This probably isn't useful to anyone, and I'm not really maintaining it as an open-source project. But I thought I'd keep it public in case anyone wants to see...
Tested on Debian Buster
su
apt install sudo
usermod -aG sudo anish
Edit /etc/apt/sources.list
and add contrib
and non-free
to the end of all the lines.
sudo apt install intel-microcode
sudo apt install firmware-realtek
sudo apt install firmware-bnx2 firmware-qlogic
This firmware (bnx2) is required to use the NICs at all, so it's needed during install to configure the network. Using the Debian distribution that includes non-free firmware solves this problem (because it includes bnx2).
Look at "quick cron example".
Add ssh key to ~/.ssh/authorized_keys
.
Disable SSH password login in /etc/ssh/sshd_config
by setting
PasswordAuthentication no
.
sudo apt update
sudo apt install \
curl git mercurial vim htop iotop axel aria2 silversearcher-ag \
build-essential pkg-config \
autojump python3-pip python3-venv python3-dev \
vnstat lm-sensors rsync zsh tmux
sudo apt install python-numpy python-scipy \
gfortran libblas-dev liblapack-dev \
libjpeg-dev zlib1g-dev python-opencv
chsh -s /usr/bin/zsh
ssh-keygen -t ed25519 -a 100
Note that the -a 100
is the number of rounds of the KDF to use, so it doesn't
matter for an unencrypted key.
Add ~/.ssh/id_ed25519.pub
to GitHub account.
mkdir -p ~/src
cd ~/src
git clone [email protected]:anishathalye/dotfiles
cd dotfiles
./install
cd ~/src
git clone [email protected]:anishathalye/dotfiles-local
cd dotfiles-local
git checkout linux-server
./install
cd ~/src/dotfiles-local
git checkout gpu
./install
- NVIDIA drivers
- CUDA toolkit
- Get kernel headers first:
sudo apt install linux-headers-$(uname -r)
- Using the Ubuntu
.run
file should work - Reboot afterwards
- Get kernel headers first:
- cuDNN
- Accelerated Computing Developer Program membership required
- Copy files to
/usr/local/cuda