Does Qubes play well with Computer vision, and ML?

68 views
Skip to first unread message

digita...@tutanota.com

unread,
Jun 1, 2016, 1:55:37 AM6/1/16
to qubes-users
I'm a comp sci student who recently installed qubes as my sole OS. I'm curious how well it handles computer vision and machine learning, as these are the areas I plan to focus the development of my programming skills. I know that deep learning(which isn't the same thing as ML), and computer vision (in some cases) uses the gpu for 3d rendering of images.

john.david.r.smith

unread,
Jun 1, 2016, 3:27:23 AM6/1/16
to digita...@tutanota.com, qubes-users
i never tried, but you can pass your pci devices to an appvm (e.g. your
gpu).
https://www.qubes-os.org/doc/assigning-devices/

frameworks will often allow you to use your cpu in case you have no gpu.
as long as your models are small enough you can train that on cpu. (but
training a big nn (neuronal network, but i guess you know that) will
take a long time and cnns (and other deep nns) are big).

so passing through your gpu would be the best course of action.
(maybe your university provides some pcs with standard dev software and
ssh-access. you could try using those.)

-john

Arqwer

unread,
Jun 1, 2016, 5:53:19 AM6/1/16
to qubes-users, digita...@tutanota.com
AFAIK you will need to gave 2 GPUs if you want to pass one to VM. On PC you can buy one more GPU, but on laptop that might be a problem. Some laptops have 2 GPUs: one discrete, and one inside a CPU. Still, on some laptops (MSI GT72 for example), there are 2 GPUs, but they can not run simultaneously, only one at a time, what makes GPU passthrough impossible.

Btw, I am studing ML and CV for two years, and still never faced a necessity of using GPU. (May be it's just me). If there would be a case, when I will need a GPU, I can walkaround this problem by buying Amazon's virtual server with gpu.
Reply all
Reply to author
Forward
0 new messages