Not enough on 4GB vRAM

[lllyasviel]

Not enough on 4GB vRAM

[AiSaurabhPatil]

It's working in my case
I have 16GB RAM and Nvidia GTX 1650 with 4GB VRAM
example :

Yeah but that's true , it make my system unusable !!

[lllyasviel]

Both GTX 1650 and 850 has 4GB vram, but it seems that 1650 works and is "smarter" to use vram in more efficiency way.
Probably 850 is really very difficult to support. We will probably add some descriptions in the Readme to make it clearer

[Tigwin]

I have a 3080FE and it has 10gb of ram. I think we need a medram setting cause it takes a looooong time to render some stuff. But if I keep it simple (low poly, "whale"), it renders about as fast as I'd expect.

[NoMansPC]

It eats up almost all of my RAM too, making my computer unusable. Sadly, for now, buying a new card is still the proper way to use SDXL.

[Tigwin]

I just found out a temp fix for the vram issue on my desktop. Disable the video card then enable it, freed up almost half my vram (powershell)

Get-PnpDevice -FriendlyName "NVIDIA GeForce RTX 3080" | Disable-PnpDevice
Sleep -Seconds 5
Get-PnpDevice -FriendlyName "NVIDIA GeForce RTX 3080" | Enable-PnpDevice

[iLKke]

Fails for me the same way as for the OP.
I have 16GB RAM and an GTX 1050 Ti with 4GB VRAM
I noticed that it will gobble all of free space on C: that (cca 7GB) before failing due to not enough memory.

[sangshuduo]

I have 32GB RAM and an 3050Ti with 4GB VRAM. Wish it can be supported.

[iLKke]

OK as suspected, I needed to free enough disk space and now it will run.
The error message was misleading as it's not VRAM or RAM but disk space that was causing it to fail.

Curiously enough, it seems to only be using about 2GB of VRAM

[mimze]

16GB of RAM and 4GB of VRAM (GTX 950m). I get no errors but it just gets stuck when I hit Generate.

[game-alle]

I have an rtx 3050ti, and it's just stuck in this state and not progressing.

[xjdeng]

Also fails on mine: GTX 960M, 4GB VRAM, 16 GB RAM:

D:\JJ\python\Fooocus_win64_2-1-791>.\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic
Already up-to-date
Update succeeded.
[System ARGV] ['Fooocus\\entry_with_update.py', '--preset', 'realistic']
Python 3.10.9 (tags/v3.10.9:1dd9be6, Dec  6 2022, 20:01:21) [MSC v.1934 64 bit (AMD64)]
Fooocus version: 2.1.802
Running on local URL:  http://127.0.0.1:7865

To create a public link, set `share=True` in `launch()`.
Total VRAM 4096 MB, total RAM 16250 MB
Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram
Set vram state to: LOW_VRAM
Disabling smart memory management
Device: cuda:0 NVIDIA GeForce GTX 960M : native
VAE dtype: torch.float32
Using pytorch cross attention
Refiner unloaded.
model_type EPS
adm 2816
Using pytorch attention in VAE
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Using pytorch attention in VAE
extra keys {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale', 'cond_stage_model.clip_g.transformer.text_model.embeddings.position_ids'}
Base model loaded: D:\JJ\python\Fooocus_win64_2-1-791\Fooocus\models\checkpoints\realisticStockPhoto_v10.safetensors
Request to load LoRAs [('SDXL_FILM_PHOTOGRAPHY_STYLE_BetaV0.4.safetensors', 0.25), ('None', 0.25), ('None', 0.25), ('None', 0.25), ('None', 0.25)] for model [D:\JJ\python\Fooocus_win64_2-1-791\Fooocus\models\checkpoints\realisticStockPhoto_v10.safetensors].
Loaded LoRA [D:\JJ\python\Fooocus_win64_2-1-791\Fooocus\models\loras\SDXL_FILM_PHOTOGRAPHY_STYLE_BetaV0.4.safetensors] for model [D:\JJ\python\Fooocus_win64_2-1-791\Fooocus\models\checkpoints\realisticStockPhoto_v10.safetensors] with 1052 keys at weight 0.25.
Fooocus V2 Expansion: Vocab with 642 words.
Fooocus Expansion engine loaded for cpu, use_fp16 = False.
Requested to load SDXLClipModel
Requested to load GPT2LMHeadModel
Loading 2 new models
[Fooocus Model Management] Moving model(s) has taken 10.48 seconds
App started successful. Use the app with http://127.0.0.1:7865/ or 127.0.0.1:7865
[Parameters] Adaptive CFG = 7
[Parameters] Sharpness = 2
[Parameters] ADM Scale = 1.5 : 0.8 : 0.3
[Parameters] CFG = 3.0
[Parameters] Seed = 7801272631270716136
[Parameters] Sampler = dpmpp_2m_sde_gpu - karras
[Parameters] Steps = 30 - 15
[Fooocus] Initializing ...
[Fooocus] Loading models ...
Refiner unloaded.
[Fooocus] Processing prompts ...
[Fooocus] Preparing Fooocus text #1 ...
[Prompt Expansion] cat, cinematic, complex, highly detailed, extremely, sharp focus, beautiful, stunning composition, symmetry, great colors, aesthetic, very inspirational, colorful, deep color, inspiring, original, full bright, lovely, cute, artistic, intricate, elegant, perfect light, fine detail, clear, ambient background, professional, creative, positive, amazing, pure, wonderful, unique
[Fooocus] Preparing Fooocus text #2 ...
[Prompt Expansion] cat, very coherent, cute, cinematic, detailed, intricate, stunning, highly refined, epic composition, magical atmosphere, full color, elegant, luxury, amazing detail, professional, winning, thoughtful, calm, beautiful, unique, best, awesome, perfect, ambient light, shining, illuminated, translucent, fine, artistic, pure, positive, attractive, creative, vibrant
[Fooocus] Encoding positive #1 ...
[Fooocus] Encoding positive #2 ...
[Fooocus] Encoding negative #1 ...
[Fooocus] Encoding negative #2 ...
Preparation time: 41.61 seconds
[Sampler] refiner_swap_method = joint
[Sampler] sigma_min = 0.02916753850877285, sigma_max = 14.614643096923828
Requested to load SDXL
Loading 1 new model

D:\JJ\python\Fooocus_win64_2-1-791>pause
Press any key to continue . . .

[lllyasviel]

Yes I should modify the Readme to say that the 4GB VRAM needs to have an GPU Arch to support float16.

But how can GTX 960M and SDXL be put in one same sentence

[prime0x2]

I have 8GB RAM and 1050TI 4GB. Can I use it?

[xjdeng]

Yes I should modify the Readme to say that the 4GB VRAM needs to have an GPU Arch to support float16.

But how can GTX 960M and SDXL be put in one same sentence

Because

1. SDXL and 4 GB VRAM were in the same sentence (or at least in the same repo)
2. The 960M worked fine with fp16 1.5 models

Now it's entirely possible it's another issue and I can accept that. But it seems this repo is implying it'll work with all 4 GB Nvidia GPUs when I've found a countetexample: my old laptop's.

[AFOLcast]

There's a note I found buried on the GitHub about new versions not working with "old" computers. I downloaded the patch and installed. Now my RTX 2060 works again.

…

On Thu, Nov 30, 2023, 6:59 AM xjdeng ***@***.***> wrote: Yes I should modify the Readme to say that the 4GB VRAM needs to have an GPU Arch to support float16. But how can GTX 960M and SDXL be put in one same sentence Because 1. SDXL and 4 GB VRAM were in the same sentence in the repo 2. The 960M worked fine with fp16 1.5 models Now it's entirely possible it's another issue and I can accept that. But it seems this repo is implying it'll work with all 4 GB Nvidia GPUs when I've found a countetexample: my old laptop's. — Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BB4LFNLVN2BEGGYFWJUAJODYHBYK7AVCNFSM6AAAAAA3N4C3ASVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMZTGYZDOOBZGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[xjdeng]

Patch? Where?

…

On Thu, Nov 30, 2023 at 7:07 AM AFOLcast ***@***.***> wrote: There's a note I found buried on the GitHub about new versions not working with "old" computers. I downloaded the patch and installed. Now my RTX 2060 works again. On Thu, Nov 30, 2023, 6:59 AM xjdeng ***@***.***> wrote: > Yes I should modify the Readme to say that the 4GB VRAM needs to have an > GPU Arch to support float16. > > But how can GTX 960M and SDXL be put in one same sentence > > Because > > 1. SDXL and 4 GB VRAM were in the same sentence in the repo > 2. The 960M worked fine with fp16 1.5 models > > Now it's entirely possible it's another issue and I can accept that. But > it seems this repo is implying it'll work with all 4 GB Nvidia GPUs when > I've found a countetexample: my old laptop's. > > — > Reply to this email directly, view it on GitHub > <#14 (comment)>, > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/BB4LFNLVN2BEGGYFWJUAJODYHBYK7AVCNFSM6AAAAAA3N4C3ASVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMZTGYZDOOBZGI> > . > You are receiving this because you are subscribed to this thread.Message > ID: ***@***.***> > — Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AECXP3GW66G4WQMBXHYTWBDYHBZI7AVCNFSM6AAAAAA3N4C3ASVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMZTGY2DGNJRGI> . You are receiving this because you commented.Message ID: ***@***.***>

[rohanmandrekar]

Fails for me the same way as for the OP. I have 16GB RAM and an GTX 1050 Ti with 4GB VRAM I noticed that it will gobble all of free space on C: that (cca 7GB) before failing due to not enough memory.

same issue. any way I can prevent it from using space on C drive?

[mashb1t]

You can not as when VRAM is exhausted RAM is used and when RAM is (near) full the swap is used.
Preventing offloading from GPU is possible, but you'll not be able to run Fooocus then.
See https://github.com/lllyasviel/Fooocus/blob/main/troubleshoot.md

[cp818]

Fails for me the same way as for the OP. I have 16GB RAM and an GTX 1050 Ti with 4GB VRAM I noticed that it will gobble all of free space on C: that (cca 7GB) before failing due to not enough memory.

same issue. any way I can prevent it from using space on C drive?

Same here too. no clear instructions?

[mashb1t]

@cp818 let's do not revive a 4 months old thread, please open a discussion or a new issue and provide all necessary information such as terminal output, hardware specs etc.