@@Luinux-Tech There's one video tutorial that I would really like you to create if you have the time. How to install fluxgym for AMD GPU on Ubuntu. I really did try to install it but failed. Your tutorials are the only ones that work. And thank you again for sharing these videos.
I love you!!! Thank you a lot. The command HSA_OVERRIDE fixed my issue. I had a hipblast error when trying to run the prompt. Now i must figur out how to make it use both gpus. I have 2 rtx7900xtx
I'm glad I helped. Regarding dual GPUs, my knowledge is very limited. Maybe "SwarmUI" or "ComfyUI_NetDist" might be what you're looking for. However, I didn't see anyone saying if it worked with AMD GPUs, and I have no way of testing it.
Hi buddy, I have an RX 6800, I know it's RDNA2 not 3, just wondering how fast your iterations are with your set up and have you any ideas on making it faster with AMD? I get around 5 It/s which is ok since when I tried to get it to work on windows it was a total mess! Thanks!
When I try to generate images with 16:9 resolutions, I get the error "Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding." Any idea how to fix this ?
I understand all this instruction is hard work and i much appreciate. But wait... after all, in the end every launch from ubuntu i need write: source venv/bin/shit activate.... ? And don't tell me i need do all this halfhour procedure after every reboot.
At the end of the video, I show the command you need to run to launch ComfyUI after you have installed it. It's all very simple and quick. If you have already typed the command at least once in the terminal, just use the up and down arrow keys to navigate through your command history. This way, you don't need to memorize or type anything. If you know a little bit about shell scripting, you can also create an icon on your desktop to start ComfyUI from there.
@@Luinux-Tech "shell scripting" - will try. By the way, it all works and Comfy just flies now. Even on python3.12.7, not sure about all configurations. Thank you.
Hi buddy! Thanks again for helping me install A1111 on Linux. I was wondering do you know any methods on how to speed up the iterations on any SD version? Or do you know which version of SD is fastest for Linux? I have an AMD GPU and I'm quite happy with the speed and results but if you can make it better then obviously you may as well? I've seen a few articles talking about this but you may know? Thanks and have a fun day! ;D
The best tip I can give you is to use ComfyUI with the "efficiency-nodes-comfyui" extension. It is significantly faster than A1111. The only downside is that the interface is slightly more complex. Regarding the SD version, it depends on your hardware. In my opinion, models like SDXL or Pony produce very satisfactory results without compromising performance.
This seems to me to be a simple path error. Go to the webui folder and check if the file really exists. Also, be careful with spaces in the file or folder names. The terminal needs a "\" before every space to interpret the path correctly.
@@Luinux-Tech Yeah i get vulkan working on windows then i switched to ubuntu to try ROCm version. I though it'll boost speed gen x2 the same as ComfyUI
Hey I can't launch it. I get this error: File "/usr/local/lib/python3.11/lzma.py", line 27, in from _lzma import * ModuleNotFoundError: No module named '_lzma'
What distro are you using? Did you compile Python manually? This module "_lzma" is part of Python standard libs, so it should be included with Python. If you are using Ubuntu or Debian, try installing the package "liblzma-dev" or "xz-devel" for RHEL and Fedora.
@@Luinux-Tech thanks for the reply! I'm using Linux mint, completely new to Linux and a lot of people online said it was pretty windows-like. I followed your instructions so I'm not sure what went wrong
@@ZakiAsir Try installing the "liblzma-dev and lzma" packages on your distro (sudo apt install lzma liblzma-dev). Then, you will need to delete the venv folder and create it again, reinstalling PyTorch and the comfyUI dependencies.
I haven't tried your procedure yet, but have tried ComfyUI install for Linux AMD, without success. I wasn't even aware of using HSA_OVERRIDE_GFX_VERSION. I haven't seen it mentioned in other searches. How do I find out what peramiture to use for my particular laptop. I'm using a HP AMD Ryzen 7 PRO 5850U laptop running POP O/S. I think my GPU is a AMD Radeon RX Vega 7. BTW, in my other attempts to install ComfyUI I didn't use the venu command. I didn't see that mentioned in my other install directions. With my CURRENT installation configuration I'm getting this error, RuntimeError: No HIP GPUs are available.
According to TechPowerUP, your iGPU is a Radeon Vega 8. The GFX version has two variants. You can try using the GFX version "9.0.2" or "9.0.3". However, I am unsure if pytorch is compiled for this architecture. Additionally, please ensure that you dedicate at least 2GB of VRAM for your iGPU in the BIOS.
@@Luinux-Tech Thanks for taking the time to help out, it's appreciated. Neither GFX version worked. Looks like I'll have to install on my Windows desktop computer which has a Nvidia card.
@Luinux-Tech hii the anaconda console tells me this: Using directml with device: Total VRAM 1024 MB, total RAM 32694 MB pytorch version: 2.4.1+cpu Set vram state to: NORMAL_VRAM how can i set the vram to high? (mi GPU is 16GB Ram and the shared memory is using 1024MB)
Is there a specific reason you went for installing multiple versions of python natively through a third party PPA rather than using a conda environment for this purpose?
These are recommendations from the devs themselves. Python versions 3.10 and 3.11 are supported. Using different versions may result in errors in modules and extensions.
dear fiend thank you for such step by step information it would be very helpful if you can guide me to install Flux models and SD3.5 models in Comfy UI Thanks a lot
Sorry, I don't have knowledge of dual GPU workflows, but a single RX 580 is enough to generate pretty decent images. If you're interested, here's a video that might help you: th-cam.com/video/mpdyJjNDDjk/w-d-xo.html
Finally a tutorial that works and it's a gazillion times easier to follow. Thank you very much.
Glad it helped
@@Luinux-Tech There's one video tutorial that I would really like you to create if you have the time. How to install fluxgym for AMD GPU on Ubuntu. I really did try to install it but failed. Your tutorials are the only ones that work. And thank you again for sharing these videos.
Big Thank You, straight to the point and works like a charm !
I love you!!! Thank you a lot. The command HSA_OVERRIDE fixed my issue. I had a hipblast error when trying to run the prompt. Now i must figur out how to make it use both gpus. I have 2 rtx7900xtx
I'm glad I helped. Regarding dual GPUs, my knowledge is very limited. Maybe "SwarmUI" or "ComfyUI_NetDist" might be what you're looking for. However, I didn't see anyone saying if it worked with AMD GPUs, and I have no way of testing it.
this is really underrated, now i can do batch of 4 832x1248 SDXL images on my RX 7600 8GB using tiled VAE, without using --medvram or --lowvram
Hi buddy, I have an RX 6800, I know it's RDNA2 not 3, just wondering how fast your iterations are with your set up and have you any ideas on making it faster with AMD? I get around 5 It/s which is ok since when I tried to get it to work on windows it was a total mess! Thanks!
Thank you very much! I will give it a try.
When I try to generate images with 16:9 resolutions, I get the error "Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding." Any idea how to fix this ?
Actually never mind. I just needed to add the --highvram flag. Thanks for the tutorial.
I understand all this instruction is hard work and i much appreciate. But wait... after all, in the end every launch from ubuntu i need write: source venv/bin/shit activate.... ? And don't tell me i need do all this halfhour procedure after every reboot.
At the end of the video, I show the command you need to run to launch ComfyUI after you have installed it. It's all very simple and quick. If you have already typed the command at least once in the terminal, just use the up and down arrow keys to navigate through your command history. This way, you don't need to memorize or type anything. If you know a little bit about shell scripting, you can also create an icon on your desktop to start ComfyUI from there.
@@Luinux-Tech "shell scripting" - will try. By the way, it all works and Comfy just flies now. Even on python3.12.7, not sure about all configurations. Thank you.
Hi buddy! Thanks again for helping me install A1111 on Linux. I was wondering do you know any methods on how to speed up the iterations on any SD version? Or do you know which version of SD is fastest for Linux? I have an AMD GPU and I'm quite happy with the speed and results but if you can make it better then obviously you may as well? I've seen a few articles talking about this but you may know? Thanks and have a fun day! ;D
The best tip I can give you is to use ComfyUI with the "efficiency-nodes-comfyui" extension. It is significantly faster than A1111. The only downside is that the interface is slightly more complex. Regarding the SD version, it depends on your hardware. In my opinion, models like SDXL or Pony produce very satisfactory results without compromising performance.
Would my 8gb vram of my 7600 be enough ?
Yes, it will work perfectly.
it worked for me with python 3.12
It doesnt work with 6800xt i had to get rocm 5.7 to make it work but great tutorial
Thanks, the RX 6800XT uses exactly the gfx1030, so I don't know what could have happened. Anyway, I'm glad you got it.
i get this any fix?
python: can't open file '/home/ghost/Desktop/sd/stable-diffusion-webui/main.py': [Errno 2] No such file or directory
This seems to me to be a simple path error. Go to the webui folder and check if the file really exists. Also, be careful with spaces in the file or folder names. The terminal needs a "\" before every space to interpret the path correctly.
Make oobabooga text-generation-webui/koboldcpp video pls. I can't run llm GGUF models on my rx 6600
Have you tried this: th-cam.com/video/sGopcVfgfk8/w-d-xo.html It uses Vulkan, and from what I've tested, it has very good speed.
@@Luinux-Tech Yeah i get vulkan working on windows then i switched to ubuntu to try ROCm version. I though it'll boost speed gen x2 the same as ComfyUI
@@sundowner6191 Okay, so you want to use KoboldCPP with ROCm. I got it. I'll make a video soon.
@@Luinux-Tech hey i bought nvidia gpu so don't bother yourself
@@sundowner6191 Nice, you'll definitely have a better experience with AI tasks on NVIDIA GPUs. Good luck.
Hey I can't launch it. I get this error:
File "/usr/local/lib/python3.11/lzma.py", line 27, in
from _lzma import *
ModuleNotFoundError: No module named '_lzma'
What distro are you using? Did you compile Python manually? This module "_lzma" is part of Python standard libs, so it should be included with Python. If you are using Ubuntu or Debian, try installing the package "liblzma-dev" or "xz-devel" for RHEL and Fedora.
@@Luinux-Tech thanks for the reply! I'm using Linux mint, completely new to Linux and a lot of people online said it was pretty windows-like. I followed your instructions so I'm not sure what went wrong
@@ZakiAsir Try installing the "liblzma-dev and lzma" packages on your distro (sudo apt install lzma liblzma-dev). Then, you will need to delete the venv folder and create it again, reinstalling PyTorch and the comfyUI dependencies.
What about Windows?
I haven't tried your procedure yet, but have tried ComfyUI install for Linux AMD, without success. I wasn't even aware of using HSA_OVERRIDE_GFX_VERSION. I haven't seen it mentioned in other searches. How do I find out what peramiture to use for my particular laptop. I'm using a HP AMD Ryzen 7 PRO 5850U laptop running POP O/S. I think my GPU is a AMD Radeon RX Vega 7. BTW, in my other attempts to install ComfyUI I didn't use the venu command. I didn't see that mentioned in my other install directions. With my CURRENT installation configuration I'm getting this error, RuntimeError: No HIP GPUs are available.
According to TechPowerUP, your iGPU is a Radeon Vega 8. The GFX version has two variants. You can try using the GFX version "9.0.2" or "9.0.3". However, I am unsure if pytorch is compiled for this architecture. Additionally, please ensure that you dedicate at least 2GB of VRAM for your iGPU in the BIOS.
@@Luinux-Tech Thanks for taking the time to help out, it's appreciated. Neither GFX version worked. Looks like I'll have to install on my Windows desktop computer which has a Nvidia card.
@Luinux-Tech hii the anaconda console tells me this:
Using directml with device:
Total VRAM 1024 MB, total RAM 32694 MB
pytorch version: 2.4.1+cpu
Set vram state to: NORMAL_VRAM
how can i set the vram to high? (mi GPU is 16GB Ram and the shared memory is using 1024MB)
Is there a specific reason you went for installing multiple versions of python natively through a third party PPA rather than using a conda environment for this purpose?
No, for the purpose of the video, a simple venv is enough. You can use conda if you want.
Do you really need to use Python 3.11 for this?
These are recommendations from the devs themselves. Python versions 3.10 and 3.11 are supported. Using different versions may result in errors in modules and extensions.
dear fiend thank you for such step by step information
it would be very helpful if you can guide me to install Flux models and SD3.5 models in Comfy UI
Thanks a lot
Sure. I'll research how to do it, and as soon as possible I'll release the video.
it support on rx5700xt?
I can't guarantee it, but try using the flag "HSA_OVERRIDE_GFX_VERSION=10.3.0".
Hi bro i have two rtx580 which sum up to 16 g vram what should i do???
Sorry, I don't have knowledge of dual GPU workflows, but a single RX 580 is enough to generate pretty decent images. If you're interested, here's a video that might help you: th-cam.com/video/mpdyJjNDDjk/w-d-xo.html
@@Luinux-Tech Thanks for your Nice Video ,i watched the linked you shared but Do you have any video For Polaris gpus comfyui running in Windows?
@@hphp8347 Yes, you can watch this video: th-cam.com/video/W75iBfnFmnU/w-d-xo.html. In it, I use a Polaris GPU on Windows 11.
New install zluda amd for win