having a GPU always available locally was worth a little money.
Even if you use an external TB3 enclosure (it's not difficult) MacOS doesn't support Nvidia cards from Pascal upwards, and seems they are actively blocking Nvidia from releasing working drivers. They are absolutely shooting themselves in the foot with this.
TBH there isn't much of a point on having an Nvidia chip if Nvidia isn't also allowed to provide proper drivers for macOS and has to rely on Apple's shoddy implementation of pretty much everything.
NVIDIA provided drivers themselves including a CUDA runtime they still do for High Sierra and older OSes, with Mojave tho Apple locked them out.
What does this comment have to do with the article? There are no mentions of Apple or Nvidia..
>"We have seen companies signing up for Linux desktop support, because they want to have fleets of Ubuntu desktop for their artificial intelligence engineers."

Those engineers would be coming from Macbooks due to weakened support for Nvidia CUDA applications.

Or perhaps they want to use Linux for other reasons? The point is the article does not say what their motivation is, so GP comment is off topic at best.