Seemingly infinite (or just really, really long) hang on first launch

[Vistril]

After finally figuring out how to get this working with my AMD RX 5700XT on Ubuntu 22.04.4, I was able to get the web interface running. Queuing my first prompt resulted in this:

$ HSA_OVERRIDE_GFX_VERSION=10.3.0 python3 main.py > ... got prompt model_type EPS adm 2816 Using split attention in VAE Working with z of shape (1, 4, 32, 32) = 4096 dimensions. Using split attention in VAE clip missing: ['clip_l.text_projection', 'clip_l.logit_scale', 'clip_g.logit_scale'] Requested to load SDXLClipModel Loading 1 new model

And it just hangs here, for quite a while! Do I wait it out? Or is something stuck? It's super late enough in the day I'm much too tired to keep troubleshooting, so I'm making a GitHub issue and seeing what this thread looks like tomorrow. It's probably some no-brainer problem that I'm too tired to notice right now. :(

Here's any relevant info that may help. The ROCMinfo Ryzen 7 5700X 16C RX 5700XT Ultra 8GB Weaksauce 16GB 2400MHz DDR4 Ubuntu 22.04.4 LTS x86_64

I'm new to SD so I'll happily provide any other log files or supporting material you may need. TIA:)

[Symber13]

Did you ever get this working? I'm having the same issue with a 7800XT. I've seen success with your card by downgrading rocm to 5.2 as well a few other dependancies (sorry, I need 5.5+ so I didn't pay much attention, but I'm sure you can google it). Just curious if I'm doing something dumb. I have Automatic1111 running fine, but Comfy is just giving me the same error you have.

[Vistril]

nope ive shelved it since

On Fri, Mar 1, 2024 at 12:18 PM Symber @.***> wrote:

Did you ever get this working? I'm having the same issue with a 7800XT. I've seen success with your card by downgrading rocm to 5.2 as well a few other dependancies (sorry, I need 5.5+ so I didn't pay much attention, but I'm sure you can google it). Just curious if I'm doing something dumb. I have Automatic1111 running fine, but Comfy is just giving me the same error you have.

— Reply to this email directly, view it on GitHub https://github.com/comfyanonymous/ComfyUI/issues/2901#issuecomment-1973778124, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJGDTEYREZXWRJMJ7LS47LTYWDH7TAVCNFSM6AAAAABDYTPDWKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZTG43TQMJSGQ . You are receiving this because you authored the thread.Message ID: @.***>

[MaxTran96]

same, I also just stuck at Loading 1 new model, and the vram crash and goes back to 0

[lcandotti]

Same issue here with the following setup:

• CPU: Ryzen 9 5900X
• GPU: RX 6950XT
• OS: Archlinux
• Kernel: 6.8.5-arch1-1
• Python version: 3.11.8
• PyTorch version: 2.4.0.dev20240416+rocm6.0
• RocM version: 6.0.2-1

Rarely, I have the following error message getting print when reaching a KSampler node Memory access fault by GPU node-1. Reason: Page not present or supervisor privilege. Otherwise and most of the time it just reach Clip Text Encode (Prompt) or KSampler and get stuck at Loading 1 new model