How to Install Stable Diffusion on AMD GPUs (NEW)

[@LuckThorfin]

Valeu!

[@TrentKingdom]

Thank You!!!

[@3starsfell]

I kept getting the same message , didn't know where the venv files are. So I deleted Python. Went to the python download page on the Python site. After it was deleted off my PC. Then I downloaded version 3.9.0, and set this as my new path when prompted to upon installing. Then went to the webui and double clicked. Works perfectly I'm now in sd

[@andrealvesf]

como resolvo essa questão de rumtimeerror ? você pode me ajudar

[@HosannaSookra]

The runtime error, maybe you need Microsoft redistributables 14 and up

[@zygimantastauras]

Thanks! Very happy to share with you my joy!

[@_scroller]

To anyone getting a runtime error about Torch not using the GPU, go edit the webuser.bat file and add this. set COMMANDLINE_ARGS=--use-directml --disable-model-loading-ram-optimization --opt-sub-quad-attention --lowvram --disable-nan-check Trent Kingdom can you plz pin this so others can see

[@kori7795]

THANK UU

[@wenbozhao7427]

BRO YOU ARE THE FUCKING 🐐🐐🐐🐐🐐🐐

[@Freelancer177]

THANK U SO MUCH OMGGGG

[@IceLom]

@@TheFRBublitz thanks so much this worked for my 6800xt it is using the the GPU. The only difference is that i only added --use-directml to COMMANDLINE_ARGS= as i did not want to run in low vram mode. MY gpu usage hits %76 while generating.

[@hoangduong2065]

@@TheFRBublitz pls help me, I stuck this Failed to import transformers.modeling_tf_utils because of the following error (look up to see its traceback): No module named 'keras.__internal__' after making from you

[@jarod1231]

FYI if you initially ignored the python versions and just installed the newest version assuming it would work.. then attempted to uninstall the newest version of python and downgrade to the correct version you will start getting an error saying "venv python not found" or something along those lines... the fix is to delete the contents of the "venv" folder and run webui-user.bat again... it will forget that you had the wrong version of python installed and work after that

[@antylife]

Love you babe

[@RomboDawg]

THIS, THIS IS THE FUCKING BEST COMMENT. This is the only comment that fixed all my problems ❤❤

[@thomasmoreau8297]

​@@RomboDawgI don't even have an AMD card but I wanted to recognise a solid comment with a tactical upvote

[@1.618_Murphy]

This should be the pinned comment. I myself also learned it the hard way on Windows 10.

[@MuNaSu]

​@@RomboDawgI don't recommend doing that if you have a lot of files within the "venv" folder which stable diffusion puts in automatically to run properly. A much much better alternative is to go within the "venv" folder and delete the "script" or "scripts" folder (Not sure if it has an S or not lol). This is where python itself is initialized in SD.

[@Eternal-Woe]

You absolute chad, thanks for the guide to get this working on amd. For anyone who may have issues with python being detected you'll want to edit: (set PYTHON=) and direct the path to the python.exe I had the same issue for my nividia install

[@DarthLeopard]

Thank you!!

[@pdaj]

Thank you so much this worked for me!

[@Max77189]

you are a gentleman and a scholar

[@007agentas]

Which RX model do you have?

[@toyjesus]

I can't thank you enough for this. i was running stable diffusion on my CPU until now and it took ages for any images to load. everything seems to work so far, I occasionally get weird errors about the model failing to load for some reason though.

[@jonnym4670]

from my understanding you can't use normal purned models on amd is that true how did you get your models loaded

[@TrentKingdom]

If you have any errors make sure to join my Discord server - discord.gg/M5SwEXPCHq

[@ls_topher1688]

Thank you very much for the video, a question: normally, how long does it take to generate an image at 768x768 with 30 steps?

[@TrentKingdom]

@@ls_topher1688 All depends on your PC specs!

[@spikethetarget4799]

@@TrentKingdom What kind of amd card are you using??

[@anonmiswolf8193]

i was trying to install stable diffusion but it keeps showing this error code AttributeError: module 'torch' has no attribute 'storage'

[@jonnym4670]

mine keeps saying unable to load model no matter what models i use

[@brandonkelly791]

Hello thank you very much for your help. I just bought a AMD GPU and didn't know they weren't supported and I downloaded stable diffusion and it didn't notice my GPU. I found your channel and I have stable diffusion running on my system now thanks to you. I will definitely be watching your channel on how to use Stable Diffusion.

[@thundervulpes]

Actually worked for me!! Thank you so much!!!! Your video was very professionally well done and easy to follow. Good job! :)

[@nightfr09]

Absolute chad, thanks! I normally draw, but i think using AI will help me generate images for my D&D campaign much faster and more aesthetically than what I'm capable of doing on a weekly basis :)))))

[@botarakutabi1199]

Exactly same here! Been mostly trying to use stable diffusion for backgrounds and elements I'm not super excited to draw. I'm running a campaign using Low Fantasy Gaming where the party starts and maintains a cult in a town I made.

[@paulw7738]

Thanks, installing now! Can't wait to try on 23.5.2 drivers - AMD reporting 2X performance with DirectML for this driver version

[@Chino45A]

2 days searching for this tutorial thank you !

[@shushkomrinmoy2483]

Thanks A lot man, working nicely and fast in my Ryzen 5 with Asus Radeon Rx 590 8GB.

[@angelrosso7685]

Thank you very much. My trip to using stable diffussion with an AMD card has been exhausting. Several ubuntu installations, and a lot of tutorials without success. This finally worked!! If that was not enough, even the stable diffusion model you suggested is great. Thanks a lot!!!

[@SpacenSpooks]

". My trip to using stable diffussion with an AMD card has been exhausting." EXACTLY the same!

[@bhootpurvmanusya]

imagine what u did and some more part extra for rdna1 gpus

[@watersports34]

OMG love you. After 3hours searching a tutorial that works

[@TrentKingdom]

Haha np man!!

[@littlebigcommentaries9833]

Just a couple things to add. Using Python 3.11 will not work. It will send back an error. I found success using 3.10.9 and make sure to select install to PATH which is not mentioned in the video. If you do not select this you will receive a "no python found" error.

[@BulbOfSalad]

3.10.6 is working also so more options cheers

[@ItsCiRO6]

Thanks a lot man!

[@mechamochi9627]

Bro your video actually is so clear and concise, and helped me through this complicated process. Thank you for creating it, and also thank you to the helpful comments leaving tips on how to solve specific errors. Everyone here is goated 🐐🐐🐐

[@obesePETE]

Legend! I spent all day trying to install the version for NIVIDIA. was bashing my brain against brick. excellent tutorial. number 1 best excelent video.

[@kempart8991]

same :c

[@jokersbin]

Thank you! I like your calm, educative and concise voice.

[@greg_richrd]

thanks man! worked fine for me with my amd 6700XT!

[@schlechtestergtaspielerdek3851]

got an 6700xt aswell getting the error with no cuda found though

[@hdawod]

Hey man, I came here especially to say thanks a million.

[@fllomasta]

This really did work for me. I tried everything in this video BUT without actually watching the video or googling it. I got errors then watched the video, done everything again and it worked. Am I stupid? Thank you, man, saved the day.

[@lenoirx]

"RuntimeError: Torch is not able to use GPU" any help? I'm using an RX 5600 XT

[@cjimene5767]

Holy cow! Using Stable Diffusion with an ancient AMD R9 380!!!! Much faster than rendering using the CPU. Thanks for the tutorial!

[@raffygarnica7995]

Thank you man! after a couple of days of searching for a fix, it finally works! it went from 10-20 minutes of generating to about 50 seconds on my RX570!

[@GamingGD5600]

Do you think this will work for a 500?

[@Max-cg5fp]

Dude I've spent like three hours trying to fix repetitive python and pip errors but it finally seems to work :D

[@darknessblades]

Thank you for the amazing tuturial. Was looking on how to get it working on my RX7900XT Currently installing, and will test it out soon

[@granatoff3917]

me too..installed it a few ago!

[@zer0xll]

ever get it to work on that 7900XT? I was on 5700xt running just fine but upgraded to a 7900xt no matter what i get error of RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check, which make it just use the CPU

[@MarySmirnova-u1e]

@@zer0xllI own 7900xtx, it opens the browser but doesn’t allow to generate any images because of “runtimeerror found no nvidia driver on your system” did you have any experience with that?

[@Yenyen666]

@@MarySmirnova-u1e same here with 6800XT, the skip cuda argument makes it slow because it runs on CPU, can't use my GPU properly yet

[@enemystandu296]

When i try to run the .bat file, in cmd appears the following message: "Couldn't launch python exit code: 9009". And when i run in administrator the cmd just desappears and don't run.

[@Acuate18777]

same here.

[@Locosiq]

same

[@sandilekhoza798]

same issue as well

[@lukaswagner8615]

I had the same problem and i installed python from Microsoft Store. Now it works

[@Digital_Rust]

when you install python you need to check "add to PATH".

[@godakai1080]

Finallt got it working thenks to you, good video. Also if some of you are getting trouble whit the text, copy it straigh from the link in the dresciption

[@samoed998]

You explained it very clearly and concisely, thank you.

[@quini7]

Worked for me, every step is very easy to follow. Thank you a lot!

[@zoricgames]

This was so easy to do thanks to your video, thanks man!

[@dede_gibbs]

I encountered many problems in my attempts before this video. This video has become my hero, everything may sound the same but I think the biggest problem is downloading "stabble diffusion" to a random directory. In this video, since we downloaded it to our main user's directory, we can easily download and unpack packages, which probably require many permissions, via cmd, so the application runs on amd gpu without any problems.

[@BrainBITEdot]

This is legit the most helpful video I've ever watched - thanks man!

[@Nukhebuk]

Absolute king! after trying for over half a year, I can finally run SD on my 5700 xt and pretty fast at that!

[@earthlingbox]

boy are you able to do that, can you shoot your specs the power supply ??? and cpu ??

[@margobelorukova]

It worked! This is amazing, thank you for this video!

[@mathews9958]

BRO THANK YOU SOO MUCH YOURS IS THE ONLY USEFUL VIDEO I FOUND YOURE THE BEST LOVE YOU MAN

[@zeynepseyhan9144]

I have no idea what I have done but I think it works :D thanks a lot!

[@KJotah]

It works for me on r5 5600x and rx 6750 xt. Thank you. I just had to change the python directory inside the .bat file.

[@16thSD]

how did you do it? it keeps telling me to update my pip, when i do i get the No such option: -u?? any ideas?

[@KJotah]

@@16thSD I went to the properties of the python executable, copied the address. I opened the .bat file and pasted it after SET PYTHON=, that's all I did differently

[@hudbaoir3610]

OMG THANK YOU!!! I was wondering why there is nothing happening after I try to run the .bat file lol

[@arvetis]

I'm getting "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check" I suspect that adding that command line variable will cause it to run just on the CPU, right? I don't want that. How do I fix this?

[@ted_van_loon]

TLDR: add --use-directml to the COMMANDLINE_ARGS line in webui-user.bat file.

[@ROUSTAVI]

​@@ted_van_loon It kinda worked, but now its giving erro that module (name) has no atribute and torch has no dml

[@zerohcrows]

If anybody is trying to inpaint but it's not working, you'll need to have "--opt-sub-quad-attention --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling" in your commandline_args. don't think you need "--upcast-sampling" but havent compared generation speed without it yet

[@RichardKelly-p3q]

Holy fukaroli!!! It finally worked!!! I have been trying to get AMD SDiffusion running for freaking hours then found your vid...Thanks dude!

[@ishmaelisaac4579]

awesome tutorial. I wasted about a hour and a half on other tutorials with many errors. I finished your video in about 15 minutes with minimal hassle.

[@Euterpe1]

02:33 it says "could not create work tree dir 'stable...': Invalid argument"

[@Franky10127]

same

[@Facha115]

you are incredible my guy, really nice tutorial 🙌

[@alexcannella9438]

🙏🙏🙏🙏🙏 Thanks, I've been looking for a tutorial like this for a few months now. You're the best. 😁😁😁😁

[@yeastydynasty]

THANK YOU for this tutorial. I'm up and running!

[@BruceCullen]

I can't figure out why I keep getting this error when trying to launch A1111 ---> RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions. I have two GPUs but when it loads it seems to want to use the AMD not my Nvidia, but still unsure why I can't ever produce 1 image yet with any model. I have an AMD Radeon RX5700 XT (8GB) and Nvidia GeForce GT 710

[@VAKAY]

God-tier video, thank you soooo sooo much!!!

[@silvertruenoobs340]

Dude, you're just the best. I've looked through a few other guides. I was browsing a forum dedicated to problem solving. I spent almost 6 hours on it, and I still managed to get the sd card running, but it was running so slow that it was almost torture. He spent 4 minutes creating an image in 512x512 resolution. 😔 Now I have reinstalled it according to your guidance, and it not only started immediately without any errors, but also started working much faster. Now an image in the same resolution (512x512) is generated in 30 seconds🙂

[@patozarantonello8572]

Hi there , what gpu do you have , asking for refence

[@silvertruenoobs340]

@@patozarantonello8572 I had a RX 6750 XT, but I have already changed it because it did not work on it some animation modules such as deforum, as well as generation in high resolutions. When working constantly went into defense and shutdown

[@patozarantonello8572]

@@silvertruenoobs340 thanks for responding ,sad it didnt work out with amd, i guess nvidia works a lot better in that field

[@RH-so1ot]

It looks for an Nvidia driver and wont work. Not sure the work around. I was able to get past looking for the CUDA cores, but not the drivers.

[@CharlieTheA]

Thanks so much Trent. Now it's playtime.

[@adhdgamedevlog]

thanks dude, this is working great on a Ryzen 5 5600x + Radeon 6600xt 32gb Ram - less than 30 secs to generate an image with 20 sampling steps !

[@pathet11cc]

512x512?

[@A_DWallowz]

Which size?

[@ElSenorMrTaco]

For those facing the missing python error, uninstall the python version you installed from the website and install python 3.10 from the microsoft store instead. Fixed the issue for me.

[@johnlondon8462]

Had a question. At 6:35 you put the argument --medvram. I have 8GB vram and did this and am running into runtime errors due to not enough VRAM, any suggestion?

[@ted_van_loon]

first use: --lowvram instead of --medvram. -medvram is more of a setting for a rtx 4090TI or such, or a intel arc A770, most other GPU's are much better of using --lowvram. generally in AI anything with less than 16gb of vram is concidered low VRAM, even though in this version of stable diffusion, you can still run it with even integrated gpu's with 1gb or 2gb of vram by adding that --lowvram setting and staying to 512x512p and not higher. just be aware that the GPU's which are used by datacenters and such to run AI actually advertise with calling 80gb of vram low vram and advertising of up to 1.3TB of Vram for a single node, in general with almost all consumer cards there isn't much gain in not using --lowvram, and in general there is a lot to gain by using --lowvram. I know people like to boast about their systems by saying they use setting more for datacenter grade hardware, but honnestly it will in general only harm you. the only consumer gpu's which might actually work well with --medvram are cards like the a770 or rtx 4090ti and that is only if you only render low resolution images, if you render higher resolution images you should also use --lowvram on those. ofcource there are exceptions since you can offload VRAM into normal ram, that is what most people do, so they just have enough free normal ram so that they can handle better settings and resolutions by offloading in normal ram. what resolution do you try to render at? and how much normal ram does your system have? also do you have VRAM offloading enabled? first just try to render at 512x512p, first try it with DDIM as sampling method, and just try something like 12 to 20 steps. while more steps or the sampling method shouldn't affect ram requirements to much, DDIM is fast and gives good results, also lower steps go faster, with DDIM you get okay results at those steps for testing purposes, so when trying setting try to use that first, later on you can try other ones as well. also make sure VRAM offloading is enabled and that you have sufficient normal ram as well, so for example 16gb of ram so you have around 8 gb or normal ram free for stable diffusion to use. by default however in most operating systems the VRAM offloading will automatically work on a APU, but you might need to add a parameter in your case if you have a dedicated gpu, easy way to test it, is by just running it letting it generate something at a low

[@Ryan-be2nv]

Awesome! Worked. Image does take 2m on average but it works.

[@blazeaster]

9/10 tutorial my friend. try and level the sound effect to the same level as your voice if i was wearing the headphones i'd be having a buzz in my ears fs. beside that i managed to get it running. also you should clarify the problem with the link shortening in the description and that people should copy and paste the link. idk if it would work if you put it in quotes

[@CedricF33]

Finally a really good tutorial that works. You are the best thank you bro 🤩🤩🥳🥳

[@Ridersodyssey_63]

runs smoothly on amd 580 rx. thanks

[@abhinavbisht9851]

4gb version or 8gb version?

[@Ridersodyssey_63]

8 gb

[@Ridersodyssey_63]

sadly deforum doesnt run

[@RaeSanira]

Thanks for the Video, works perfectly with RX 6600

[@einstien2409]

What is your iteration speed? how long does it take you to create the image he created?

[@RaeSanira]

@@einstien2409 35 seconds

[@nocifer]

hey bro, im trying to run it on my rx 6650m, it keeps showing an issue with use-directml... any workarounds?

[@RaeSanira]

@@nocifer uninstall Nvidia drivers if you have some

[@meworlds8216]

Thank you so much for the tutorial nice job brother! Just a feedback about the sound. Please make sure your sound levels are constant. I have to change my volume cuz you are very silent sometimes then it gets loud. Not good in the night when everything is silent haha :)

[@leftylou4693]

if you add " --autolaunch" will launch the browser for you when loaded, and you have a great tutorial but with the error pointed out before, you need Python 3.10.6 for all the things (torch) in SD to work, and also you did towards the end remembere about the model about 3 mins after, but you forgot to tell people what folder to put the model into at that point, so some could have left the video after that, just for the future so it's more step by step you know, and let's be real you are not the only one with tut for AMD. But anyways, Thank you ;)

[@nickvanimpe7881]

Hey it is working but very slowly. When I look into task manager and Performance tab, I noticed that my GPU is hardly used but it is using my full CPU/ memory. I have an AMD RTX5700XT.

[@Wolf-km7bc]

same

[@DevanshSharma-wx6zy]

ImportError: cannot import name 'opts' from partially initialized module 'modules.shared' (most likely due to a circular import) (D:\Stable Diffusion\stable-diffusion-webui-directml\modules\shared.py) Any solutions ?

[@jancycoco5827]

Error code: 1 stdout: stderr: Traceback (most recent call last): File "", line 1, in AssertionError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

[@singlsrvngfrnd]

The last line you copied is telling you what to do "add --skip-torch-cuda-test to COMMANDLINE_ARGS" where your told to add "medvram" also add --skip-torch-cuda-test

[@YoinkerDoodles]

@@singlsrvngfrnd that makes it use the cpu, which makes this tutorial completely irrelevant for people wanting to use their gpu for faster generations and possibly bigger images

[@hudbaoir3610]

Thank you. And everyone in the comments that helped solve issues.

[@TuomasHenrikkiRönty]

Thank you so much! This video was super helpful!

[@sdcpp31]

Hi guys ! So I'm blocked at 2:31 and the program says: Couldn't launch python exit code: 9009 stderr: Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases. Launch unsuccessful. Exiting. Press any key to continue . . . Even though I already have python 3.10.6 installed Does anyone know why it doesn't work ?

[@ProWideoqu]

Yes you should uninstall phyton. install it again with the box "add to PATH" checked. it should work.

[@AL-gt7uy]

@@ProWideoqu thanks brother

[@TheWilliam3227]

It works on my AMD GPU. Thank you very much.

[@HoldenGoar]

Damnnnnn my guy actually made it work. Legend

[@uniquetraveller01]

Finally something worked!!!! Thank you so much...

[@MishkaLightmist]

I got it working!!! Awesome! Thank you so much ❤

[@shengsun5916]

You're a CHAMP, much appreciate your help！

[@robertgoldbornatyout]

Excellent upload, I got it working, on rx580 8 gbram, takes about 1minite to generate 512 x 512, image, now i just have to learn all about the interface. thanks for the upload liked subed,

[@LazyNinjaZ]

i have the same card and followed the same steps in the video but the final image looks burned out... did you face any issues?

[@peterecker3599]

Tanks! U deserve more likes man!

[@hsynnl4746]

when i try to run "webui-user.bat", i see this error "Couldn't launch python" ... but i launched the python... can u help me?

[@sandilekhoza798]

Getting the same error

[@coolname.]

anyone know what to do for AttributeError: module 'torch' has no attribute 'dml'

[@RJTheJunior]

i have the same issue

[@jarodhanson57]

i have the same issue

[@jarodhanson57]

@@RJTheJunior me aswell, ever find out?

[@RJTheJunior]

@@jarodhanson57 nope, it seems like nobody is willing to help anymore unfortunately

[@rafa-lk6lf]

yes it worked good but when i try to generate my prompt, the final result is always a black screen image on output :/

[@Ruszie]

Thank You Buddy it was so much helpfull..

[@dustboii]

Worked after some troubleshooting. Thanks.

[@zgraggenst]

Thank you very much for this tutorial. There are quite a few of them available, but this is the only one that works with my AMD Radeon RX 6700 XT. It took a while to find working startup options for the webui-user.bat, but with set COMMANDLINE_ARGS=--medvram --no-half --precision full --opt-sub-quad-attention I manage to create 768x768 pictures without a problem. I have one question though, how do I know which version of stable-diffusion is installed and how do I upgrade to the latest version? Thanks a lot and kind regards!

[@Blairjeworski]

it tells you at the top at start up i think. add git pull to the top of the webui bat to auto upgrade.

[@Rodritoledo94]

Thanks, it worked flawlessly

[@carlosan6965]

thanks, it worked for me on an RX 6600

[@RaeSanira]

how? I followed all the steps and is not working

[@RaeSanira]

it renders on CPU

[@RaeSanira]

It works, I discovered I had installed NVIDIA Drivers, causing the issue for the AMD installation.

[@SwoleKitchen]

This is the best tutorial bro

[@julianelagrimas6261]

No module named pip Launch unsuccessful. Exiting. Press any key to continue . . . im getting this error help

[@djb2046]

I need some help. I did everything as instructed, but I’m getting this error when attempting to launch the website.bat file. couldn't launch python exit code: 9009 stderr: Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings Manage App Execution Aliases. Launch unsuccessful. Exiting.

[@humansvd3269]

Copy the location where the python.exe file is by editing the bat file and paste it.

[@MichaelR1975]

thanks, I got SD to work with My CPU which was very slow and not to great quality. But now working much better using my RX6600.

[@savatesky]

Hi there may i ask how did you manage to do so ? cause i can't :)

[@manojkchauhan]

Thanks, Installed with RX590 8gb Intel i9, Working fine,

[@licensedviber]

nice I have the same card

[@KingofDaba]

Thanks a lot. Yesterday was half a day, and I've been trying since this morning, but Stable Diffusion didn't work well. Watching your video solved all my problems. thank you very much.😄 (Ryzen7 2700X Radeon RX590)

[@AnthonyWashingtonX2K]

Thanks for the video. I am looking at it 4 months after you posted and it's still good. I got it to work using my RX 480 with 8GC VRAM and the argument: --medvram --precision full --no-half --no-half-vae --opt-split-attention-v1 --opt-sub-quad-attention --disable-nan-check . Try to figure out max resolution. Render times are barrable. I am currently using PlagroundAI. Thanks again for a great video.

[@Kitten_Genocide]

how do you make it work without the "--onnx --backend directml"? If I don't run it with this arguments, it still run on only CPU.

[@SwoleKitchen]

You sir are the goat

[@АхматДжуккаев-ф5ь]

RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check what is the proplem?

[@dennik8374]

Great Video maybe u can add that you have to make python registered to windows path but thanks for the video works now!!!!!

[@OnryoHorrorChannel]

Can't generate anything, it gives me a float error, how do I fix it?

[@PatRose]

thank you so much it worked

[@rasavarnam]

it worked! beautiful 👌i had low vram and I ran into few errors because of that, but after adding those flags, it all seems to be working well .. now need to try out with different models .. heard amd gpus won't work with every models ..

[@thatonetryhard8427]

Im having an issue with sable diffusion. it keeps say "RuntimeError: Could not allocate tensor with 4915840 bytes" when I have 8g of Vram. Is there something wrong on my end?

[@Vay0]

Did you fix it?

[@Akuto1337]

same issue, any fix?

[@bhootpurvmanusya]

he seems like a prompter(term used for who use copy others and ad prompt of their own claiming they trained the ai) ur using rocm for amd. which is supported until rdna2 he's happy it worked cuz its 7000 series card which is top tier and works perfectly with rocm for rdna1 (5000 series) ur only bet is using thirdparty libraries which use it raw without optmisation.happy coding

[@heinskarte4235]

For some reason it says "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check" , if I add that, it works but it only uses CPU... Anyone knows how to solve this so it uses GPU?

[@thanatopsis70]

I have the same problem. I looked up other AMD tutorials that do things a bit differently, but they all give me the same error. Somewhere somebody said to put --use-directml --reinstall-torch in the webui-user.bat file under set COMMANDLINE_ARGS= but that didn't work for me either. It did actually run, but it was super slow, probably still just using CPU. This guy made a video about it, but his fix still didn't work for me either. kzbin.info/www/bejne/o3zbpWOhraloeZI

[@simondiaz184]

@@thanatopsis70

[@Vegas-z5n]

@@thanatopsis70 thank you, your advice helped

[@mikesavad]

changed python, and now the bat file is stuck on: Model loaded in 35.0s (calculate hash: 31.6s, load weights from disk: 0.3s, create model: 0.2s, apply weights to model: 0.5s, apply half(): 0.4s, move model to device: 1.9s). And hasn't moved. Tried it twice.

[@CodyCloud]

same spot did you get yours running?

[@mikesavad]

@@CodyCloud i gave up. Especially when I read the comments below, where they could only do a 512x512 sample size. Which has no value to me at all. On the plus side it went further than other installs. But it doesn't actually run or work despite downgrading python. I'll just wait longer, someone is bound to figure out a way for amd to work. I tried a number of ways to make this work, but the limitations don't make it any better than using it live.

[@PotatoWagon]

@@mikesavad You will only be able to do it with 512x512 or 512x768 or 768x512. From there you upscale it with hires fix, or with img2img, or from the extras tab. It's the same with Nvidia GPUs.

[@acedrink999]

Ît took me ages to figure it out. For me it worked, if I change the edit to "COMMANDLINE_ARGS= --opt-sub-quad-attention --medvram --autolaunch git pull" instead of what he mentioned. Hope this helps some people!

[@lukaswagner8615]

I added --autolaunch behind --disable-nan-check and it worked for me

[@jaimegonzalez1395]

WOW I really love you man!! Thank U