Lab manual for the Thornton lab at UC Irvine
Using LTS
releases is critical.
The 6 month “developer” releases can suffer from user interface lag due to the NVIDIA drivers.
We have had issues with key commands taking over one second to register in the GNOME terminal.
We have had the nouveau
drivers freeze up on us.
We’ve seen this with multiple different card of various ages.
These issues all go away (knock on wood…) with the LTS
releases.
An no, “force full composition” (in the NVIDIA X server settings) doesn’t fully solve these problems.
The procedure is simple:
LTS
release and flash it to a USB
drive.It is possible that the boot device order will need to be adjusted in the machine’s BIOS
!
Prior to such an install, it is prudent to:
/home
to one of the secondary drives.
To do so, use rsync -avr
.This section is current with 20.04
.
This is very straightforward:
sudo apt install system76-cudnn-11.1 libcusolver* python3-pip
pip3 install tensorflow-gpu
Then,
python3
import tensorflow as tf
tf.config.list_physical_devices('GPU')
The information printed to screen should tell you that the GPU have been found and “registered”.
Follow instructions here.
The command that worked is:
amdgpu-install --usecase=opencl --no-dkms