DEV Community

Hizar Arain
Hizar Arain

Posted on

AI ON ARDEON 780M!?

Hello guys to my first guide on how to install ai libaries on a radeon 780m (this guide is also for other gfx110X gpus)

REQUIREMENTS
a gfx110X combabtible gpu

16GB is mandatory or above for bigger model i know ram is very pricey

Fedora or any distros but i prefer fedora or use windows(NO TRITON OR JAX SUPPORT)

Python (preferably a newer version) python only for windows

Good internet I MEAN IT the packages are so big because they include all of rocm

ALSO GO IN YOUR BIOS AND SET THE UMA BUFFER TO HALF YOUR RAM

ON WINDOWS PLEASE USE CMD

Hip sdk if on windows HIP SDK

Time i also mean it

basic terminal knowledge and python knowledge

GUIDE

ok first things first the commands are gonna be split one for windows one for linux indicated by "w" for windows and "l" for linux

"l"

lspci | grep -i "VGA\|Display\|Radeon"
Enter fullscreen mode Exit fullscreen mode

it should return something like HawkPoint1 or similair for radeon 780m

"l"

ls /dev/dri/
Enter fullscreen mode Exit fullscreen mode

should say card1 renderD128 or similair

"l"

groups | grep -E "render|video"
Enter fullscreen mode Exit fullscreen mode

should return render and video if not do sudo

usermod -aG render,video $USER
Enter fullscreen mode Exit fullscreen mode

LOG OUT AND LOG IN AGAIN

ok now for the venv
_**

CRUCIAL: On Linux, never install these ROCm wheels globally. By using a venv, we create a sandboxed environment where the 'heavy' ROCm libraries live peacefully without conflicting with your system's desktop drivers.

**_

create the venv

"l"

python3 -m venv venv
Enter fullscreen mode Exit fullscreen mode

"w"

python -m venv venv
Enter fullscreen mode Exit fullscreen mode

activate it

"l"

source venv/bin/activate
Enter fullscreen mode Exit fullscreen mode

"w"

.\venv\Scripts\activate
Enter fullscreen mode Exit fullscreen mode

yay we got the venv if you got here GOOD JOB

NOW for the giant package installation

NOTE PLEASE HAVE MORE THAN 5 GB OF FREE SPACE AND IF IT LOOOKS FROZEN JUST WAIT

"l"

python3 -m pip install apex filelock fsspec jax-rocm7-pjrt jax-rocm7-plugin jaxlib jinja2 markupsafe mpmath networkx numpy pillow rocm rocm-profiler rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-gfx110x-all setuptools sympy torch torchaudio torchvision triton typing-extensions --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ --resume-retries 0
Enter fullscreen mode Exit fullscreen mode

"w"

python -m pip install filelock fsspec jinja2 markupsafe mpmath networkx numpy pillow setuptools sympy torch torchaudio torchvision typing-extensions rocm-sdk-libraries-gfx110x-all --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ --resume-retries 0
Enter fullscreen mode Exit fullscreen mode

YAY IF those finished you did the longest step i wnat to just remind here DONT USE THE LINUX COMMANDS ON WINDOWS because windows doesnt have triton or flax support

NOW YOU HAVE SUCESSFULY INSTALLED ALL THE AI PACKAGE IF YOU WNAT TO WRITE YOUR OWN CODE HERE IS THE END OF LINE BUT IF YOU WNATA WRAPPER LIKE comfyui than read further

COMFYUI yay

"l" and "w"

clone the repo

git clone https://github.com/comfyanonymous/ComfyUI.git
Enter fullscreen mode Exit fullscreen mode

enter comfyui

cd ComfyUI
Enter fullscreen mode Exit fullscreen mode

now install the rest requirements this will take a while also

python -m pip install -r requirements.txt
Enter fullscreen mode Exit fullscreen mode

_HOW TO LAUNCH*_*

ok this time splits again "l" for linux and "w" for windows (USE --lowvram if you have 8gb or under 8gb uma bffer or vga if you have above it rmeove the --lowvram)

"l"

python3 main.py --lowvram
Enter fullscreen mode Exit fullscreen mode

"w"

python main.py --lowvram
Enter fullscreen mode Exit fullscreen mode

IF YOU LIKED THIS THAN LET ME KNOW IT YAYYY

Top comments (0)