Install Stable Diffusion on Ubuntu for Polaris GPUs(GFX803)
Вставка
- Опубліковано 8 вер 2024
- I just saw that ROCm is now available for Ubuntu 20.04. I'm already compiling the binaries for GFX803, and I'll soon make a video for this latest version.
In this video, I show how to install Stable diffusion on Ubuntu for Polaris GPUs (e.g. RX 580, RX 570...).
All files and commands are in this github: github.com/Lin...
Hello! Very informative and useful video, it helped me a lot, thank you for the instructions. Is it possible to make the same video for Fooocus with Ubuntu 22.04?
Wouldn't MESA already have the rx580 driver automatically?
As far as I know, you need amdgpu-pro to use ROCm.
I'm sorry, but according to this instruction, what you did, the system constantly reboots after starting the generation, what could be the reason? Video card RX 470 8 GB
This type of task demands a lot from the GPU. Check if your GPU is at an acceptable temperature. I recommend, after the system reboot, running the command "sudo journalctl -b -1" to see the logs from the previous boot and possibly the reason for the reboot.
@@LinuxMadeEZ I did all this even before I wrote to you, this is some kind of error due to the connection between my video card and the amd rx 470 +Intel Xeon processor causing errors in the hve kernel
@@LinuxMadeEZ this error is not even stored in the log, so I am asking you, maybe you have encountered such a problem? If I find a solution, I will write it in the comments, maybe someone else will need it. And thank you for the good content. And do you plan to make a similar video for arch similar systems?
What generation is your Xeon? If your CPU does not have the AVX2 instruction set (Intel Ivy Bridge and older), PyTorch will not work. You will need a specific build with AVX2 disabled in the compilation. Regarding Arch, the problem is that there are no official repositories and the available ROCm is already in version 6, which removed the GFX803 arch from the binaries. An alternative would be to install it through Docker.
@@LinuxMadeEZ This is how I understood it, it will be necessary to compile pytorch from raw materials without AVX2 support. Thanks for the help, I will try.And how will the experiments be at the expense of Windows and ZLUDA, if everything is so bad with Arch?
I'm using fedora, which video should I refer back to ?
The closest guide would be the Alma Linux one, but Fedora is not officially supported, so there are no official ROCm repositories for it. As far as I know, it is possible to install ROCm-HIP, but only on Fedora 39 (pending for Fedora 38). If you are really interested, I strongly recommend using one of the supported distros (RHEL 9, Ubuntu 22.04, or SUSE 15).
@@LinuxMadeEZ Thank you but I'm not changing my distro for this. but I like how you're engaging with people and genuienly looking to help others. I'm subbed for future vids.
i have rx 580 8gb please guide me to install it on windows 11
Sure, in a few days, I'll release the video.
Your video is out. hope it helps
I've done everything like in your guide and still "Torch is not able to use GPU"
Ubuntu 22.04, Radeon RX 570 4 GB.
The only difference that I was able to spot is that after typing "dkms status" I had:
"amdgpu/6.7.0-1781449.22.04, 6.5.0-45-generic, x86_64: installed"
While in yours video you have "6.5.0-41-generic". The rest is the same.
Go to the bin folder of your venv and type in the terminal "./python3", then type "import torch", and finally "torch.cuda.get_device_name(0)". What is the output of this last command?
@@LinuxMadeEZ "Runtime error: No HIP GPUs are available"
If you run the command "/opt/rocm-5.7.3/bin/rocminfo" in the terminal, does the output show your GPU?
@@LinuxMadeEZ Nope. It only outputs Agent 1: Intel Core i7-6700k
Device type: CPU.
Search for the exact model and brand of your GPU on the TechPowerUp website. On that website, go to the GPU notes section. Check what it says in the "Graphics/Compute" option.
In Ubuntu generates images faster than Almalinux??
No, from what I tested, it is the same speed.
Where do I find these 3 files to download? Torch ,torchvideo ,torchaudio
Video description, in the github link.
@@LinuxMadeEZ the files I refer to are from step 6, these 3 files in the pytorch installation! (torch-2.3.0) (torchvision-0.18) (torchaudio-2.3.1) are not on github
@@LinuxMadeEZ I've already looked at the link you left on GitHub, but these 3 torch files aren't there for download.
you need to access the "releases" tab, here is the direct link: github.com/LinuxMadeEZ/PyTorch-Ubuntu-GFX803/releases