Something not covered in this video, but it is on the ReadMe of ComfyUI ZLuda is that you need to make sure that you edit your System Variable settings in Windows. The variable HIP_Path was there already for me, but I needed to add the one to the Path system Variable. Specifics are in the ReadMe. Otherwise this worked great with Flux. Thank you so much!
@fangoriously1
Ай бұрын
Thank you!
@noiseandblur
Ай бұрын
Thanks for this video, mate. I oddly appreciate when there’s a lack of cuts/editing, so you can see a lot of details that might help to figure out some issues. I really regret buying my 7900 XTX every day while trying to play with AI...
@unclemusclez
Ай бұрын
dont give up yet. there is more news to come soon.
@noiseandblur
Ай бұрын
@@unclemusclez Really, you saved my AMD card from being sold xD
@khangvinh4656
10 күн бұрын
@@noiseandblurI have an 7900xtx too. I almost try to sell mine for a 4080 but realized that I can still use it plentyful on Zluda. I’m currently using Vlad fork for SDnext dev branch and it’s pretty simple to set up, most importantly the speed is amazing.
@noiseandblur
9 күн бұрын
@@khangvinh4656 Glad to read this! I managed to handle zluda on my XTX, but it's still slow. Will give the Vlad fork its chance, thanks for the tip. Something I'm a bit worry about is the lack of support (plugin/updates/compatibility) when diving into any fork.
@khangvinh4656
9 күн бұрын
@@noiseandblur i worried about features when I first download vlad version but it seems that his version is actually extremely fleshed out as well. I’m pretty a majority of plug-in and functionality is also readily available for it. Just make sure to install the right version of python and set up the right pathing for the hip SDK. Also if the first time you generate an image and it seems like it’s stuck, give it at least 5 to 15 minutes. This only happens for the first ever generation, subsequent one is much faster. Also if you are able, it’s suggested to use Zluda in Linux or at least dual-booting Linux to try it out. The Linux version is around double to tripled that of the window counterpart (which is insane).
@MrLutijen
2 ай бұрын
Super confusing, jumping through many menues. Will somebody in the world finally create a step-by-step tutorial on how to make SD3 run properly on AMD cards?
@unclemusclez
2 ай бұрын
its not natively/officially supported. Zluda also does not technically support PyTorch, nor will it. I would not suggest attempting this if anything i did looked daunting, and perhaps just install PyTorch applications on a Linux machine which it is designed for, or wait for AMD to upgrade HIP for windows.
@khangvinh4656
10 күн бұрын
@@unclemusclezim pretty sure Linux is the most desirable for SD on AMD rn. Dual booting does wonders in this case for 3x the performance (window is still great but Linux is just next level)
@CapaUno1322
9 күн бұрын
@@khangvinh4656 I've just about to install a version of Linux so that I can install SD and use ROCm properly, do you have any advice on the best distro to use? Also how is the performance on Linux? How does it compare to native Nvidia and CUDA? It's so hard to find examples? Thanks!
@khangvinh4656
8 күн бұрын
@@CapaUno1322 I haven’t tried it yet but everyone over in vladmandic discord (this is the fork I used for Zluda) said it outrun the window version by a lot.
@CapaUno1322
9 күн бұрын
I followed the video and when I run start.bat it just flashes through a few lines of code and the CMD window closes, I don't have time to read anything in the window? Any ideas? Thanks for the video, it all installed ok as a far as I could tell? I cleared the PIP Cache as per the install page error troubleshoot....? I'm going to try Linux, I really haven't had much joy with this or SD.Next, I'm beginning to think I'm jinxed! 😟
@gkoogz9877
3 ай бұрын
Thanks. I saw sd3 was released yesterday and i was wondering about AMD compatibility
@mgwach
3 ай бұрын
So when I click Queue Prompt, it says Loading 1 new model and then I get a rocBLAS error: Cannot read... and then a path to TensileLibrary.dat: No such file or directory for GPU arch : gfx1036. I'm running a 7900XT like you.
@Alexander_rekaX
2 ай бұрын
It’s your integrated GPU. You need to play with HIP_VISIBLE_DEVICES setting
@bankmanager
2 ай бұрын
AMD documentation suggests disabling iGPUs in the UEFI/BIOS until they fix the issue.
@zygimantastauras
Ай бұрын
I can run it but I do not managed to get a batch with 6800 and a lot of other problems, And the model is very censored, and really worse than others
@nobleloren
2 сағат бұрын
This is the most confusing one so far.
@DoctorrMetal
Ай бұрын
it works better in linux
@gkoogz9877
2 ай бұрын
Honestly you should redo this video. Provide exact commands to paste in CMD in the description and tell exactly which models to download. You can mention the text encode as an aside but you should definitely not leave it in a wishy washy "well... download which one you want" without recommending one definitively.
@unclemusclez
2 ай бұрын
i hear you but please take into consideration that it's not an exact science it will greatly depend on your environment. This is not supported by ZLUDA, ComfyUI, SD3, AMD, Windows, nor NVIDIA. if you guys want to see more content, feel free to contribute to the channel. Another version of this video is coming with 3D, IMG2VID, AUDIO, TTS, and some ample workflows... once i figure out how to get it working.
@ImacojapieTV
26 күн бұрын
@@unclemusclez well still your tutorial is pure dog
@xt-cj7jg
3 ай бұрын
FIRST!
@rudigerhopperbentz7037
Ай бұрын
With all due respect for the work you put in, skip the parts where everything doesn't work and re edit...
@unclemusclez
Ай бұрын
please feel free to donate or support the channel, particularly if you have requests!
Пікірлер: 27